Loading video...

Video Failed to Load

Go Home

BREAKING $AMD Dr. Lisa Su Fresh Interview 🚀🚀 Note: This was after CES AMD ANALysts concluded OpenAI has no money, but $40B Cash and Raising $100B where they already secured $22B from SoftBank. This morning they also concluded Saudi has $0 money backing and LumaAI for 2GW AMD chips...

MikeLongTerm's profile picture

Mike

22,162 subscribers

35,337 views • 7 months ago •via X (Twitter)

LIVE

Anya Rossi• Live Now

Private livecam show

0 Comments

No comments available

Comments from the original post will appear here

Related Videos

$AMD Dr. Su full press in Taiwan 🆕🚨🚨🚨 I dont think AMD analysts or bears are ready for H2 2026 and 2027. The level of growth will be so massive that everyone would be shocked. Well mostly because they didnt read Mike threads. CPU:GPU Ratio is much higher than 1:1, most likely in the 3-5:1 moving to 5-10:1 . Enterprises are demanding more and more agents running tasks longer and more intensive. Dr. Su is too nice to say it out loud. But that does not mean training gonna go away. Just growing significantly slower than Inference.
6:50

$AMD Dr. Su full press in Taiwan 🆕🚨🚨🚨 I dont think AMD analysts or bears are ready for H2 2026 and 2027. The level of growth will be so massive that everyone would be shocked. Well mostly because they didnt read Mike threads. CPU:GPU Ratio is much higher than 1:1, most likely in the 3-5:1 moving to 5-10:1 . Enterprises are demanding more and more agents running tasks longer and more intensive. Dr. Su is too nice to say it out loud. But that does not mean training gonna go away. Just growing significantly slower than Inference.

Mike

233,700 views • 3 months ago

$AMD $620/share is too conservative for 2026 🧵 Some quick facts before I dive into this super long thread: $META allocated 42% GPUs to $AMD and 58% to $NVDA OpenAI allocated 6GW(38%) to $AMD and 10GW to $NVDA My $620 PT below by end of 2026 was only for 10-15% market share. I believe $AMD is going to have much much higher market share than I projected. The AI accelerator market is exploding, projected to reach $500 billion by 2028(is now heading $1Tril), driven by insatiable demand for training and inference compute in large language models (LLMs), recommendation systems, and autonomous systems. Nvidia ($NVDA) has long held a stranglehold, commanding over 90% market share through its CUDA ecosystem and superior rack-scale solutions. However, AMD is mounting a formidable challenge, leveraging cost advantages, open-source software momentum, and hyperscaler partnerships to erode Nvidia's moat. Recent deals—such as Meta's ($META) allocation of 42% of its GPU capacity to AMD and OpenAI's commitment to 6GW of AMD compute (versus 10GW for Nvidia)—signal a tipping point. At the forefront is AMD's Instinct MI450 series, a next-generation AI GPU slated for H2 2026 launch, which promises "no-excuses" leadership in training, inference, and distributed workloads. This analysis dissects how AMD will capture more market share and why hyperscalers like $Meta , xAI , Oracle , and others are poised to become voracious buyers of the MI450. AMD's AI GPU revenue has surged from negligible levels in 2022 to an estimated $4-5 billion in 2025, capturing ~6% of the data center GPU market. This growth stems from the Instinct MI300X, which offers 141GB of HBM3 memory and competitive FP8/FP16 performance at 20-30% lower cost than Nvidia's H100. Hyperscalers, facing NVIDIA 's overcharging, have turned to AMD for diversification. Meta, for instance, plans 600,000 H100-equivalent GPUs by end-2024, with ~42% (or 250,000+ units) sourced from AMD's MI300 series for inference tasks like image editing and AI assistants. Similarly, OpenAI's recent multi-year deal commits to 6GW of AMD compute—equivalent to ~300,000-400,000 MI450 GPUs—starting with 1GW in 2026, explicitly to counterbalance its 10GW Nvidia allocation. These aren't one-offs. Microsoft Azure, Amazon AWS, and Oracle Cloud Infrastructure (OCI) have integrated MI300X for AI workloads, with Oracle deploying 30,000 MI355X units in zettascale clusters. xAI, Elon Musk Musk's AI venture, ran 30% of Grok-1's production traffic on MI300X GPUs and has confirmed ongoing purchases. Collectively, these partners represent over $400 billion in projected AI infrastructure spend through 2028, with AMD targeting up to 40% market share. For those that subscribed, I wrote a specific thread on how AMD "secret weapon" is going to change the game in 2026 with an improved designs on all its products, yes AMD has patent on it. Software is the linchpin. AMD's ROCm platform, once derided as "half-baked," now supports day-zero integration for Llama-4, DeepSeek V3, and GPT-OSS models—closing the CUDA gap. Benchmarks show MI355X (MI450 precursor) outperforming Nvidia's B200 in inference by 1.5-2x on memory-bound tasks, at 25-35% lower TCO. For training, MI450's rack-scale IF128 configuration (128 GPUs, 1.4 PB/s intra-rack bandwidth) rivals Nvidia's VR200 NVL144, enabling clusters like xAI's Colossus (scaling to 1M GPUs). My below thread projected Etimated conservative FY 25 revenue: $34-$36B Estimated conservative FY 26 revenue: $55B-$62B Below is why $AMD is revenue is going to be much higher after OpenAI deal. 1. OpenAI 1GW in 2026. With high demand for MI355X at $30,000k+ per unit, with MI450 is likely to be sold in the $45k-$55k. We can safely calcuate 1GW would require roughly 400,000 MI450 GPUs. or Roughly ~$20B revenue in 2026 alone from OpenAI. That would mean $AMD would hit $56B just from one partnership(OpenAI) in 2026 2. $META, the biggest spender on AI Infrastructure right now, Daddy Zuckerberg bought 250,000+ MI300, and is buying MI355X for recommendation engines and Llama training. It is very unlikely for Daddy Zuck to slow down AMD Chips, due to its Inference superiority to NVDA Chips. Most likely we will see at least 300,000-400,000 MI355X ordered from now toward end of H1 2025. And another 300,000-500,000 MI450 by H2 2025. Or ~$20B from just Meta in H2 alone, excluded H1. 3. xAI : Musk confirmed "AMD GPUs work very well" for Grok's small/medium models, with 30% of Grok-1 on MI300X. xAI's Colossus (200K+ GPUs, targeting 1M) and Oracle partnership (via OCI's MI355X cluster) position it for MI450 trials in H1 2026. With $6B funding and Grok integration into Oracle services, xAI could allocate 10-20% ($10B-$15B) to MI450 for distributed inference. We haven't heard the detail from Daddy Elon Musk yet, but most likely not going to be spending less than OpenAI or Sam Altman 4. Oracle ($ORCL): A multi-billion-dollar MI355X deal powers OCI's AI superclusters, with $500B+ remaining performance obligations. Larry Ellison's zettascale ambitions and xAI/OpenAI integrations make Oracle a MI450 anchor tenant—projected 50-100k units ($15B+ spend) for enterprise AI platforms. $ORCL is likely to spend more on the new "secret weapon" due to its capability in AI inference and cost advantage for $500B backlog. 5. Others ( Microsoft , Amazon , Saudi+other countries): Microsoft (Azure MI300X for training) and Amazon ($148B 15-year spend) test MI450 via Stargate ($500B with Oracle/SoftBank). Emerging buyers like G42 (5GW UAE campus), Crusoe, and Hot Aisle add 5-10GW demand. These potentially would add $15B-$30B in 2026 alone. We also need to factor in $TSM supply constraint( $NVDA is TSMC favorite), so $AMD market cap/growth is being tamed by TSMC. So what are you saying Mike, well $AMD 2026 revenue could hit $90-$100B by end of 2026 or nearly 185% growth YoYo. So what does that mean for valuation? I have no idea how Mr. Market gonna value AMD in 2026 with 3 digits growth. My Conservative $620 was my best projection until today with OpenAI partnership. I'm telling you as one of the biggest AMD bull, that I will leave it to "smart money" and other investors to do the price discovery while I'm chilling and writing DDs daily. Lastly, AMD's MI450 isn't hype—it's a calibrated strike at Nvidia's vulnerabilities, amplified by hyperscaler bets like Meta's 42% allocation and OpenAI's 6GW lifeline. By prioritizing inference efficiency, rack-scale innovation, and open ecosystems, AMD will siphon 10-15% share in 2026, scaling to 20%+ as TCO trumps CUDA loyalty. Meta, xAI, Oracle et al. aren't passive; they're active co-designers, betting billions on MI450 to fuel AGI pursuits without Nvidia's premium. For investors, this is AMD's inflection Per Dr. Lisa Su Not Financial Advice!
1:06

$AMD $620/share is too conservative for 2026 🧵 Some quick facts before I dive into this super long thread: $META allocated 42% GPUs to $AMD and 58% to $NVDA OpenAI allocated 6GW(38%) to $AMD and 10GW to $NVDA My $620 PT below by end of 2026 was only for 10-15% market share. I believe $AMD is going to have much much higher market share than I projected. The AI accelerator market is exploding, projected to reach $500 billion by 2028(is now heading $1Tril), driven by insatiable demand for training and inference compute in large language models (LLMs), recommendation systems, and autonomous systems. Nvidia ($NVDA) has long held a stranglehold, commanding over 90% market share through its CUDA ecosystem and superior rack-scale solutions. However, AMD is mounting a formidable challenge, leveraging cost advantages, open-source software momentum, and hyperscaler partnerships to erode Nvidia's moat. Recent deals—such as Meta's ($META) allocation of 42% of its GPU capacity to AMD and OpenAI's commitment to 6GW of AMD compute (versus 10GW for Nvidia)—signal a tipping point. At the forefront is AMD's Instinct MI450 series, a next-generation AI GPU slated for H2 2026 launch, which promises "no-excuses" leadership in training, inference, and distributed workloads. This analysis dissects how AMD will capture more market share and why hyperscalers like $Meta , xAI , Oracle , and others are poised to become voracious buyers of the MI450. AMD's AI GPU revenue has surged from negligible levels in 2022 to an estimated $4-5 billion in 2025, capturing ~6% of the data center GPU market. This growth stems from the Instinct MI300X, which offers 141GB of HBM3 memory and competitive FP8/FP16 performance at 20-30% lower cost than Nvidia's H100. Hyperscalers, facing NVIDIA 's overcharging, have turned to AMD for diversification. Meta, for instance, plans 600,000 H100-equivalent GPUs by end-2024, with ~42% (or 250,000+ units) sourced from AMD's MI300 series for inference tasks like image editing and AI assistants. Similarly, OpenAI's recent multi-year deal commits to 6GW of AMD compute—equivalent to ~300,000-400,000 MI450 GPUs—starting with 1GW in 2026, explicitly to counterbalance its 10GW Nvidia allocation. These aren't one-offs. Microsoft Azure, Amazon AWS, and Oracle Cloud Infrastructure (OCI) have integrated MI300X for AI workloads, with Oracle deploying 30,000 MI355X units in zettascale clusters. xAI, Elon Musk Musk's AI venture, ran 30% of Grok-1's production traffic on MI300X GPUs and has confirmed ongoing purchases. Collectively, these partners represent over $400 billion in projected AI infrastructure spend through 2028, with AMD targeting up to 40% market share. For those that subscribed, I wrote a specific thread on how AMD "secret weapon" is going to change the game in 2026 with an improved designs on all its products, yes AMD has patent on it. Software is the linchpin. AMD's ROCm platform, once derided as "half-baked," now supports day-zero integration for Llama-4, DeepSeek V3, and GPT-OSS models—closing the CUDA gap. Benchmarks show MI355X (MI450 precursor) outperforming Nvidia's B200 in inference by 1.5-2x on memory-bound tasks, at 25-35% lower TCO. For training, MI450's rack-scale IF128 configuration (128 GPUs, 1.4 PB/s intra-rack bandwidth) rivals Nvidia's VR200 NVL144, enabling clusters like xAI's Colossus (scaling to 1M GPUs). My below thread projected Etimated conservative FY 25 revenue: $34-$36B Estimated conservative FY 26 revenue: $55B-$62B Below is why $AMD is revenue is going to be much higher after OpenAI deal. 1. OpenAI 1GW in 2026. With high demand for MI355X at $30,000k+ per unit, with MI450 is likely to be sold in the $45k-$55k. We can safely calcuate 1GW would require roughly 400,000 MI450 GPUs. or Roughly ~$20B revenue in 2026 alone from OpenAI. That would mean $AMD would hit $56B just from one partnership(OpenAI) in 2026 2. $META, the biggest spender on AI Infrastructure right now, Daddy Zuckerberg bought 250,000+ MI300, and is buying MI355X for recommendation engines and Llama training. It is very unlikely for Daddy Zuck to slow down AMD Chips, due to its Inference superiority to NVDA Chips. Most likely we will see at least 300,000-400,000 MI355X ordered from now toward end of H1 2025. And another 300,000-500,000 MI450 by H2 2025. Or ~$20B from just Meta in H2 alone, excluded H1. 3. xAI : Musk confirmed "AMD GPUs work very well" for Grok's small/medium models, with 30% of Grok-1 on MI300X. xAI's Colossus (200K+ GPUs, targeting 1M) and Oracle partnership (via OCI's MI355X cluster) position it for MI450 trials in H1 2026. With $6B funding and Grok integration into Oracle services, xAI could allocate 10-20% ($10B-$15B) to MI450 for distributed inference. We haven't heard the detail from Daddy Elon Musk yet, but most likely not going to be spending less than OpenAI or Sam Altman 4. Oracle ($ORCL): A multi-billion-dollar MI355X deal powers OCI's AI superclusters, with $500B+ remaining performance obligations. Larry Ellison's zettascale ambitions and xAI/OpenAI integrations make Oracle a MI450 anchor tenant—projected 50-100k units ($15B+ spend) for enterprise AI platforms. $ORCL is likely to spend more on the new "secret weapon" due to its capability in AI inference and cost advantage for $500B backlog. 5. Others ( Microsoft , Amazon , Saudi+other countries): Microsoft (Azure MI300X for training) and Amazon ($148B 15-year spend) test MI450 via Stargate ($500B with Oracle/SoftBank). Emerging buyers like G42 (5GW UAE campus), Crusoe, and Hot Aisle add 5-10GW demand. These potentially would add $15B-$30B in 2026 alone. We also need to factor in $TSM supply constraint( $NVDA is TSMC favorite), so $AMD market cap/growth is being tamed by TSMC. So what are you saying Mike, well $AMD 2026 revenue could hit $90-$100B by end of 2026 or nearly 185% growth YoYo. So what does that mean for valuation? I have no idea how Mr. Market gonna value AMD in 2026 with 3 digits growth. My Conservative $620 was my best projection until today with OpenAI partnership. I'm telling you as one of the biggest AMD bull, that I will leave it to "smart money" and other investors to do the price discovery while I'm chilling and writing DDs daily. Lastly, AMD's MI450 isn't hype—it's a calibrated strike at Nvidia's vulnerabilities, amplified by hyperscaler bets like Meta's 42% allocation and OpenAI's 6GW lifeline. By prioritizing inference efficiency, rack-scale innovation, and open ecosystems, AMD will siphon 10-15% share in 2026, scaling to 20%+ as TCO trumps CUDA loyalty. Meta, xAI, Oracle et al. aren't passive; they're active co-designers, betting billions on MI450 to fuel AGI pursuits without Nvidia's premium. For investors, this is AMD's inflection Per Dr. Lisa Su Not Financial Advice!

Mike

711,006 views • 10 months ago

$AMD $5 Trillion MC Is Inevitable Long Term👑 This thread will focus more on Inference! 2026 EPYC "Venice" $TSM 2nm to save Large GW Scale Inference by 40% more than Prior Turin gen. Context: EPYC Turin achieves ~$0.001 per million tokens for batch inference vs $0.02-$0.12/ million tokens as I wrote the thread below. Venice is going to lower cost down to $0.0005-$0.0006/Million Tokens. OpenAI spent roughly $20B on Inference and Training, where 80-90% of that was for Inference per Analysts. AKA Renting Compute is Expensive AF! In this thread, I want to focus on why most analysts and investors are underestimating the role EPYC "Venice" and future Gen on overall Data center revenue. And $TSM ramping up 2nm supply early is a confirmation that AMD will be a major buyer long term. I will also link the thread the Gap between AMD Analysts & Reality and 2nm Ramp Thread so you have more comprehensive view of what I'm writing here. Before I go into detail this is my 2026 Projection: AI GPUs: $35-$50B EPYC Data Center: $15B-$17B Client Segment: $12-$13B Gaming: $6B Embedded: $4B-$5B Total Revenue $70-$100B Non-GAAP net income $18B-$25B Non-GAAP EPS $10.97-$15.40 Foward P/E 55x-70x= $603-$1,078 AMD's Analysts are projecting $0 Revenue for MI450 and sluggish EPYC Growth. Meaning, all analysts are either full of 💩 or Sexist, you decide! Analysts are also projecting 0% growth on AMD "Secret Weapon" Chip as $MSFT said we are at significant Windows refresh and upgrade cycle. Do you think TSMC would allocate more 2nm supply to $AMD at $0 MI450 revenue and sluggish EPYC? 1. EPYC is going to be the leader in lowest Inference! Current Turin cost saving is 95% vs $NVDA or 98-99% on Inference cost when you factor in renting Inference compute from Amazon Web Services, Microsoft Azure, or $NVDA Neocloud pets. TSMC claimed: 10-15% higher performance at iso-power, 25-30% lower power at iso-speed, and ~15% higher transistor density compared to 3nm. This reduces operational expenses (energy, cooling) while increasing throughput per chip. EPYC Turin achieves ~$0.001 per million tokens for batch inference (via vLLM on models like Llama 3 70B), driven by high core counts and low hardware costs. EPYC Venice offers ~1.7x overall performance and up to 70% more compute capability per core, with up to 256 cores (512 threads). Enhanced vector/AI instructions and open-source firmware (openSIL) optimize for inference workloads. AMD Incorporates AI Engines (now part of AMD's XDNA) for on-chip acceleration, improving efficiency for low-latency and edge inference. This reduces reliance on discrete GPUs, lowering system complexity and TCO. Venice SKUs are projected at $3,000-$15,000 ($5,000 for 256-core flagship), far below NVIDIA Rubin ($50,000-$90,000) or AMD's own MI450 GPUs ($40,000-$50,000). High memory bandwidth (up to 1.6 TB/s) supports efficient batch inference. Venice is designed exactly for Large customers that want to lower Inference Cost and MI450 Helios is for Customers that want Training at lowest TCO, TDP as well as lower Upfront 1GW scale(Full build $35-$40B vs $NVDA $55B-$80B). 2. Real World Example: OpenAI's 2025 inference spend reached ~$20B, escalating to even higher total compute rental (mostly inference) amid token volume growth(from video generating). By 2026, with usage doubling (consistent with industry trends: token demand grows 2-5x YoY), assume OpenAI processes ~1,800 billion million-tokens annually $NVDA Blackwell at $0.02-$0.12 is $36B(most optimized) Rubin is projected to be at $0.01/million tokens or $18B annual Inference Cost vs $AMD Venice $0.0005/million tokens or $0.9B annual Inference Cost => Massive saving for OpenAI or anyone that are paying 80-90% Annual Bill for Inference compute. In short, it is unsustainable to pay this much rent vs owning for all current AI players for the medium to long term. Rubin excels in low-latency decode (if Groq integration from $20B deal in 2027-2028), but Venice dominates batch (80% of inference by 2030). Actual savings depend on deployment scale (OpenAI's 6GW AMD plans), electricity rates, and software maturity. If Rubin only hits $0.03, savings swell to $53.1B vs. $17.1B. 3. Will running Inference on Venice and future Gen slow down response generation in 2026 and beyond? Human perception of "fast enough" for chat, agents, search augmentation, summarization, coding assistance is roughly Meaning, EPYC may generate $100B a year on data center revenue, Hence $MSFT $AMZN $META $GOOGL OpenAI xAI and 42+ Countries are leaning AMD for Inference, because the cost saving is MASSIVE! 4. Regular users (you, me, people using ChatGPT, Claude, Gemini, Grok, Perplexity...) are extremely unlikely to notice any slowdown and in many cases might even experience slightly faster or more consistent response times if the industry heavily shifts toward AMD EPYC for inference. What actually happens when companies save massively on inference? When OpenAI , Anthropic , Gemini , Grok Meta .... save billions on the batch/enterprise/RAG layer using EPYC Venice, they typically do one or more of these things with the savings, none of which make your chat slower but enhancing their bottom line(Profit) ~Keep prices the same → make more profit ~Lower subscription prices / increase free tier limits ~Train bigger & better models more frequently ~Offer longer context windows ~Add more reasoning steps / tool calls / agents per query ~Improve multimodal capabilities ~Build more data centers / reduce throttling during peaks In practice the consumer experience usually gets better, not worse, when inference becomes dramatically cheaper. Prime example is $META leaning AMD heavily or currently AMD largest customer. or Grok 2 to Grok 3 heavily used AMD for Inference saving. And most Grok Users reported Groke responses snappier, not slower. 5. What does this mean for potential Revenue? Noted that TSMC is massively ramping 2nm supply for $AMD both MI450 and EPYC. EPYC Conservative projection: FY2025: $10.5B(best Est) FY2026: $16B FY2027: $29B FY2028: $49B FY2029: $75B FY2030: $100B Large customers: $META OpenAI $MSFT $AMZN $GOOGL xAI (Apple?) Smaller customer: $DELL $HPE $SMCI and 42+ other countries. The roadmap to $5 Trillion is very much inevitable as Inference Cost from Renting or owning $NVDA are too high, but $NVDA will still dominate Training market share, where MI families are likely to take 15-20% market share, but the TAM is also expanding Rapidly. Most Institutions are projecting $2-$3Trillion TAM by 2030. $NVDA said $4 Trillion. Dr. Lisa Su said $1 Trillion+ by 2030. So you decide on how much TAM. If you enjoy this kind of analysis, Slap the Like/Repost and Bookmark to please the X Algo as it is Free.99! If you want to support my work further, consider subscribe to see more in-depth analysis! Alright, that is it. Not Financial Advice!
3:09

$AMD $5 Trillion MC Is Inevitable Long Term👑 This thread will focus more on Inference! 2026 EPYC "Venice" $TSM 2nm to save Large GW Scale Inference by 40% more than Prior Turin gen. Context: EPYC Turin achieves ~$0.001 per million tokens for batch inference vs $0.02-$0.12/ million tokens as I wrote the thread below. Venice is going to lower cost down to $0.0005-$0.0006/Million Tokens. OpenAI spent roughly $20B on Inference and Training, where 80-90% of that was for Inference per Analysts. AKA Renting Compute is Expensive AF! In this thread, I want to focus on why most analysts and investors are underestimating the role EPYC "Venice" and future Gen on overall Data center revenue. And $TSM ramping up 2nm supply early is a confirmation that AMD will be a major buyer long term. I will also link the thread the Gap between AMD Analysts & Reality and 2nm Ramp Thread so you have more comprehensive view of what I'm writing here. Before I go into detail this is my 2026 Projection: AI GPUs: $35-$50B EPYC Data Center: $15B-$17B Client Segment: $12-$13B Gaming: $6B Embedded: $4B-$5B Total Revenue $70-$100B Non-GAAP net income $18B-$25B Non-GAAP EPS $10.97-$15.40 Foward P/E 55x-70x= $603-$1,078 AMD's Analysts are projecting $0 Revenue for MI450 and sluggish EPYC Growth. Meaning, all analysts are either full of 💩 or Sexist, you decide! Analysts are also projecting 0% growth on AMD "Secret Weapon" Chip as $MSFT said we are at significant Windows refresh and upgrade cycle. Do you think TSMC would allocate more 2nm supply to $AMD at $0 MI450 revenue and sluggish EPYC? 1. EPYC is going to be the leader in lowest Inference! Current Turin cost saving is 95% vs $NVDA or 98-99% on Inference cost when you factor in renting Inference compute from Amazon Web Services, Microsoft Azure, or $NVDA Neocloud pets. TSMC claimed: 10-15% higher performance at iso-power, 25-30% lower power at iso-speed, and ~15% higher transistor density compared to 3nm. This reduces operational expenses (energy, cooling) while increasing throughput per chip. EPYC Turin achieves ~$0.001 per million tokens for batch inference (via vLLM on models like Llama 3 70B), driven by high core counts and low hardware costs. EPYC Venice offers ~1.7x overall performance and up to 70% more compute capability per core, with up to 256 cores (512 threads). Enhanced vector/AI instructions and open-source firmware (openSIL) optimize for inference workloads. AMD Incorporates AI Engines (now part of AMD's XDNA) for on-chip acceleration, improving efficiency for low-latency and edge inference. This reduces reliance on discrete GPUs, lowering system complexity and TCO. Venice SKUs are projected at $3,000-$15,000 ($5,000 for 256-core flagship), far below NVIDIA Rubin ($50,000-$90,000) or AMD's own MI450 GPUs ($40,000-$50,000). High memory bandwidth (up to 1.6 TB/s) supports efficient batch inference. Venice is designed exactly for Large customers that want to lower Inference Cost and MI450 Helios is for Customers that want Training at lowest TCO, TDP as well as lower Upfront 1GW scale(Full build $35-$40B vs $NVDA $55B-$80B). 2. Real World Example: OpenAI's 2025 inference spend reached ~$20B, escalating to even higher total compute rental (mostly inference) amid token volume growth(from video generating). By 2026, with usage doubling (consistent with industry trends: token demand grows 2-5x YoY), assume OpenAI processes ~1,800 billion million-tokens annually $NVDA Blackwell at $0.02-$0.12 is $36B(most optimized) Rubin is projected to be at $0.01/million tokens or $18B annual Inference Cost vs $AMD Venice $0.0005/million tokens or $0.9B annual Inference Cost => Massive saving for OpenAI or anyone that are paying 80-90% Annual Bill for Inference compute. In short, it is unsustainable to pay this much rent vs owning for all current AI players for the medium to long term. Rubin excels in low-latency decode (if Groq integration from $20B deal in 2027-2028), but Venice dominates batch (80% of inference by 2030). Actual savings depend on deployment scale (OpenAI's 6GW AMD plans), electricity rates, and software maturity. If Rubin only hits $0.03, savings swell to $53.1B vs. $17.1B. 3. Will running Inference on Venice and future Gen slow down response generation in 2026 and beyond? Human perception of "fast enough" for chat, agents, search augmentation, summarization, coding assistance is roughly Meaning, EPYC may generate $100B a year on data center revenue, Hence $MSFT $AMZN $META $GOOGL OpenAI xAI and 42+ Countries are leaning AMD for Inference, because the cost saving is MASSIVE! 4. Regular users (you, me, people using ChatGPT, Claude, Gemini, Grok, Perplexity...) are extremely unlikely to notice any slowdown and in many cases might even experience slightly faster or more consistent response times if the industry heavily shifts toward AMD EPYC for inference. What actually happens when companies save massively on inference? When OpenAI , Anthropic , Gemini , Grok Meta .... save billions on the batch/enterprise/RAG layer using EPYC Venice, they typically do one or more of these things with the savings, none of which make your chat slower but enhancing their bottom line(Profit) ~Keep prices the same → make more profit ~Lower subscription prices / increase free tier limits ~Train bigger & better models more frequently ~Offer longer context windows ~Add more reasoning steps / tool calls / agents per query ~Improve multimodal capabilities ~Build more data centers / reduce throttling during peaks In practice the consumer experience usually gets better, not worse, when inference becomes dramatically cheaper. Prime example is $META leaning AMD heavily or currently AMD largest customer. or Grok 2 to Grok 3 heavily used AMD for Inference saving. And most Grok Users reported Groke responses snappier, not slower. 5. What does this mean for potential Revenue? Noted that TSMC is massively ramping 2nm supply for $AMD both MI450 and EPYC. EPYC Conservative projection: FY2025: $10.5B(best Est) FY2026: $16B FY2027: $29B FY2028: $49B FY2029: $75B FY2030: $100B Large customers: $META OpenAI $MSFT $AMZN $GOOGL xAI (Apple?) Smaller customer: $DELL $HPE $SMCI and 42+ other countries. The roadmap to $5 Trillion is very much inevitable as Inference Cost from Renting or owning $NVDA are too high, but $NVDA will still dominate Training market share, where MI families are likely to take 15-20% market share, but the TAM is also expanding Rapidly. Most Institutions are projecting $2-$3Trillion TAM by 2030. $NVDA said $4 Trillion. Dr. Lisa Su said $1 Trillion+ by 2030. So you decide on how much TAM. If you enjoy this kind of analysis, Slap the Like/Repost and Bookmark to please the X Algo as it is Free.99! If you want to support my work further, consider subscribe to see more in-depth analysis! Alright, that is it. Not Financial Advice!

Mike

102,223 views • 7 months ago

$AMD's heading to $5T MC LT| Lowest $/M tokens 🧵 The real reason why Institutions are FOMOing into AMD while other Semi stocks are underperforming ($NVDA $AVGO) Not Financial Advice! DYOR! Under Dr. Lisa Su’s leadership, AMD has transformed from a distant challenger into a formidable force in AI infrastructure, delivering the industry’s most compelling TCO story for high-volume inference. Her clear vision open ecosystems, aggressive annual roadmaps, rack-scale innovation, and relentless focus on tokens-per-dollar has positioned AMD’s Helios racks as the go-to solution for hyperscalers and AI natives struggling with exploding token costs, collapsing the cost down to $0.0003-$0.0005/M tokens. I will link various threads on this analysis to supply chain and wafer ratio if you are interested in understanding the full picture. In the last 3-4 months, explosive Agentic AI demand significantly increased Inference demand for Agentic AI models with 5-10 agents. If you are a listener of CNBC or Bloomberg, u should know enterprises and companies are complaining abt cost of token, and how it starts to spike up way too much to make sense. The fact that most data center today are run by $NVDA Chips, where the cost is way too high for Training or Inference. 1. Token cost Here are some quick comp, so u understand why $META OpenAI Anthropic $MSFT $AMZN Softbank $GOOGL and many more small to medium AI Natives are buying AMD CPUs and GPUs as much as they want, or pretty much AMD chips are sold out for the next 3-5 years. Inference (Cost per Million Tokens) ~$NVDA B200 / HGX: ~$0.02–$0.08 on optimized workloads (FP4/MXFP4, speculative decoding). Significant improvement over Hopper but still premium-priced. GB200 NVL72 rack-scale: $0.05–$0.25+ ~$AMD Helios Racks: $0.0003-$0.0005 per M tokens, dramatically lower than NVIDIA equivalents in owned infra. MI355X node-level: Up to 40% more tokens per dollar vs. competing solutions ( B200), driven by higher memory capacity (up to 288GB+ HBM), strong bandwidth, and lower acquisition costs. Training ~$NVDA Rubin Rack is estimated $0.7-$1.2/M Tokens ~$AMD Helios Rack is estimated $0.65-$1.0/M Tokens 2. Why Hyperscalers and AI Natives Are Choosing AMD Token consumption (especially Agentic) is outpacing even NVIDIA’s efficiency gains, making diversification mandatory for economic viability. Massive deals reflect this reality like $META, OpenAI, $MSFT, Softbank, $AMZN, Oracle, LumaAI, G42... Dr. Lisa Su’s Vision in Action: Since taking the helm, Su has driven AMD’s turnaround with disciplined execution, annual GPU cadence (MI300 → MI350 → MI400), full-stack software (ROCm 7), open ecosystems (UALink, OCP designs), and customer-centric rack-scale solutions like Helios. Her emphasis on “tokens per dollar” and TCO has turned AMD into the pragmatic choice for sustainable AI scaling. Power/Energy Efficiency: ~Helios Rack-level is estimated at 120kW-140kW with 50% more HBM4 where Inference and Training cost matter ~Rubin Rack-Level is estimated at 160kW-230kw AMD Helios shines in owned TCO, memory density, and energy flexibility at hyperscale. Cost to build 1GW data center 1GW Helios Rack full build is estimated $30-$35B 1GW Rubin Rack full build is estimated $45-$55B 3. Superior CPUs to pair with GPUs on massive scale 5-10-20GW Agentic AI. autonomous, multi-step workflows with orchestration, tool use, parallel agents, data movement, and enterprise integration has dramatically increased the importance of strong host CPUs alongside GPUs. This shifts the CPU-to-GPU ratio higher and makes balanced systems critical toward 1:1 to 5:1 as enterprises testing more than 5-10 agents. AMD EPYC Venice excels ~Leadership core density (up to 256 Zen 6 cores per socket) for running many agents in parallel, orchestration layers, and high-throughput control-plane tasks. ~Superior performance-per-core and power efficiency ( up to 2.1x higher perf/core and 2.26x better SPECpower vs. NVIDIA Grace in benchmarks). ~Tight integration in Helios: One Venice CPU + multiple MI450 GPUs per node, enabling efficient data feeding to GPUs ("zero-copy"), parallel execution, and full rack utilization for complex agentic loops. Hyperscalers (Meta, Microsoft, Amazon, Google, Softbank) and AI natives (OpenAI, Anthropic...) are adopting high-core EPYC at scale specifically for these agentic demands, as CPUs now handle a larger share of non-model work (orchestration, policy enforcement, tool calls). This complements AMD’s lower-cost GPUs for overall TCO wins. Conclusion: NVIDIA’s Vera Rubin cannot compete with a 2 years old EPYC Turin, but AMD under Dr. Lisa Su has engineered the lowest cost-per-million-tokens, highly competitive energy-efficient solutions, and superior CPU orchestration for agentic AI at scale with Helios. Dr. Su has championed this shift since at least 2023, foreseeing the rise of agentic workflows that demand far more orchestration, parallel agents, and balanced compute well before the industry fully embraced it. Her long-term vision of AI moving from simple prompts to always-on, multi-agent systems has driven AMD’s investments in high-core EPYC CPUs and integrated rack-scale solutions, perfectly positioning the company for today’s realities. Hyperscalers and AI natives effectively have no choice but to buy more AMD system for Agentic AI as leadership in economical, power-aware, high-volume internal + agentic use. However, due to supply constraints where Supply is far behind Demand, this makes multi-vendor reality along with in-house chips drive faster industry progress, lower overall costs, and better sustainability. Not Financial Advice! DYOR! Video source: Microsoft Build 2026
0:49

$AMD's heading to $5T MC LT| Lowest $/M tokens 🧵 The real reason why Institutions are FOMOing into AMD while other Semi stocks are underperforming ($NVDA $AVGO) Not Financial Advice! DYOR! Under Dr. Lisa Su’s leadership, AMD has transformed from a distant challenger into a formidable force in AI infrastructure, delivering the industry’s most compelling TCO story for high-volume inference. Her clear vision open ecosystems, aggressive annual roadmaps, rack-scale innovation, and relentless focus on tokens-per-dollar has positioned AMD’s Helios racks as the go-to solution for hyperscalers and AI natives struggling with exploding token costs, collapsing the cost down to $0.0003-$0.0005/M tokens. I will link various threads on this analysis to supply chain and wafer ratio if you are interested in understanding the full picture. In the last 3-4 months, explosive Agentic AI demand significantly increased Inference demand for Agentic AI models with 5-10 agents. If you are a listener of CNBC or Bloomberg, u should know enterprises and companies are complaining abt cost of token, and how it starts to spike up way too much to make sense. The fact that most data center today are run by $NVDA Chips, where the cost is way too high for Training or Inference. 1. Token cost Here are some quick comp, so u understand why $META OpenAI Anthropic $MSFT $AMZN Softbank $GOOGL and many more small to medium AI Natives are buying AMD CPUs and GPUs as much as they want, or pretty much AMD chips are sold out for the next 3-5 years. Inference (Cost per Million Tokens) ~$NVDA B200 / HGX: ~$0.02–$0.08 on optimized workloads (FP4/MXFP4, speculative decoding). Significant improvement over Hopper but still premium-priced. GB200 NVL72 rack-scale: $0.05–$0.25+ ~$AMD Helios Racks: $0.0003-$0.0005 per M tokens, dramatically lower than NVIDIA equivalents in owned infra. MI355X node-level: Up to 40% more tokens per dollar vs. competing solutions ( B200), driven by higher memory capacity (up to 288GB+ HBM), strong bandwidth, and lower acquisition costs. Training ~$NVDA Rubin Rack is estimated $0.7-$1.2/M Tokens ~$AMD Helios Rack is estimated $0.65-$1.0/M Tokens 2. Why Hyperscalers and AI Natives Are Choosing AMD Token consumption (especially Agentic) is outpacing even NVIDIA’s efficiency gains, making diversification mandatory for economic viability. Massive deals reflect this reality like $META, OpenAI, $MSFT, Softbank, $AMZN, Oracle, LumaAI, G42... Dr. Lisa Su’s Vision in Action: Since taking the helm, Su has driven AMD’s turnaround with disciplined execution, annual GPU cadence (MI300 → MI350 → MI400), full-stack software (ROCm 7), open ecosystems (UALink, OCP designs), and customer-centric rack-scale solutions like Helios. Her emphasis on “tokens per dollar” and TCO has turned AMD into the pragmatic choice for sustainable AI scaling. Power/Energy Efficiency: ~Helios Rack-level is estimated at 120kW-140kW with 50% more HBM4 where Inference and Training cost matter ~Rubin Rack-Level is estimated at 160kW-230kw AMD Helios shines in owned TCO, memory density, and energy flexibility at hyperscale. Cost to build 1GW data center 1GW Helios Rack full build is estimated $30-$35B 1GW Rubin Rack full build is estimated $45-$55B 3. Superior CPUs to pair with GPUs on massive scale 5-10-20GW Agentic AI. autonomous, multi-step workflows with orchestration, tool use, parallel agents, data movement, and enterprise integration has dramatically increased the importance of strong host CPUs alongside GPUs. This shifts the CPU-to-GPU ratio higher and makes balanced systems critical toward 1:1 to 5:1 as enterprises testing more than 5-10 agents. AMD EPYC Venice excels ~Leadership core density (up to 256 Zen 6 cores per socket) for running many agents in parallel, orchestration layers, and high-throughput control-plane tasks. ~Superior performance-per-core and power efficiency ( up to 2.1x higher perf/core and 2.26x better SPECpower vs. NVIDIA Grace in benchmarks). ~Tight integration in Helios: One Venice CPU + multiple MI450 GPUs per node, enabling efficient data feeding to GPUs ("zero-copy"), parallel execution, and full rack utilization for complex agentic loops. Hyperscalers (Meta, Microsoft, Amazon, Google, Softbank) and AI natives (OpenAI, Anthropic...) are adopting high-core EPYC at scale specifically for these agentic demands, as CPUs now handle a larger share of non-model work (orchestration, policy enforcement, tool calls). This complements AMD’s lower-cost GPUs for overall TCO wins. Conclusion: NVIDIA’s Vera Rubin cannot compete with a 2 years old EPYC Turin, but AMD under Dr. Lisa Su has engineered the lowest cost-per-million-tokens, highly competitive energy-efficient solutions, and superior CPU orchestration for agentic AI at scale with Helios. Dr. Su has championed this shift since at least 2023, foreseeing the rise of agentic workflows that demand far more orchestration, parallel agents, and balanced compute well before the industry fully embraced it. Her long-term vision of AI moving from simple prompts to always-on, multi-agent systems has driven AMD’s investments in high-core EPYC CPUs and integrated rack-scale solutions, perfectly positioning the company for today’s realities. Hyperscalers and AI natives effectively have no choice but to buy more AMD system for Agentic AI as leadership in economical, power-aware, high-volume internal + agentic use. However, due to supply constraints where Supply is far behind Demand, this makes multi-vendor reality along with in-house chips drive faster industry progress, lower overall costs, and better sustainability. Not Financial Advice! DYOR! Video source: Microsoft Build 2026

Mike

145,992 views • 2 months ago

$AMD is ready to break $1 Trillion MC| $TSM 2nm🧵 TLDR FY 2026(Excluding China AI Revenue) AI GPUs: $35-$50B EPYC Data Center: $15B-$17B Client Segment: $12-$13B Gaming: $6B Embedded: $4B-$5B Total Revenue $70-$100B Non-GAAP net income $18B-$25B Non-GAAP EPS $10.97-$15.40 Foward P/E 55x-70x= $603-$1,078 The semiconductor industry is at a pivotal juncture, with advanced process nodes like TSMC's 2nm technology becoming the battleground for leadership in artificial intelligence and high-performance computing (HPC). Amid this landscape, AMD stands poised to secure early production and higher allocation of its Venice (EPYC ) and MI450 (Instinct GPUs) on TSMC's 2nm process. This strategic advantage is not merely a product of timing but a culmination of a robust partnership, market demand, technical superiority, and geopolitical dynamics. The AI and HPC markets are experiencing unprecedented growth, with inference workloads projected to constitute 80-90% of AI compute by 2030. AMD's EPYC processors and Instinct GPUs are uniquely positioned to capitalize on this trend, particularly given the demand from hyperscalers such as OpenAI , $META , $MSFT, $AMZN, and $ORCL. With $TSM starting 2nm Mass Production in Taiwan is ensuring AMD to meet FY2026 $70B to $100B revenue, driven by non-GAAP net income of $18B to $25B highlights the scale of this opportunity, starkly contrasting with analyst revenue consensus of $39-$45B. This discrepancy arises from analysts' failure to account for major orders, notably from OpenAI(Today SoftBank secured OpenAI a massive cash balance of $55-$62B).OpenAI is raising $100B, so this left $77B from UAE, Saudi, $MSFT, and others. $AMD is on track to receive higher allocation of EPYC Venice and Mi450 in 2026. AMD's acquisition of Xilinx has significantly strengthened its position in AI inference, particularly through adaptive computing technologies like FPGA-based AI Engines. The upcoming Zen 6 "Venice" generation (on TSMC 2nm, launching with MI450 in 2026) promises ~1.7× performance uplift, enhanced vector/AI capabilities, greater thread density, and open firmware innovations positioning EPYC to maintain its inference leadership while powering massive hybrid AI superclusters. TSMC's Fab 22 in Kaohsiung, Taiwan, is now the epicenter of 2nm mass production, a earlier strategic move to meet soaring demand from $AMD and $AAPL. Early production slots are typically reserved for customers with the highest revenue potential and strategic importance. AMD's early tape-out of Venice and the MI450's role as the first AMD GPU on 2nm place it at the forefront of this allocation. The 2nm process offers 10-15% higher performance or 25-30% lower power use compared to 3nm, a critical advantage for AI and HPC applications(TSMC claimed) Moreover, TSMC's recent 20% yield improvement in Versal production, as mentioned in related discussions, indicates efficient scaling. Higher yields translate to more chips produced per wafer, reducing costs and increasing allocation for key customers like AMD. This efficiency is particularly important given the aggressive timelines of customers like OpenAI, who require rapid scaling to meet their computational needs. The reopening of the China market adds another layer of demand pressure. Vendors and hyperscalers are begging for allocation of AMD's MI308X, MI300X, and MI355X, and the 2nm capacity will be critical to meet this need. TSMC's early production of 2nm ensures AMD can capitalize on this opportunity, securing higher allocation to fulfill these orders. Dr. Lisa Su's emphasis on disciplined supply chain planning for multiple gigawatt-scale customers, such as OpenAI, demonstrates AMD's readiness to scale. TSMC's confidence in AMD's ability to absorb this capacity is evident in the early 2nm production allocation. This discipline is particularly important in a market where demand outstrips supply by 10-12x. TSMC's competitors, such as Samsung and Intel, are still in the early stages of their 2nm and equivalent processes. Samsung's 2nm GAA transistors and Intel's 18A process are not yet in mass production, giving TSMC and AMD a first-mover advantage. Nvidia's acquisition of Groq Inc. is a defensive move to diversify into inference, but it does not immediately address the 2nm gap. AMD EPYC Venice and future Gen are already ahead of lowest cost for Inference along with MI450 has TCO of $0.65 to $1.00 per million inference tokens, significantly lower than Nvidia's Rubik (H2 2026) at $0.70 to $1.20 and Broadcom's XPU (2027-2029) at $0.70 to $1.30. Additionally, the MI450's TDP is estimated at 1000-1800W, compared to Nvidia's 2300-3600W (Ultra), reducing operational costs and energy consumption(TSMC 2nm vs TSMC 3nm). The MI450 features 432GB of HBM4 memory and 19.6 TB/s bandwidth, surpassing Nvidia's Rubik (288GB HBM4, 16 TB/s) and Broadcom's XPU (192/256GB HBM4, 7 TB/s est). This enhanced memory and bandwidth capacity is essential for handling the complex, data-intensive workloads of large language models and other AI applications. AMD's full-stack vision, combining EPYC hosts with Instinct accelerators, offers the lowest total cost of ownership (TCO) and thermal design power (TDP). This synergy is unbeatable for both training and inference, further justifying TSMC's prioritization. The 2nm process amplifies these advantages, ensuring AMD can maintain its competitive edge over rivals like Nvidia, whose Rubin GPUs are still on N3P (a 3nm derivative). Today, TSMC just secured $AMD to join the top 10 largest companies in the world as it begins 2nm mass production in Taiwan. AMD and Apple are to receive highest allocation. The long-standing partnership with TSMC, massive demand from hyperscalers, technical advantages of 2nm, and disciplined supply chain planning all point to AMD's favored position. The 2nm process's early mass production at Fab 22, combined with AMD's revenue potential and competitive edge, justifies TSMC's prioritization. This allocation is critical for AMD to meet aggressive demand, capture market share, and solidify its position as a leader in AI and HPC, especially in the inference-dominated future. Dr. Lisa Su "We will multiple customers/hyperscalers at GW scale" Not Financial Advice!
2:38

$AMD is ready to break $1 Trillion MC| $TSM 2nm🧵 TLDR FY 2026(Excluding China AI Revenue) AI GPUs: $35-$50B EPYC Data Center: $15B-$17B Client Segment: $12-$13B Gaming: $6B Embedded: $4B-$5B Total Revenue $70-$100B Non-GAAP net income $18B-$25B Non-GAAP EPS $10.97-$15.40 Foward P/E 55x-70x= $603-$1,078 The semiconductor industry is at a pivotal juncture, with advanced process nodes like TSMC's 2nm technology becoming the battleground for leadership in artificial intelligence and high-performance computing (HPC). Amid this landscape, AMD stands poised to secure early production and higher allocation of its Venice (EPYC ) and MI450 (Instinct GPUs) on TSMC's 2nm process. This strategic advantage is not merely a product of timing but a culmination of a robust partnership, market demand, technical superiority, and geopolitical dynamics. The AI and HPC markets are experiencing unprecedented growth, with inference workloads projected to constitute 80-90% of AI compute by 2030. AMD's EPYC processors and Instinct GPUs are uniquely positioned to capitalize on this trend, particularly given the demand from hyperscalers such as OpenAI , $META , $MSFT, $AMZN, and $ORCL. With $TSM starting 2nm Mass Production in Taiwan is ensuring AMD to meet FY2026 $70B to $100B revenue, driven by non-GAAP net income of $18B to $25B highlights the scale of this opportunity, starkly contrasting with analyst revenue consensus of $39-$45B. This discrepancy arises from analysts' failure to account for major orders, notably from OpenAI(Today SoftBank secured OpenAI a massive cash balance of $55-$62B).OpenAI is raising $100B, so this left $77B from UAE, Saudi, $MSFT, and others. $AMD is on track to receive higher allocation of EPYC Venice and Mi450 in 2026. AMD's acquisition of Xilinx has significantly strengthened its position in AI inference, particularly through adaptive computing technologies like FPGA-based AI Engines. The upcoming Zen 6 "Venice" generation (on TSMC 2nm, launching with MI450 in 2026) promises ~1.7× performance uplift, enhanced vector/AI capabilities, greater thread density, and open firmware innovations positioning EPYC to maintain its inference leadership while powering massive hybrid AI superclusters. TSMC's Fab 22 in Kaohsiung, Taiwan, is now the epicenter of 2nm mass production, a earlier strategic move to meet soaring demand from $AMD and $AAPL. Early production slots are typically reserved for customers with the highest revenue potential and strategic importance. AMD's early tape-out of Venice and the MI450's role as the first AMD GPU on 2nm place it at the forefront of this allocation. The 2nm process offers 10-15% higher performance or 25-30% lower power use compared to 3nm, a critical advantage for AI and HPC applications(TSMC claimed) Moreover, TSMC's recent 20% yield improvement in Versal production, as mentioned in related discussions, indicates efficient scaling. Higher yields translate to more chips produced per wafer, reducing costs and increasing allocation for key customers like AMD. This efficiency is particularly important given the aggressive timelines of customers like OpenAI, who require rapid scaling to meet their computational needs. The reopening of the China market adds another layer of demand pressure. Vendors and hyperscalers are begging for allocation of AMD's MI308X, MI300X, and MI355X, and the 2nm capacity will be critical to meet this need. TSMC's early production of 2nm ensures AMD can capitalize on this opportunity, securing higher allocation to fulfill these orders. Dr. Lisa Su's emphasis on disciplined supply chain planning for multiple gigawatt-scale customers, such as OpenAI, demonstrates AMD's readiness to scale. TSMC's confidence in AMD's ability to absorb this capacity is evident in the early 2nm production allocation. This discipline is particularly important in a market where demand outstrips supply by 10-12x. TSMC's competitors, such as Samsung and Intel, are still in the early stages of their 2nm and equivalent processes. Samsung's 2nm GAA transistors and Intel's 18A process are not yet in mass production, giving TSMC and AMD a first-mover advantage. Nvidia's acquisition of Groq Inc. is a defensive move to diversify into inference, but it does not immediately address the 2nm gap. AMD EPYC Venice and future Gen are already ahead of lowest cost for Inference along with MI450 has TCO of $0.65 to $1.00 per million inference tokens, significantly lower than Nvidia's Rubik (H2 2026) at $0.70 to $1.20 and Broadcom's XPU (2027-2029) at $0.70 to $1.30. Additionally, the MI450's TDP is estimated at 1000-1800W, compared to Nvidia's 2300-3600W (Ultra), reducing operational costs and energy consumption(TSMC 2nm vs TSMC 3nm). The MI450 features 432GB of HBM4 memory and 19.6 TB/s bandwidth, surpassing Nvidia's Rubik (288GB HBM4, 16 TB/s) and Broadcom's XPU (192/256GB HBM4, 7 TB/s est). This enhanced memory and bandwidth capacity is essential for handling the complex, data-intensive workloads of large language models and other AI applications. AMD's full-stack vision, combining EPYC hosts with Instinct accelerators, offers the lowest total cost of ownership (TCO) and thermal design power (TDP). This synergy is unbeatable for both training and inference, further justifying TSMC's prioritization. The 2nm process amplifies these advantages, ensuring AMD can maintain its competitive edge over rivals like Nvidia, whose Rubin GPUs are still on N3P (a 3nm derivative). Today, TSMC just secured $AMD to join the top 10 largest companies in the world as it begins 2nm mass production in Taiwan. AMD and Apple are to receive highest allocation. The long-standing partnership with TSMC, massive demand from hyperscalers, technical advantages of 2nm, and disciplined supply chain planning all point to AMD's favored position. The 2nm process's early mass production at Fab 22, combined with AMD's revenue potential and competitive edge, justifies TSMC's prioritization. This allocation is critical for AMD to meet aggressive demand, capture market share, and solidify its position as a leader in AI and HPC, especially in the inference-dominated future. Dr. Lisa Su "We will multiple customers/hyperscalers at GW scale" Not Financial Advice!

Mike

43,219 views • 8 months ago

$AMD Massive Rotation from $NVDA $INTC🧵 Not Financial Advice! DYOR! 5-10 minutes before the bell today, last trading day of May 2026, massive rotation out of $INTC and $NVDA into $AMD. I wrote this thread this morning on what $TSM said on Energy Efficiency is now TOP Priotity and why AMD is the biggest winner. Of course I did not have influence on this rebalancing, I was just pointing out why Dr. Su saw this coming years ago. (Check the picture to understand more). I been talking about Agentic AI for like 3-4 years now. OpenClaw broke the CPU:GPU Ratio 1:4 narrative to 1:1 to 5:1 in late Jan and Feb 2026. I will link various threads where you can understand the full picture from supply chain, to TSMC expansion, and different Wafer Ratio for EPYC Venice and MI455X. Energy efficiency is a structural, long-term driver behind institutional rotation from $NVDA and $INTC into $AMD (with spillover strength in $AVGO for complementary networking/custom silicon). This isn't just short-term rebalancing, it's a massive bet on the shift from AI training (performance-at-any-cost) to inference, deployment, and embodied/agentic systems (where total cost of ownership, power draw, and scalability dominate). Precisely What I been writing about $AMD for years now, probably at least more than 5,000 threads.This is the FOMO from Institutions to own $AMD. Do know that AMD is the least owned Semi Stock among vs Peers. AI infrastructure is moving beyond massive training clusters to widespread inference for Agentic AI (running models 24/7) and embodied AI (robots, autonomous agents, edge devices). These workloads prioritize: ~Tokens-per-watt and performance-per-watt ~Lower total power consumption for data centers facing grid constraints ~Better economics at scale (cost-per-token, TCO) ~Thermal and power efficiency for on-device/robotics use Hyperscalers are now thinking more about Margin, Profitability, and $/M Tokens At $516/share. AMD Fwd PEG Ratio is still 35/100+= 0.35 AKA very cheap IMO for the growth and potential. A. Why institutions rotated out of $NVDA? Because Agentic AI is going to dominated by CPUs for years to come, moving violently to 5-10-20:1 CPU:GPU Ratio as enterprises are demanding more than 10-20 agents to run tasks. Now, that does not mean training is going away, Inference is just going to grow much faster. B. Why instiutitons rotated out of $INTC? Because AMD x86 unit share is only at 30-31% but Revenue share is already at 46.2% according to Mercury Research. And Dr. Su wants 50-60% market share, and that would mean 60-70%+ Revenue share where the CPUs TAM Is now already at $200B in 2026 and projected to be $500B by 2030. C. Why $AMD? Because AMD secured meaningful 2nm Capacity, Advanced Packaging and Memory through 2027-2028. And TSMC is expanding 2 primary 2nm Fabs toward 60-65k WPM each, and speeding up 5 2nm Fabs in Taiwan. With total up to 12 2nm Fabs through 2027/2028. 2nm Capacity is expected to be 140k+ WPM toward end of 2026, and 220-240k WPM by end of 2027. Apple has secured 35-45k WPM. And AMD does not have to worry about allocation competition until late 2027 from $AVGO for $META and $GOOGL(This may change) D. Agentic AI will evolve to 24/7 Autonomous Agent, and that will become the foundational layer for Robotic or Physical AI. Agentic AI (autonomous systems that plan, reason, use tools, self-correct, pursue long-horizon goals, and adapt) provides the high-level cognitive architecture. It turns raw perception and low-level control into useful, general-purpose behavior in the physical world. Physical AI (or Embodied AI) refers to AI that senses, understands, and acts directly in the real world through robots, actuators, and sensors. Agentic capabilities are what make this scalable and useful beyond narrow, scripted tasks. Reactive/programmed machines → To proactive, goal-oriented autonomous agents. How does this work? Autonomous Agent layer is the brain ~Vision-Language-Action models or robotics foundation models. ~Agentic loops: Planning, chain-of-thought reasoning, reflection, tool use (simulators, APIs), multi-step task decomposition. ~Persistent 24/7 operation with Memory, world modeling, continuous learning. Institutions may not like $AMD from 2022-2025, but they cannot stop this evolution and it is inevitable. Part of my main thesis for AMD to get to $5 Trillion Market Cap Long Term. Conclusion: Institutions are rotating capital toward AMD not merely for tactical rebalancing, but because Dr. Lisa Su and her team anticipated this exact inflection years in advance and have been methodically engineering AMD’s platform to dominate it. Dr. Su has long championed the convergence of Agentic AI as the high-level cognitive foundation for Physical AI and robotics. As far back as her 2023/2024 CES keynote and earlier strategic commentary, she described Physical AI (including humanoid robotics and edge autonomy) as “the next big thing”; a natural extension of agentic workflows moving from digital reasoning to real-world action. She emphasized that enabling persistent, 24/7 autonomous agents requires a full-stack approach: high-performance CPUs for orchestration and motion control, dedicated accelerators for real-time vision and multimodal inference, and open software ecosystems for rapid development. This vision aligns precisely with the structural drivers we’ve discussed. As AI shifts from training to massive-scale inference and embodiment, energy efficiency, total cost of ownership, and heterogeneous compute become first-order advantages. AMD’s Instinct MI350/MI355 series, Ryzen AI Embedded processors, and EPYC platforms deliver superior performance-per-watt and balanced CPU + GPU + NPU integration ideal for power-constrained robots that must run sophisticated agentic reasoning loops without excessive thermal or battery drain. Dr. Su has repeatedly highlighted the rising importance of CPUs in agentic systems (moving toward 1:1 or even CPU-heavy ratios with GPUs), positioning AMD’s strengths in orchestration, memory handling, and efficiency as critical for the next phase of growth. AMD is engineered for the deployment realities of embodied agents: scalable, efficient, and deployable at the edge and in physical systems. The institutional flows out of NVDA and INTC into AMD reflect recognition of this prepared leadership. Dr. Su didn’t just see the future of Agentic AI powering robotics, she has spent years building the silicon, software, and partnerships to make it practical and economically viable. This rotation signals confidence that the companies best positioned for the physical, always-on intelligence layer will capture the highest-volume opportunities in the coming decade. Not Financial Advice! DYOR!
2:25

$AMD Massive Rotation from $NVDA $INTC🧵 Not Financial Advice! DYOR! 5-10 minutes before the bell today, last trading day of May 2026, massive rotation out of $INTC and $NVDA into $AMD. I wrote this thread this morning on what $TSM said on Energy Efficiency is now TOP Priotity and why AMD is the biggest winner. Of course I did not have influence on this rebalancing, I was just pointing out why Dr. Su saw this coming years ago. (Check the picture to understand more). I been talking about Agentic AI for like 3-4 years now. OpenClaw broke the CPU:GPU Ratio 1:4 narrative to 1:1 to 5:1 in late Jan and Feb 2026. I will link various threads where you can understand the full picture from supply chain, to TSMC expansion, and different Wafer Ratio for EPYC Venice and MI455X. Energy efficiency is a structural, long-term driver behind institutional rotation from $NVDA and $INTC into $AMD (with spillover strength in $AVGO for complementary networking/custom silicon). This isn't just short-term rebalancing, it's a massive bet on the shift from AI training (performance-at-any-cost) to inference, deployment, and embodied/agentic systems (where total cost of ownership, power draw, and scalability dominate). Precisely What I been writing about $AMD for years now, probably at least more than 5,000 threads.This is the FOMO from Institutions to own $AMD. Do know that AMD is the least owned Semi Stock among vs Peers. AI infrastructure is moving beyond massive training clusters to widespread inference for Agentic AI (running models 24/7) and embodied AI (robots, autonomous agents, edge devices). These workloads prioritize: ~Tokens-per-watt and performance-per-watt ~Lower total power consumption for data centers facing grid constraints ~Better economics at scale (cost-per-token, TCO) ~Thermal and power efficiency for on-device/robotics use Hyperscalers are now thinking more about Margin, Profitability, and $/M Tokens At $516/share. AMD Fwd PEG Ratio is still 35/100+= 0.35 AKA very cheap IMO for the growth and potential. A. Why institutions rotated out of $NVDA? Because Agentic AI is going to dominated by CPUs for years to come, moving violently to 5-10-20:1 CPU:GPU Ratio as enterprises are demanding more than 10-20 agents to run tasks. Now, that does not mean training is going away, Inference is just going to grow much faster. B. Why instiutitons rotated out of $INTC? Because AMD x86 unit share is only at 30-31% but Revenue share is already at 46.2% according to Mercury Research. And Dr. Su wants 50-60% market share, and that would mean 60-70%+ Revenue share where the CPUs TAM Is now already at $200B in 2026 and projected to be $500B by 2030. C. Why $AMD? Because AMD secured meaningful 2nm Capacity, Advanced Packaging and Memory through 2027-2028. And TSMC is expanding 2 primary 2nm Fabs toward 60-65k WPM each, and speeding up 5 2nm Fabs in Taiwan. With total up to 12 2nm Fabs through 2027/2028. 2nm Capacity is expected to be 140k+ WPM toward end of 2026, and 220-240k WPM by end of 2027. Apple has secured 35-45k WPM. And AMD does not have to worry about allocation competition until late 2027 from $AVGO for $META and $GOOGL(This may change) D. Agentic AI will evolve to 24/7 Autonomous Agent, and that will become the foundational layer for Robotic or Physical AI. Agentic AI (autonomous systems that plan, reason, use tools, self-correct, pursue long-horizon goals, and adapt) provides the high-level cognitive architecture. It turns raw perception and low-level control into useful, general-purpose behavior in the physical world. Physical AI (or Embodied AI) refers to AI that senses, understands, and acts directly in the real world through robots, actuators, and sensors. Agentic capabilities are what make this scalable and useful beyond narrow, scripted tasks. Reactive/programmed machines → To proactive, goal-oriented autonomous agents. How does this work? Autonomous Agent layer is the brain ~Vision-Language-Action models or robotics foundation models. ~Agentic loops: Planning, chain-of-thought reasoning, reflection, tool use (simulators, APIs), multi-step task decomposition. ~Persistent 24/7 operation with Memory, world modeling, continuous learning. Institutions may not like $AMD from 2022-2025, but they cannot stop this evolution and it is inevitable. Part of my main thesis for AMD to get to $5 Trillion Market Cap Long Term. Conclusion: Institutions are rotating capital toward AMD not merely for tactical rebalancing, but because Dr. Lisa Su and her team anticipated this exact inflection years in advance and have been methodically engineering AMD’s platform to dominate it. Dr. Su has long championed the convergence of Agentic AI as the high-level cognitive foundation for Physical AI and robotics. As far back as her 2023/2024 CES keynote and earlier strategic commentary, she described Physical AI (including humanoid robotics and edge autonomy) as “the next big thing”; a natural extension of agentic workflows moving from digital reasoning to real-world action. She emphasized that enabling persistent, 24/7 autonomous agents requires a full-stack approach: high-performance CPUs for orchestration and motion control, dedicated accelerators for real-time vision and multimodal inference, and open software ecosystems for rapid development. This vision aligns precisely with the structural drivers we’ve discussed. As AI shifts from training to massive-scale inference and embodiment, energy efficiency, total cost of ownership, and heterogeneous compute become first-order advantages. AMD’s Instinct MI350/MI355 series, Ryzen AI Embedded processors, and EPYC platforms deliver superior performance-per-watt and balanced CPU + GPU + NPU integration ideal for power-constrained robots that must run sophisticated agentic reasoning loops without excessive thermal or battery drain. Dr. Su has repeatedly highlighted the rising importance of CPUs in agentic systems (moving toward 1:1 or even CPU-heavy ratios with GPUs), positioning AMD’s strengths in orchestration, memory handling, and efficiency as critical for the next phase of growth. AMD is engineered for the deployment realities of embodied agents: scalable, efficient, and deployable at the edge and in physical systems. The institutional flows out of NVDA and INTC into AMD reflect recognition of this prepared leadership. Dr. Su didn’t just see the future of Agentic AI powering robotics, she has spent years building the silicon, software, and partnerships to make it practical and economically viable. This rotation signals confidence that the companies best positioned for the physical, always-on intelligence layer will capture the highest-volume opportunities in the coming decade. Not Financial Advice! DYOR!

Mike

104,109 views • 3 months ago

$AMD $NVDA & the AMD Bear SemiAnalysis 🧵 Here are some facts: $META allocated 42% AI GPUs to $AMD OpenAI allocated 6GW(38%) to $AMD 1. Model-Specific Bias: Llama 3.3 70B graph favored NVIDIA due to TRT-LLM optimizations, highlighting throughput and latency where Blackwell excels. In contrast, the GPT-OSS 120B chart shifts focus to cost and interactivity, where MI355X shines. This selective model choice clearly suggests SemiAnalysis tailors benchmarks to reinforce narratives—NVIDIA’s dominance in speed (Llama 3.3) and AMD’s niche in cost (GPT-OSS). GPT-OSS 120B, with its sparse attention mechanisms (similar to DeepSeek-V3.2-Exp), shows AMD’s CDNA 4 architecture, while Llama 3.3’s dense attention favors NVIDIA’s Tensor Cores. SemiAnalysis’ decision to emphasize Llama 3.3 initially could reflect its AMD bear stance. 2. The way Data is presented The Llama 3.3 graph focused on raw performance metrics (throughput vs. latency), downplaying cost, where AMD holds an edge. This new chart, buried in follow-up posts, reveals AMD’s strength but receives less prominence, suggesting a curated narrative. Labeling variability (e.g., B200 with/without TRT) and the lack of uniform scaling across graphs indicate potential cherry-picking of configurations to favor NVIDIA’s optimized setups. 3. Historical Context: SemiAnalysis’ past critiques of AMD’s R&D and ROCm (web results from May 2025) align with a bearish outlook. Their own hype/brand around NVIDIA’s 15x ROI contrasts with muted coverage of AMD’s cost advantages, reinforcing bias. Despite AMD’s participation in InferenceMAX, the benchmark’s framing (e.g., prioritizing Blackwell’s ROI) reflect SemiAnalysis’ market predictions rather than balanced analysis. Lastly, AMD’s Instinct MI355X proves superior in inference and cost per million tokens for the GPT-OSS 120B model, offering a 25% cost advantage over NVIDIA’s H200 at moderate-to-high interactivity levels. This efficiency, driven by AMD’s memory bandwidth and FP4 support, makes it a better choice for cost-sensitive, multi-user deployments over a three-year horizon. However, SemiAnalysis’ sole focus(presentation graph) on Llama 3.3—where NVIDIA excels demonstrates a pattern of cherry-picking models and data to favor NVIDIA , consistent with its historical AMD bearish stance. This selective presentation risks misleading stakeholders by overshadowing AMD economic strengths. My personal take: I would trust Dr. Lisa Su, and Greg Brockman Sam Altman take on AMD and how they viewed and allocated 6GW for AMD over SemiAnalysis . At the end of the day, Large customers pay when it works. $Meta allocated 42% AI GPUs to $AMD for a reason. And the "secret weapon" will improve energy consumption by 20-50%, meaning at 6GW, OpenAI would be able to deploy 25-50% more MI450 at a much better cost advantage, higher memory bandwidth, and the queen of Inference! Oh and ROCm 8 is expected to be on par with CUDA in 2026.
2:46

$AMD $NVDA & the AMD Bear SemiAnalysis 🧵 Here are some facts: $META allocated 42% AI GPUs to $AMD OpenAI allocated 6GW(38%) to $AMD 1. Model-Specific Bias: Llama 3.3 70B graph favored NVIDIA due to TRT-LLM optimizations, highlighting throughput and latency where Blackwell excels. In contrast, the GPT-OSS 120B chart shifts focus to cost and interactivity, where MI355X shines. This selective model choice clearly suggests SemiAnalysis tailors benchmarks to reinforce narratives—NVIDIA’s dominance in speed (Llama 3.3) and AMD’s niche in cost (GPT-OSS). GPT-OSS 120B, with its sparse attention mechanisms (similar to DeepSeek-V3.2-Exp), shows AMD’s CDNA 4 architecture, while Llama 3.3’s dense attention favors NVIDIA’s Tensor Cores. SemiAnalysis’ decision to emphasize Llama 3.3 initially could reflect its AMD bear stance. 2. The way Data is presented The Llama 3.3 graph focused on raw performance metrics (throughput vs. latency), downplaying cost, where AMD holds an edge. This new chart, buried in follow-up posts, reveals AMD’s strength but receives less prominence, suggesting a curated narrative. Labeling variability (e.g., B200 with/without TRT) and the lack of uniform scaling across graphs indicate potential cherry-picking of configurations to favor NVIDIA’s optimized setups. 3. Historical Context: SemiAnalysis’ past critiques of AMD’s R&D and ROCm (web results from May 2025) align with a bearish outlook. Their own hype/brand around NVIDIA’s 15x ROI contrasts with muted coverage of AMD’s cost advantages, reinforcing bias. Despite AMD’s participation in InferenceMAX, the benchmark’s framing (e.g., prioritizing Blackwell’s ROI) reflect SemiAnalysis’ market predictions rather than balanced analysis. Lastly, AMD’s Instinct MI355X proves superior in inference and cost per million tokens for the GPT-OSS 120B model, offering a 25% cost advantage over NVIDIA’s H200 at moderate-to-high interactivity levels. This efficiency, driven by AMD’s memory bandwidth and FP4 support, makes it a better choice for cost-sensitive, multi-user deployments over a three-year horizon. However, SemiAnalysis’ sole focus(presentation graph) on Llama 3.3—where NVIDIA excels demonstrates a pattern of cherry-picking models and data to favor NVIDIA , consistent with its historical AMD bearish stance. This selective presentation risks misleading stakeholders by overshadowing AMD economic strengths. My personal take: I would trust Dr. Lisa Su, and Greg Brockman Sam Altman take on AMD and how they viewed and allocated 6GW for AMD over SemiAnalysis . At the end of the day, Large customers pay when it works. $Meta allocated 42% AI GPUs to $AMD for a reason. And the "secret weapon" will improve energy consumption by 20-50%, meaning at 6GW, OpenAI would be able to deploy 25-50% more MI450 at a much better cost advantage, higher memory bandwidth, and the queen of Inference! Oh and ROCm 8 is expected to be on par with CUDA in 2026.

Mike

104,194 views • 10 months ago

Jensen Huang is warning that China is winning the AI energy race. $NVDA $AMD $IREN $NBIS 🇨🇳 🇺🇸 🇪🇺 The US is generations ahead on chips, but that doesn't matter if there is no energy to power them. Europe is completely irrelevant in AI and doomed already. It takes 10 years to build a new data center in Germany, as there is no power, and they are turning off nuclear power plants. The US won't let China win, so we could be at the beginning stages of an energy boom. The question is which stocks will benefit from this boom the most?
6:50

Jensen Huang is warning that China is winning the AI energy race. $NVDA $AMD $IREN $NBIS 🇨🇳 🇺🇸 🇪🇺 The US is generations ahead on chips, but that doesn't matter if there is no energy to power them. Europe is completely irrelevant in AI and doomed already. It takes 10 years to build a new data center in Germany, as there is no power, and they are turning off nuclear power plants. The US won't let China win, so we could be at the beginning stages of an energy boom. The question is which stocks will benefit from this boom the most?

Ray Myers

84,029 views • 8 months ago

MY 5 BIGGEST TAKEAWAYS FROM $AMD ANALYST DAY 1. The pivot is official that AMD it’s building a platform with a goal is to sell full AI infrastructure: silicon, interconnect, software & racks that hyperscalers can deploy at scale. It’s chasing $NVDA vertical model but through an open ecosystem. 2. Data center is the company with Lisa guiding for ~35% CAGR, 57% gross margins & >$20 EPS by 2030. That requires DC biz to grow from ~$16B today to near $100B within five years which is a 6x run that only works if MI450 + Helios hit ramp targets & hyperscaler orders stack fast. 3. GPU is the upside but ROCm is the gatekeeper. The MI450 & yearly GPU cadence will define whether AMD becomes a real platform or stays a second source. Success hinges on ROCm feeling as stable and integrated as CUDA but partnerships with $META, $ORCL & OpenAI are the first real signs of ecosystem lift. 4. CPUs fund the fight with the server share near 40% gives AMD the profit engine it needs. As AI inference workloads push compute back toward CPUs, crossing 50% share does actually look realistic & that stable cash flow supports GPU & software expansion without breaking margins. 5. Michael Burry is wrong.
1:01

MY 5 BIGGEST TAKEAWAYS FROM $AMD ANALYST DAY 1. The pivot is official that AMD it’s building a platform with a goal is to sell full AI infrastructure: silicon, interconnect, software & racks that hyperscalers can deploy at scale. It’s chasing $NVDA vertical model but through an open ecosystem. 2. Data center is the company with Lisa guiding for ~35% CAGR, 57% gross margins & >$20 EPS by 2030. That requires DC biz to grow from ~$16B today to near $100B within five years which is a 6x run that only works if MI450 + Helios hit ramp targets & hyperscaler orders stack fast. 3. GPU is the upside but ROCm is the gatekeeper. The MI450 & yearly GPU cadence will define whether AMD becomes a real platform or stays a second source. Success hinges on ROCm feeling as stable and integrated as CUDA but partnerships with $META, $ORCL & OpenAI are the first real signs of ecosystem lift. 4. CPUs fund the fight with the server share near 40% gives AMD the profit engine it needs. As AI inference workloads push compute back toward CPUs, crossing 50% share does actually look realistic & that stable cash flow supports GPU & software expansion without breaking margins. 5. Michael Burry is wrong.

Shay Boloor

244,819 views • 9 months ago

Peter Thiel says the quiet part loud: AI chips will get commoditized. This is why I avoid $NVDA. It made an immense amount of money marking up its GPUs by 10x and others had to pay as alternatives weren’t good enough. Now they are getting good enough. $AMD has already caught up with $NVDA in hardware performance, and ASICS are proving unexpectedly competitive due to their efficiency. $GOOG trained Gemini 3 solely on TPUs, and Anthropic is running a significant part of its training and inference workloads on Trainium clusters. As alternatives become even stronger, $NVDA margins and volumes will erode, and profits will shift from hardware to the application layer.
1:42

Peter Thiel says the quiet part loud: AI chips will get commoditized. This is why I avoid $NVDA. It made an immense amount of money marking up its GPUs by 10x and others had to pay as alternatives weren’t good enough. Now they are getting good enough. $AMD has already caught up with $NVDA in hardware performance, and ASICS are proving unexpectedly competitive due to their efficiency. $GOOG trained Gemini 3 solely on TPUs, and Anthropic is running a significant part of its training and inference workloads on Trainium clusters. As alternatives become even stronger, $NVDA margins and volumes will erode, and profits will shift from hardware to the application layer.

Oguz Erkan

460,614 views • 8 months ago

$PLTR $AMD | Dr. Karp and Dr. Su were right! ✍️ Companies are now fighting back. Dr. Karp, Palantir CEO, recently told CNBC that enterprises are privately "unhappy" with frontier AI labs like OpenAI and Anthropic, accusing them of prioritizing "tokenmaxxing" or maximizing AI token consumption to signal activity over delivering real business value and understanding customer needs. Uber, Coinbases routing to capping token usage or routing to cheaper models to keep cost under control. or Microsoft revoked Claude Code licenses companywide, Priceline imposed token limits after sharp cost spikes, and reports cite Meta, Salesforce, and multiple unnamed firms facing 3x+ budget overruns or $ hundreds of millions in unexpected spend by mid-2026. Analysts note this as an emerging industry pattern, with FinOps and executives describing "existential crises" over token bills; dozens of enterprises are now adding guardrails, though public complaints remain concentrated among high-profile tech firms experimenting at scale. Dr. Lisa Su anticipated the pivot to inference economics and CPU-dense systems for agentic AI, correctly predicting that token costs, power efficiency, and deployability on standard platforms would determine scalable adoption long before the current enterprise pushback. Dr. Alex Karp accurately diagnosed the disconnect in frontier labs' approach, calling out "tokenmaxxing" as activity without outcomes; enterprises are indeed demanding real implementation and business-specific value rather than raw volume that inflates bills without proportional ROI. Together, their independent foresight validates the maturing AI thesis, efficient infrastructure (AMD Helios/EPYC optimized for lowest TCO & $/M Tokens) paired with outcome-focused platforms (Palantir AIP/Foundry) positions both companies to benefit as the market shifts from hype-driven consumption to sustainable, value-driven deployment. Yes it may look good on the revenue growth for AI Labs to show off on IPOs investors/bankers, but the customers have to find value in those tokens spent where $NVDA & In-house chips on inference claims are just false. At the end of the day, ~Token cost needs to go down more & more particularly inference by owning more AMD chips/racks. In-house chips can make all kind of claims for years, but the bills enterprises paid have to obey economic. ~Enterprises want a thick software OS or solution focused, they do not want to have unlimited budget for "tokenmaxxing" where it is leading to high costs with limited business transformation; success increasingly depends on implementation layers that route tasks, enforce policies, and connect AI to existing workflows. Not Financial Advice! DYOR!
2:07

$PLTR $AMD | Dr. Karp and Dr. Su were right! ✍️ Companies are now fighting back. Dr. Karp, Palantir CEO, recently told CNBC that enterprises are privately "unhappy" with frontier AI labs like OpenAI and Anthropic, accusing them of prioritizing "tokenmaxxing" or maximizing AI token consumption to signal activity over delivering real business value and understanding customer needs. Uber, Coinbases routing to capping token usage or routing to cheaper models to keep cost under control. or Microsoft revoked Claude Code licenses companywide, Priceline imposed token limits after sharp cost spikes, and reports cite Meta, Salesforce, and multiple unnamed firms facing 3x+ budget overruns or $ hundreds of millions in unexpected spend by mid-2026. Analysts note this as an emerging industry pattern, with FinOps and executives describing "existential crises" over token bills; dozens of enterprises are now adding guardrails, though public complaints remain concentrated among high-profile tech firms experimenting at scale. Dr. Lisa Su anticipated the pivot to inference economics and CPU-dense systems for agentic AI, correctly predicting that token costs, power efficiency, and deployability on standard platforms would determine scalable adoption long before the current enterprise pushback. Dr. Alex Karp accurately diagnosed the disconnect in frontier labs' approach, calling out "tokenmaxxing" as activity without outcomes; enterprises are indeed demanding real implementation and business-specific value rather than raw volume that inflates bills without proportional ROI. Together, their independent foresight validates the maturing AI thesis, efficient infrastructure (AMD Helios/EPYC optimized for lowest TCO & $/M Tokens) paired with outcome-focused platforms (Palantir AIP/Foundry) positions both companies to benefit as the market shifts from hype-driven consumption to sustainable, value-driven deployment. Yes it may look good on the revenue growth for AI Labs to show off on IPOs investors/bankers, but the customers have to find value in those tokens spent where $NVDA & In-house chips on inference claims are just false. At the end of the day, ~Token cost needs to go down more & more particularly inference by owning more AMD chips/racks. In-house chips can make all kind of claims for years, but the bills enterprises paid have to obey economic. ~Enterprises want a thick software OS or solution focused, they do not want to have unlimited budget for "tokenmaxxing" where it is leading to high costs with limited business transformation; success increasingly depends on implementation layers that route tasks, enforce policies, and connect AI to existing workflows. Not Financial Advice! DYOR!

Mike

253,557 views • 2 months ago

Dylan Patel on the importance of memory and storage Two key quotes: "An $NVDA GPU is faster than an $AMD GPU in most cases, but because AMD GPUs have more memory, they can outperform Nvidia in certain workloads." “It is a difficult, multivariable problem. Generally, you need the best GPU, such as a GB300, but you also need the best storage solutions. I will not spoil who comes out on top, but storage solutions matter a lot, memory solutions matter a lot, and frontend networking also matters significantly" Full Quote: “We have over $80 million of compute: GPUs from $NVDA and $AMD, TPUs from Google, and Trainium from Amazon. We constantly run this benchmark using the newest inference engines, drivers, PyTorch versions, and other software. It runs every day through automated CI across the latest Chinese models from GLM, Zhipu, Moonshot, Kimi, Alibaba, and others. Initially, when we were benchmarking the differences between these chips, inference engines, and parallelism schemes, we used fixed context lengths. But with Agent X, we have now analyzed more than $5 million worth of Claude Code traces. This is real production traffic that users have donated to us, combined with internally generated data, so we now understand what an actual agent workload looks like. When we implement those workloads and run the benchmarks, it turns out that the chip you are using is very important, but how you handle memory offload can be even more important. An Nvidia GPU is faster than an AMD GPU in most cases, but because AMD GPUs have more memory, they can outperform Nvidia in certain workloads. Similarly, you can use a less powerful GPU with a much better storage solution and outperform the best GPU when it lacks those solutions. Simply buying the newest GPU does not necessarily give you the best inference economics. You need to layer in other innovations, including storage and memory.” Interviewer: “Who is the top player on your chart? Can you tell us?” Dylan Patel: “It is a difficult, multivariable problem. Generally, you need the best GPU, such as a GB300, but you also need the best storage solutions. I will not spoil who comes out on top, but storage solutions matter a lot, memory solutions matter a lot, and frontend networking also matters significantly.”
2:27

Dylan Patel on the importance of memory and storage Two key quotes: "An $NVDA GPU is faster than an $AMD GPU in most cases, but because AMD GPUs have more memory, they can outperform Nvidia in certain workloads." “It is a difficult, multivariable problem. Generally, you need the best GPU, such as a GB300, but you also need the best storage solutions. I will not spoil who comes out on top, but storage solutions matter a lot, memory solutions matter a lot, and frontend networking also matters significantly" Full Quote: “We have over $80 million of compute: GPUs from $NVDA and $AMD, TPUs from Google, and Trainium from Amazon. We constantly run this benchmark using the newest inference engines, drivers, PyTorch versions, and other software. It runs every day through automated CI across the latest Chinese models from GLM, Zhipu, Moonshot, Kimi, Alibaba, and others. Initially, when we were benchmarking the differences between these chips, inference engines, and parallelism schemes, we used fixed context lengths. But with Agent X, we have now analyzed more than $5 million worth of Claude Code traces. This is real production traffic that users have donated to us, combined with internally generated data, so we now understand what an actual agent workload looks like. When we implement those workloads and run the benchmarks, it turns out that the chip you are using is very important, but how you handle memory offload can be even more important. An Nvidia GPU is faster than an AMD GPU in most cases, but because AMD GPUs have more memory, they can outperform Nvidia in certain workloads. Similarly, you can use a less powerful GPU with a much better storage solution and outperform the best GPU when it lacks those solutions. Simply buying the newest GPU does not necessarily give you the best inference economics. You need to layer in other innovations, including storage and memory.” Interviewer: “Who is the top player on your chart? Can you tell us?” Dylan Patel: “It is a difficult, multivariable problem. Generally, you need the best GPU, such as a GB300, but you also need the best storage solutions. I will not spoil who comes out on top, but storage solutions matter a lot, memory solutions matter a lot, and frontend networking also matters significantly.”

Daniel Romero

38,220 views • 1 month ago

$AMD $AMZN partnership will 🚀 in 2026 🔥 Amazon/AMD partnership is hidden among hot headlines from OpenAI $NVDA $ORCL... TLDR: Amazon refused to bid up the overpriced $NVDA chips among other hyperscalers, and decided to work closely with $AMD. Amazon is expected to spend up to $10-$20B a year on 2026 EPYC breakthrough Gen and Future Gen. Dr. Su confirmed "we have plenty for other large customers". For its 2026 EPYC "Venice" processors, AMD is using a multi-node manufacturing strategy: the CPU core complex dies (CCDs) are built on TSMC's 2 nm-class node (N2), while the I/O die (IOD) uses the N3P (3 nm) process. Context: Andy Jassy Amazon Web Services has been working with AMD on EPYC processors since November 2018. With this "secret weapon" breakthrough(patented), this long time partnership has expanded to New breakthrough 2026 EPYC Gen. AMD's 6th Gen EPYC "Venice" processors, slated for 2026, introduce New Chiplet design breakthrough. a revolutionary chiplet interconnect fabric that redefines server scalability for AI. This isn't just faster silicon; it's a paradigm shift for AWS, enabling hyper-efficient, rack-scale AI inference that slashes costs and latency while boosting throughput. AMD to benefit AWS's $100B+ AI opportunity along with $ORCL $MSFT $GOOGL $META Saudi, UAE ,38+ countries and startups. In early October, Amazon/AWS announced the new EC2 M8a instances as their latest-generation, general-purpose compute instances now powered by AMD EPYC 9005 "Turin" processors. Amazon announced the M8a as having up to 30% higher performance and up to 19% better price performance over M7a. With my testing of both at 32 vCPUs, the new AMD EPYC Turin instance provided 1.59x the performance over the prior-generation EPYC Genoa instance! How will this impact AWS AI Inference? ~Cost Efficiency: Inference is 80%+ of AI workloads and latency-sensitive (e.g., chatbots need <1s responses). "Secret weapon" enables 35x better inference perf (per AMD's CDNA roadmap tie-in), cutting AWS's energy use by 50%+ in clusters. With $118B 2025 capex, this could save $20–$30B annually in OPEX, boosting margins to 35%-40%. ~Scalability for Agentic AI: Supports "Helios" rack-scale platforms (up to 128 GPUs + EPYC hosts), delivering 3.58x FP6 perf for distributed inference. AWS can run 700K+ more tokens/sec in 1,000-node clusters (via EPYC 9575F boosts), enabling real-time apps like personalized search or fraud detection at enterprise scale. ~Adoption Catalysts: Early partners like Oracle signal broad uptake; AWS's existing AMD instances G4ad with Radeon GPUs) pave the way. By 2026, EPYC could power 40%+ of AWS AI infra, outpacing Nvidia's GPU lock-in via open standards (ROCm 8 software). Lastly, Amazon’s trajectory toward a $320 stock price is not a speculative leap but a grounded projection rooted in its unmatched fundamentals and strategic AI leadership. With Amazon Web Services poised to surpass $100 billion in annual revenue by 2026, driven by explosive AI inference demand, Amazon is redefining cloud computing’s future. The adoption of AMD’s 2026 EPYC processors with "Secret" architecture is a game-changer, slashing costs by up to 50% and boosting inference throughput 3x, enabling AWS to dominate enterprise AI workloads with unmatched efficiency. This technological edge, combined with Amazon’s e-commerce dominance and high-margin advertising growth, supports a valuation rerating to 22x EV/EBITDA, and it is still a discount to historical highs. Trading at $222, $AMZN is undervalued for its 15–20% revenue CAGR and 25%+ EPS growth through 2030.
0:49

$AMD $AMZN partnership will 🚀 in 2026 🔥 Amazon/AMD partnership is hidden among hot headlines from OpenAI $NVDA $ORCL... TLDR: Amazon refused to bid up the overpriced $NVDA chips among other hyperscalers, and decided to work closely with $AMD. Amazon is expected to spend up to $10-$20B a year on 2026 EPYC breakthrough Gen and Future Gen. Dr. Su confirmed "we have plenty for other large customers". For its 2026 EPYC "Venice" processors, AMD is using a multi-node manufacturing strategy: the CPU core complex dies (CCDs) are built on TSMC's 2 nm-class node (N2), while the I/O die (IOD) uses the N3P (3 nm) process. Context: Andy Jassy Amazon Web Services has been working with AMD on EPYC processors since November 2018. With this "secret weapon" breakthrough(patented), this long time partnership has expanded to New breakthrough 2026 EPYC Gen. AMD's 6th Gen EPYC "Venice" processors, slated for 2026, introduce New Chiplet design breakthrough. a revolutionary chiplet interconnect fabric that redefines server scalability for AI. This isn't just faster silicon; it's a paradigm shift for AWS, enabling hyper-efficient, rack-scale AI inference that slashes costs and latency while boosting throughput. AMD to benefit AWS's $100B+ AI opportunity along with $ORCL $MSFT $GOOGL $META Saudi, UAE ,38+ countries and startups. In early October, Amazon/AWS announced the new EC2 M8a instances as their latest-generation, general-purpose compute instances now powered by AMD EPYC 9005 "Turin" processors. Amazon announced the M8a as having up to 30% higher performance and up to 19% better price performance over M7a. With my testing of both at 32 vCPUs, the new AMD EPYC Turin instance provided 1.59x the performance over the prior-generation EPYC Genoa instance! How will this impact AWS AI Inference? ~Cost Efficiency: Inference is 80%+ of AI workloads and latency-sensitive (e.g., chatbots need <1s responses). "Secret weapon" enables 35x better inference perf (per AMD's CDNA roadmap tie-in), cutting AWS's energy use by 50%+ in clusters. With $118B 2025 capex, this could save $20–$30B annually in OPEX, boosting margins to 35%-40%. ~Scalability for Agentic AI: Supports "Helios" rack-scale platforms (up to 128 GPUs + EPYC hosts), delivering 3.58x FP6 perf for distributed inference. AWS can run 700K+ more tokens/sec in 1,000-node clusters (via EPYC 9575F boosts), enabling real-time apps like personalized search or fraud detection at enterprise scale. ~Adoption Catalysts: Early partners like Oracle signal broad uptake; AWS's existing AMD instances G4ad with Radeon GPUs) pave the way. By 2026, EPYC could power 40%+ of AWS AI infra, outpacing Nvidia's GPU lock-in via open standards (ROCm 8 software). Lastly, Amazon’s trajectory toward a $320 stock price is not a speculative leap but a grounded projection rooted in its unmatched fundamentals and strategic AI leadership. With Amazon Web Services poised to surpass $100 billion in annual revenue by 2026, driven by explosive AI inference demand, Amazon is redefining cloud computing’s future. The adoption of AMD’s 2026 EPYC processors with "Secret" architecture is a game-changer, slashing costs by up to 50% and boosting inference throughput 3x, enabling AWS to dominate enterprise AI workloads with unmatched efficiency. This technological edge, combined with Amazon’s e-commerce dominance and high-margin advertising growth, supports a valuation rerating to 22x EV/EBITDA, and it is still a discount to historical highs. Trading at $222, $AMZN is undervalued for its 15–20% revenue CAGR and 25%+ EPS growth through 2030.

Mike

511,082 views • 10 months ago

J-Cal: Nvidia is taking the gloves off with its open source model. They will own the whole stack. “You will not be able to tell the difference between Jensen Huang's open source LLM and Claude for 95% of your searches, I guarantee you. Now, why has Nvidia and Jensen downplayed their open source model until this moment? Why would he do that? Why would he never bring it up in the All-In interview, never bring it up? Because his top customers (OpenAI, Anthropic, etc.) were very concerned, from what I understand, about the fact that Nvidia had made so much progress on their open source model. But suddenly, after OpenAI announced their Jalapeno chips, after Anthropic started making chips, after AMD did successful projects with both of these companies, after Elon said he's going to do his own fab, Nvidia's taking the gloves off. They are going to own the whole stack. You get the hardware from them, and you're going to get a model that's competitive with OpenAI's for free.”
0:54

J-Cal: Nvidia is taking the gloves off with its open source model. They will own the whole stack. “You will not be able to tell the difference between Jensen Huang's open source LLM and Claude for 95% of your searches, I guarantee you. Now, why has Nvidia and Jensen downplayed their open source model until this moment? Why would he do that? Why would he never bring it up in the All-In interview, never bring it up? Because his top customers (OpenAI, Anthropic, etc.) were very concerned, from what I understand, about the fact that Nvidia had made so much progress on their open source model. But suddenly, after OpenAI announced their Jalapeno chips, after Anthropic started making chips, after AMD did successful projects with both of these companies, after Elon said he's going to do his own fab, Nvidia's taking the gloves off. They are going to own the whole stack. You get the hardware from them, and you're going to get a model that's competitive with OpenAI's for free.”

The All-In Podcast

227,477 views • 1 month ago

$AMD| $META is using $GOOGL to negotiate 🧵 The Ironwood pod is 5.1–10x more expensive annually ($148.3 million ÷ $14.87–$29.04 million) and 5.1–10x more expensive monthly ($12.36 million ÷ $1.24–$2.42 million) than renting 15 MI450 racks for equivalent compute. The rapidly evolving landscape of artificial intelligence infrastructure presents a complex interplay of technological innovation, market dynamics, and strategic maneuvering among major players. Recent leaked information suggesting that Meta Platforms ($META) might work with Google's Tensor Processing Unit (TPU) in 2027 has sparked speculation about its true intent. This leak is likely a strategic move by Meta to negotiate more favorable terms with AMD , leveraging the competitive dynamics of the AI hardware market to optimize its substantial investment in AI infrastructure. By examining the key elements of this scenario Meta's investment strategy, the comparative advantages of AMD's MI450 and Google's Ironwood TPU, and the broader market context; we can discern the potential beneficiaries and the strategic implications of this information. Meta's aggressive pursuit of AI capabilities is underscored by its planned expenditure of $66-72 billion on AI infrastructure in 2025, with expectations to escalate significantly in 2026. This investment is part of a broader strategy to build "titan clusters" like Prometheus, which are projected to reach 1 gigawatt of compute power by 2026. Such a scale of investment reflects Meta's recognition of the critical role that AI will play in its future growth, particularly in enhancing its social media platforms and developing new AI-driven applications. However, the financial burden of this infrastructure buildout necessitates a careful consideration of cost-effectiveness and scalability, which brings us to the leaked information about potential collaboration with Google's Ironwood TPU. Google's Ironwood TPU, introduced as the seventh-generation ASIC optimized for TensorFlow-based inference, represents a high-cost, cloud-locked solution priced at $445 million per pod (9,216 chips) over three years. This model, while offering significant performance gains and power efficiency, is tailored for pod-scale deployment and integrated with Google's cloud services, limiting flexibility and increasing costs for customers. In contrast, AMD's MI450 GPU, priced at $30,000–$40,000 per unit, provides a modular, open ROCm ecosystem that delivers comparable compute capacity at a fraction of the cost. Renting 15 MI450 racks could achieve similar 42+ exaFLOPS inference compute at 5–10x lower cost than renting a single Ironwood pod, underscoring AMD's competitive edge in terms of total cost of ownership (TCO). The leaked information about Meta's potential TPU deployment in 2027, therefore, can be interpreted as a negotiating tactic rather than a definitive shift in strategy. By signaling interest in Google's solution, Meta may be attempting to pressure AMD into offering more favorable terms/prices for 5-10GW. This tactic aligns with Meta's broader goal to finance most of its AI spend internally while exploring partnerships that can reduce costs and enhance flexibility. The post's emphasis on MI450's TCO advantage and its partnerships with major players like OpenAI, Microsoft, and Meta itself suggests that AMD is a critical component of Meta's AI infrastructure strategy. The threat of working with Google's TPU could prompt AMD to reassess its pricing, provide additional support, or offer incentives to retain Meta as a customer, thereby securing or expanding its market share. From a logical standpoint, Meta stands to benefit the most from this strategy. As a major buyer in a high-stakes market projected to surpass $1 trillion in annual spending by 2030, Meta's negotiating power is significant. The leaked information could lead to substantial cost savings on its $66-72 billion investment, enhancing its financial flexibility and allowing for further investment in AI capabilities. Moreover, this tactic reinforces Meta's position as a leader in the AI infrastructure race, potentially attracting more external financing for its data center projects and strengthening its competitive stance against other hyperscalers like Amazon and Microsoft. AMD could also benefit from this scenario. The negotiation pressure might lead to small short-term concessions, but it could also solidify long-term partnerships with Meta, ensuring continued demand for MI450 and other AI hardware solutions. Initially Meta's 42% allocation to AMD MI300X and its partnerships with Oracle, Dell, and HP indicates a deep integration of AMD's technology into Meta's infrastructure, which could be leveraged to maintain this relationship. For AMD, retaining Meta as a large key customer is crucial to capturing a larger share of the rapidly growing data center infrastructure market, driven by the insatiable demand for AI compute power. Google, on the other hand, faces a more limited benefit from this leaked information. While securing Meta as a customer would reinforce its position in the AI hardware market, the high cost and ecosystem lock-in of the Ironwood TPU might deter Meta from fully committing to this solution. The leaked information could prompt Google to reconsider its pricing or ecosystem strategy to remain competitive, but the immediate impact is likely to be minimal compared to the potential gains for Meta and AMD. Investors and market analysts also stand to benefit from this information, as it provides insights into the competitive dynamics of the AI hardware market. Adjustments in portfolios based on anticipated shifts in market share and profitability could lead to opportunities for those who correctly anticipate outcomes. The negotiation dynamic might introduce volatility, but it also highlights the strategic importance of cost-effective solutions in the AI infrastructure space. Lastly, the leaked information about Meta potentially working with Google's TPU in 2027 is likely a strategic move to negotiate with AMD, leveraging the competitive landscape to optimize its AI infrastructure investment. Meta, as the primary negotiator, stands to gain the most by securing better terms from AMD, reducing costs, and enhancing its financial flexibility. AMD, while initially at risk, could benefit from retaining a key customer and solidifying its market position. Google faces limited immediate benefits but may need to adapt its strategy to remain competitive. This scenario underscores the complex interplay of technology, market dynamics, and strategic maneuvering in the AI hardware market, where cost-effectiveness and scalability are paramount. As the data center infrastructure market continues to grow, the outcomes of such negotiations will shape the future of AI development and deployment.
1:12

$AMD| $META is using $GOOGL to negotiate 🧵 The Ironwood pod is 5.1–10x more expensive annually ($148.3 million ÷ $14.87–$29.04 million) and 5.1–10x more expensive monthly ($12.36 million ÷ $1.24–$2.42 million) than renting 15 MI450 racks for equivalent compute. The rapidly evolving landscape of artificial intelligence infrastructure presents a complex interplay of technological innovation, market dynamics, and strategic maneuvering among major players. Recent leaked information suggesting that Meta Platforms ($META) might work with Google's Tensor Processing Unit (TPU) in 2027 has sparked speculation about its true intent. This leak is likely a strategic move by Meta to negotiate more favorable terms with AMD , leveraging the competitive dynamics of the AI hardware market to optimize its substantial investment in AI infrastructure. By examining the key elements of this scenario Meta's investment strategy, the comparative advantages of AMD's MI450 and Google's Ironwood TPU, and the broader market context; we can discern the potential beneficiaries and the strategic implications of this information. Meta's aggressive pursuit of AI capabilities is underscored by its planned expenditure of $66-72 billion on AI infrastructure in 2025, with expectations to escalate significantly in 2026. This investment is part of a broader strategy to build "titan clusters" like Prometheus, which are projected to reach 1 gigawatt of compute power by 2026. Such a scale of investment reflects Meta's recognition of the critical role that AI will play in its future growth, particularly in enhancing its social media platforms and developing new AI-driven applications. However, the financial burden of this infrastructure buildout necessitates a careful consideration of cost-effectiveness and scalability, which brings us to the leaked information about potential collaboration with Google's Ironwood TPU. Google's Ironwood TPU, introduced as the seventh-generation ASIC optimized for TensorFlow-based inference, represents a high-cost, cloud-locked solution priced at $445 million per pod (9,216 chips) over three years. This model, while offering significant performance gains and power efficiency, is tailored for pod-scale deployment and integrated with Google's cloud services, limiting flexibility and increasing costs for customers. In contrast, AMD's MI450 GPU, priced at $30,000–$40,000 per unit, provides a modular, open ROCm ecosystem that delivers comparable compute capacity at a fraction of the cost. Renting 15 MI450 racks could achieve similar 42+ exaFLOPS inference compute at 5–10x lower cost than renting a single Ironwood pod, underscoring AMD's competitive edge in terms of total cost of ownership (TCO). The leaked information about Meta's potential TPU deployment in 2027, therefore, can be interpreted as a negotiating tactic rather than a definitive shift in strategy. By signaling interest in Google's solution, Meta may be attempting to pressure AMD into offering more favorable terms/prices for 5-10GW. This tactic aligns with Meta's broader goal to finance most of its AI spend internally while exploring partnerships that can reduce costs and enhance flexibility. The post's emphasis on MI450's TCO advantage and its partnerships with major players like OpenAI, Microsoft, and Meta itself suggests that AMD is a critical component of Meta's AI infrastructure strategy. The threat of working with Google's TPU could prompt AMD to reassess its pricing, provide additional support, or offer incentives to retain Meta as a customer, thereby securing or expanding its market share. From a logical standpoint, Meta stands to benefit the most from this strategy. As a major buyer in a high-stakes market projected to surpass $1 trillion in annual spending by 2030, Meta's negotiating power is significant. The leaked information could lead to substantial cost savings on its $66-72 billion investment, enhancing its financial flexibility and allowing for further investment in AI capabilities. Moreover, this tactic reinforces Meta's position as a leader in the AI infrastructure race, potentially attracting more external financing for its data center projects and strengthening its competitive stance against other hyperscalers like Amazon and Microsoft. AMD could also benefit from this scenario. The negotiation pressure might lead to small short-term concessions, but it could also solidify long-term partnerships with Meta, ensuring continued demand for MI450 and other AI hardware solutions. Initially Meta's 42% allocation to AMD MI300X and its partnerships with Oracle, Dell, and HP indicates a deep integration of AMD's technology into Meta's infrastructure, which could be leveraged to maintain this relationship. For AMD, retaining Meta as a large key customer is crucial to capturing a larger share of the rapidly growing data center infrastructure market, driven by the insatiable demand for AI compute power. Google, on the other hand, faces a more limited benefit from this leaked information. While securing Meta as a customer would reinforce its position in the AI hardware market, the high cost and ecosystem lock-in of the Ironwood TPU might deter Meta from fully committing to this solution. The leaked information could prompt Google to reconsider its pricing or ecosystem strategy to remain competitive, but the immediate impact is likely to be minimal compared to the potential gains for Meta and AMD. Investors and market analysts also stand to benefit from this information, as it provides insights into the competitive dynamics of the AI hardware market. Adjustments in portfolios based on anticipated shifts in market share and profitability could lead to opportunities for those who correctly anticipate outcomes. The negotiation dynamic might introduce volatility, but it also highlights the strategic importance of cost-effective solutions in the AI infrastructure space. Lastly, the leaked information about Meta potentially working with Google's TPU in 2027 is likely a strategic move to negotiate with AMD, leveraging the competitive landscape to optimize its AI infrastructure investment. Meta, as the primary negotiator, stands to gain the most by securing better terms from AMD, reducing costs, and enhancing its financial flexibility. AMD, while initially at risk, could benefit from retaining a key customer and solidifying its market position. Google faces limited immediate benefits but may need to adapt its strategy to remain competitive. This scenario underscores the complex interplay of technology, market dynamics, and strategic maneuvering in the AI hardware market, where cost-effectiveness and scalability are paramount. As the data center infrastructure market continues to grow, the outcomes of such negotiations will shape the future of AI development and deployment.

Mike

182,273 views • 9 months ago

David Sacks just said what every honest analyst in Silicon Valley is already thinking (Save this). Nobody has ever seen anything like this. Anthropic has grown at 10x per year for three straight years and going into 2026, the conventional wisdom was that the rate of growth had to slow at this level of scale but then the numbers came in. Q1 alone is $10B ARR to $30B, in April, $30B to $44B and that's $96 million in new ARR added every single day. Inference margins are now above 70%, up from 38% last year and the only thing holding them back was compute. That's solved now, the SpaceX deal and others Anthropic has been quietly signing unlocks the supply side. This is exactly why we are bullish on Nebius and AMD. When a single company is adding nearly $100M in ARR per day, the real trade isn't the frontier lab but rather the infrastructure underneath it. Nebius, one of the fastest-growing neoclouds on the planet posted 547% YoY revenue growth in Q4 2025, exited the year with $1.25B ARR, and is guiding for $7–9B ARR by year-end 2026. Their revenue backlog has reached $46B, with projections of $16B in revenue by 2028 and NVIDIA locked in a $2 billion stock buy agreement with them giving Nebius early access to cutting-edge chips while every other cloud scrambles for supply. AMD is the other side of the same coin. Data center revenue hit $5.78B in Q1, up 57% year-over-year with total company revenue at $10.25B, up 38%. Meta has committed to deploying up to 6 gigawatts of AMD Instinct GPUs. Data center GPU revenue is forecast to surge 114% year over year to $15B in 2026. MI400-series chips hit the market in H2 and analysts project segment operating margins climbing to 31% as the next generation ramps. The model is simple, Anthropic is printing revenue and that that revenue pays for compute. That compute flows through companies like Nebius and AMD. This is why Milk Road PRO remains bullish on them and our positions are up massively. Our analysts have broken down the full thesis, the allocations, and the price targets. Go PRO at Milk Road to see everything, link below!
1:03

David Sacks just said what every honest analyst in Silicon Valley is already thinking (Save this). Nobody has ever seen anything like this. Anthropic has grown at 10x per year for three straight years and going into 2026, the conventional wisdom was that the rate of growth had to slow at this level of scale but then the numbers came in. Q1 alone is $10B ARR to $30B, in April, $30B to $44B and that's $96 million in new ARR added every single day. Inference margins are now above 70%, up from 38% last year and the only thing holding them back was compute. That's solved now, the SpaceX deal and others Anthropic has been quietly signing unlocks the supply side. This is exactly why we are bullish on Nebius and AMD. When a single company is adding nearly $100M in ARR per day, the real trade isn't the frontier lab but rather the infrastructure underneath it. Nebius, one of the fastest-growing neoclouds on the planet posted 547% YoY revenue growth in Q4 2025, exited the year with $1.25B ARR, and is guiding for $7–9B ARR by year-end 2026. Their revenue backlog has reached $46B, with projections of $16B in revenue by 2028 and NVIDIA locked in a $2 billion stock buy agreement with them giving Nebius early access to cutting-edge chips while every other cloud scrambles for supply. AMD is the other side of the same coin. Data center revenue hit $5.78B in Q1, up 57% year-over-year with total company revenue at $10.25B, up 38%. Meta has committed to deploying up to 6 gigawatts of AMD Instinct GPUs. Data center GPU revenue is forecast to surge 114% year over year to $15B in 2026. MI400-series chips hit the market in H2 and analysts project segment operating margins climbing to 31% as the next generation ramps. The model is simple, Anthropic is printing revenue and that that revenue pays for compute. That compute flows through companies like Nebius and AMD. This is why Milk Road PRO remains bullish on them and our positions are up massively. Our analysts have broken down the full thesis, the allocations, and the price targets. Go PRO at Milk Road to see everything, link below!

Milk Road AI

184,643 views • 3 months ago

Alex Krainer (Alex (Sasha) Krainer) explains why Saudi Arabia, after watching the West freeze nearly $300 billion of Russian assets, is now paying a protection racket to Turkey and Pakistan. The Gulf states understand their own trillions in Western-held sovereign wealth are next. The new alliance is not about defending Israel—it is about acquiring new protectors before the imperial system loots everything. Host Zulfikar, who has direct access to the relevant deliberations, confirms the assessment as simply “protection money.” Krainer: “Everybody saw how the Western powers didn’t hesitate a moment to freeze almost 300 billion dollars’ worth of Russian assets in the Western financial institutions. If they are brazen enough, if they are ruthless enough to do this to the number-one nuclear power in the world, then what is Saudi Arabia to them? It’s nothing. It’s a maggot they will squash whenever they choose to do so. Saudi Arabia not only has far more than Russia—last I looked they had something like 800 or 900 billion dollars in direct assets in the United States—but they also have these enormous sovereign wealth funds. The combined assets of the Saudi, UAE, Qatari, Bahraini and Kuwaiti sovereign wealth funds in the West are something close to five trillion dollars. One thing we can take as a near-100% certainty is that before the Western empire fails, it is going to loot all of that money and take it for its own ends. After Ukraine and after what they did to Russia, these rulers may be corrupt, they may be brutal, they may be ruthless and oppressive, but they are not stupid. They see the writing on the wall. They see what is coming their way. The way I interpret this new alliance between Pakistan, Turkey and Saudi Arabia is that Saudi Arabia is deciding to pay a protection racket to Turkey and Pakistan. Two of these three nations are major military powers; one of them is a pipsqueak. Why would Pakistan and Turkey sign an alliance with Saudi Arabia—an attack on one is an attack on all? If somebody attacks Turkey, is Saudi Arabia going to come to their aid? That’s a joke. I think the Saudis initiated this by saying: we are already screwed. We are going to get skinned alive. All our sovereign asset funds—we are probably never going to see them again. So what we can do is, since you are already at least on paper allied with the United States, you can become our protectors. We will pay you for it with these funds that are in the United States. If we say ‘send us a trillion dollars,’ Trump is going to say no—why don’t we send you a trillion dollars’ worth of Patriot missiles that we will deliver in 2046? But if Turkey or Pakistan say ‘send us a trillion dollars,’ the United States may have to send the money. Saudi Arabia gets new protectors, and these new protectors get some of the Saudi money.” Zulfikar: “Alex, a lot of nonsense has been said about this, but you have cut to the chase. You have uttered the words that are the truest words describing what has just taken place. I say this sitting in a position where I have actually listened to the deliberations between people who put their names on the piece of paper. I am that close to the scene. Exactly what you have said is the truth. It is very simple. Saudi Arabia is buying protection. It is protection money. Easy, very simple to understand. And Saudi Arabia watched with great dismay the destruction of the United States by Iran—under 50 years of sanctions.” The Gulf states have read the writing on the wall. Paying a protection racket to real military powers is their attempt to escape the coming looting of the imperial system.
5:19

Alex Krainer (Alex (Sasha) Krainer) explains why Saudi Arabia, after watching the West freeze nearly $300 billion of Russian assets, is now paying a protection racket to Turkey and Pakistan. The Gulf states understand their own trillions in Western-held sovereign wealth are next. The new alliance is not about defending Israel—it is about acquiring new protectors before the imperial system loots everything. Host Zulfikar, who has direct access to the relevant deliberations, confirms the assessment as simply “protection money.” Krainer: “Everybody saw how the Western powers didn’t hesitate a moment to freeze almost 300 billion dollars’ worth of Russian assets in the Western financial institutions. If they are brazen enough, if they are ruthless enough to do this to the number-one nuclear power in the world, then what is Saudi Arabia to them? It’s nothing. It’s a maggot they will squash whenever they choose to do so. Saudi Arabia not only has far more than Russia—last I looked they had something like 800 or 900 billion dollars in direct assets in the United States—but they also have these enormous sovereign wealth funds. The combined assets of the Saudi, UAE, Qatari, Bahraini and Kuwaiti sovereign wealth funds in the West are something close to five trillion dollars. One thing we can take as a near-100% certainty is that before the Western empire fails, it is going to loot all of that money and take it for its own ends. After Ukraine and after what they did to Russia, these rulers may be corrupt, they may be brutal, they may be ruthless and oppressive, but they are not stupid. They see the writing on the wall. They see what is coming their way. The way I interpret this new alliance between Pakistan, Turkey and Saudi Arabia is that Saudi Arabia is deciding to pay a protection racket to Turkey and Pakistan. Two of these three nations are major military powers; one of them is a pipsqueak. Why would Pakistan and Turkey sign an alliance with Saudi Arabia—an attack on one is an attack on all? If somebody attacks Turkey, is Saudi Arabia going to come to their aid? That’s a joke. I think the Saudis initiated this by saying: we are already screwed. We are going to get skinned alive. All our sovereign asset funds—we are probably never going to see them again. So what we can do is, since you are already at least on paper allied with the United States, you can become our protectors. We will pay you for it with these funds that are in the United States. If we say ‘send us a trillion dollars,’ Trump is going to say no—why don’t we send you a trillion dollars’ worth of Patriot missiles that we will deliver in 2046? But if Turkey or Pakistan say ‘send us a trillion dollars,’ the United States may have to send the money. Saudi Arabia gets new protectors, and these new protectors get some of the Saudi money.” Zulfikar: “Alex, a lot of nonsense has been said about this, but you have cut to the chase. You have uttered the words that are the truest words describing what has just taken place. I say this sitting in a position where I have actually listened to the deliberations between people who put their names on the piece of paper. I am that close to the scene. Exactly what you have said is the truth. It is very simple. Saudi Arabia is buying protection. It is protection money. Easy, very simple to understand. And Saudi Arabia watched with great dismay the destruction of the United States by Iran—under 50 years of sanctions.” The Gulf states have read the writing on the wall. Paying a protection racket to real military powers is their attempt to escape the coming looting of the imperial system.

🅰pocalypsis 🅰pocalypseos 🇷🇺 🇨🇳 🅉

60,453 views • 13 days ago

Today, we’re announcing the first major discovery made by our AI Scientist with the lab in the loop: a promising new treatment for dry AMD, a major cause of blindness. Our agents generated the hypotheses, designed the experiments, analyzed the data, iterated, even made figures for the paper. The resulting manuscript is a first-of-a-kind in the natural sciences, in which everything that needed to be done to write the paper was done by AI agents, apart from actually conducting the physical experiments in the lab and writing the final manuscript. We are also introducing Robin, the first multi-agent system that fully automates the in-silico components of scientific discovery, which made this discovery. This is the first time that we are aware of that hypothesis generation, experimentation, and data analysis have been joined up in closed loop, and is the beginning of a massive acceleration in the pace of scientific discovery that will be driven by these agents. We will be open-sourcing the code and data next week. Robin is a multi-agent system that uses Crow, Falcon, and Finch, the agents on our platform, to generate novel hypotheses, plan experiments, and analyze data. We asked Robin to find a new treatment for dry age-related macular degeneration. Robin considered the disease mechanisms associated with dry AMD, proposed a specific experimental assay that could be used to evaluate hypotheses in the wet lab, and proposed specific molecules we could test in that assay. We tested the molecules and gave it the resulting data, which it analyzed before proposing more experiments. In the end, it identified Ripasudil, a Rho Kinase inhibitor (ROCK inhibitor) that is approved in Japan for several other diseases, which seems very promising as potential treatment for dry AMD. It also identified specific molecular mechanisms that might underlie the effects of Ripasudil in RPE cells, from an RNA sequencing experiment it proposed. To be clear, no one has proposed using ROCK inhibitors to treat dry AMD in the literature before, as far as we can find, and I think it would have been very difficult for us to come up with this hypothesis without the agents. We have also run the proposed treatment by several experts in AMD, who confirm that it is interesting and novel. Moreover, this project was fast: with Robin in hand, the entire project took about 10 weeks, which is way shorter than it would have taken if we had been doing all of the in-silico components ourselves. Important caveats: We are real biologists at FutureHouse, so I want to be clear that although the discovery here is exciting, we are not claiming that we have cured dry AMD. Fully validating this hypothesis as a treatment for dry AMD will take human trials, which will take much longer. Also, this discovery is cool, but it is not yet a "move 37"-style discovery. At the current rate of progress, I'm sure we will get to that level soon. Congratulations to the team. Congratulations in particular to Robin, which generated the hypotheses, proposed the experiments, analyzed the data and generated the figures. And major congratulations also to the human team, which built Robin: Michaela Hinks, Ali Ghareeb, Benjamin Chang, Ludovico Mitchener, Mo Razzak, Kiki Szostkiewicz, and Angela Yiu.
5:57

Today, we’re announcing the first major discovery made by our AI Scientist with the lab in the loop: a promising new treatment for dry AMD, a major cause of blindness. Our agents generated the hypotheses, designed the experiments, analyzed the data, iterated, even made figures for the paper. The resulting manuscript is a first-of-a-kind in the natural sciences, in which everything that needed to be done to write the paper was done by AI agents, apart from actually conducting the physical experiments in the lab and writing the final manuscript. We are also introducing Robin, the first multi-agent system that fully automates the in-silico components of scientific discovery, which made this discovery. This is the first time that we are aware of that hypothesis generation, experimentation, and data analysis have been joined up in closed loop, and is the beginning of a massive acceleration in the pace of scientific discovery that will be driven by these agents. We will be open-sourcing the code and data next week. Robin is a multi-agent system that uses Crow, Falcon, and Finch, the agents on our platform, to generate novel hypotheses, plan experiments, and analyze data. We asked Robin to find a new treatment for dry age-related macular degeneration. Robin considered the disease mechanisms associated with dry AMD, proposed a specific experimental assay that could be used to evaluate hypotheses in the wet lab, and proposed specific molecules we could test in that assay. We tested the molecules and gave it the resulting data, which it analyzed before proposing more experiments. In the end, it identified Ripasudil, a Rho Kinase inhibitor (ROCK inhibitor) that is approved in Japan for several other diseases, which seems very promising as potential treatment for dry AMD. It also identified specific molecular mechanisms that might underlie the effects of Ripasudil in RPE cells, from an RNA sequencing experiment it proposed. To be clear, no one has proposed using ROCK inhibitors to treat dry AMD in the literature before, as far as we can find, and I think it would have been very difficult for us to come up with this hypothesis without the agents. We have also run the proposed treatment by several experts in AMD, who confirm that it is interesting and novel. Moreover, this project was fast: with Robin in hand, the entire project took about 10 weeks, which is way shorter than it would have taken if we had been doing all of the in-silico components ourselves. Important caveats: We are real biologists at FutureHouse, so I want to be clear that although the discovery here is exciting, we are not claiming that we have cured dry AMD. Fully validating this hypothesis as a treatment for dry AMD will take human trials, which will take much longer. Also, this discovery is cool, but it is not yet a "move 37"-style discovery. At the current rate of progress, I'm sure we will get to that level soon. Congratulations to the team. Congratulations in particular to Robin, which generated the hypotheses, proposed the experiments, analyzed the data and generated the figures. And major congratulations also to the human team, which built Robin: Michaela Hinks, Ali Ghareeb, Benjamin Chang, Ludovico Mitchener, Mo Razzak, Kiki Szostkiewicz, and Angela Yiu.

Sam Rodriques

1,107,754 views • 1 year ago