Загрузка видео...

Не удалось загрузить видео

На главную

Gavin Baker (Gavin Baker) says the disaggregation of inference can extend GPU useful lives from 3-4 years to 10-15. That may single-handedly save private credit and reduce the financing rates for GPUs, which will drive demand and help finance the build-out. "The disaggregation of prefill and inference is going...

InvestLikeBest's profile picture

Invest Like the Best

36,939 subscribers

211,336 просмотров • 3 месяцев назад •via X (Twitter)

Наука и технологииФинансы
LIVE

Anya Rossi• Live Now

Private livecam show

Комментарии: 0

Нет доступных комментариев

Здесь появятся комментарии из оригинального поста

Похожие видео

Spotted a chip depreciation narrative violation in the wild on the show today CoreWeave founder brannin mcbee: "The A100 SKU. This is a 2020 SKU. We still have clients coming in, asking specifically for that SKU." "It's not like they're coming in asking for Hopper, and it's like, 'Oh, we only have Ampere,' or they're asking for Blackwell, and it's like, 'Oh, we only have Ampere.'" "They are saying, 'No, we want Ampere for workloads,' because that is the most performant platform for their workload." "We signed a contract in the quarter that goes out through 2029. It'll be a 9-year-old chip by the end. And this is a take-or-pay, fixed price contract. And pricing, on A100s, has held solid since early 2025." "One of the big criticisms was, 'Well, it's 2- to 3-year compute, and it's useless after that.' But we've been very consistent that a 6-year depreciable life is accurate for this infrastructure." "And I think that there is a really strong opportunity for it to have material useful life beyond that. And we see it every day." "There isn't one GPU to rule them all. It's just this matrix of different sizes of workloads relative to different sizes of GPUs."
2:19

Spotted a chip depreciation narrative violation in the wild on the show today CoreWeave founder brannin mcbee: "The A100 SKU. This is a 2020 SKU. We still have clients coming in, asking specifically for that SKU." "It's not like they're coming in asking for Hopper, and it's like, 'Oh, we only have Ampere,' or they're asking for Blackwell, and it's like, 'Oh, we only have Ampere.'" "They are saying, 'No, we want Ampere for workloads,' because that is the most performant platform for their workload." "We signed a contract in the quarter that goes out through 2029. It'll be a 9-year-old chip by the end. And this is a take-or-pay, fixed price contract. And pricing, on A100s, has held solid since early 2025." "One of the big criticisms was, 'Well, it's 2- to 3-year compute, and it's useless after that.' But we've been very consistent that a 6-year depreciable life is accurate for this infrastructure." "And I think that there is a really strong opportunity for it to have material useful life beyond that. And we see it every day." "There isn't one GPU to rule them all. It's just this matrix of different sizes of workloads relative to different sizes of GPUs."

TBPN

44,015 просмотров • 8 дней назад

Jensen - "We know that every single GPU that Nvidia has in the cloud, across aws and gcp, is rented out. They’re all fully utilized. Demand for gpus is so high that even the spot prices of gpus sold a couple of years ago are going up."
0:52

Jensen - "We know that every single GPU that Nvidia has in the cloud, across aws and gcp, is rented out. They’re all fully utilized. Demand for gpus is so high that even the spot prices of gpus sold a couple of years ago are going up."

The AI Investor

69,299 просмотров • 6 месяцев назад

CoreWeave $CRWV CEO Michael Intrator, on the Big Technology Podcast, providing what I believe is the most compelling rebuttal to the narrative that GPUs only have a 2 to 3-year useful life. CoreWeave's customers are the world's most sophisticated users of GPUs, and they are signing contracts to utilize today's GPUs for the next 5 years, despite Nvidia's new product release cadence resulting in more efficient GPUs every 1-2 years. The use-case for the GPUs might change over time as Nvidia rolls out new products (e.g. older chips are moved from training to inference workloads), but the "usefulness" does not. "The most important tool that I have for understanding what the depreciation curve or the obsolescence curve of compute is not what I think, right? It's not what some historic short (seller) thinks. It's what are the buyers, the most sophisticated companies in the world are willing to pay for today. And when they come to me and they put in a contract for a 5-year deal or 6-year deal, in what world do I not think that they who are the consumers of this understand that there are new, more powerful chips coming out? Of course they do. They understand it, but they also understand what their various use cases are. And they're saying to themselves, "I'm going to buy this because I'm going to need it today. I'm going to need it in 3 years, and I'm going to need it in 5 years"... My opinions around depreciation are informed by the only entities that get to vote in my world, which are the folks that are paying for the compute over time."
1:33

CoreWeave $CRWV CEO Michael Intrator, on the Big Technology Podcast, providing what I believe is the most compelling rebuttal to the narrative that GPUs only have a 2 to 3-year useful life. CoreWeave's customers are the world's most sophisticated users of GPUs, and they are signing contracts to utilize today's GPUs for the next 5 years, despite Nvidia's new product release cadence resulting in more efficient GPUs every 1-2 years. The use-case for the GPUs might change over time as Nvidia rolls out new products (e.g. older chips are moved from training to inference workloads), but the "usefulness" does not. "The most important tool that I have for understanding what the depreciation curve or the obsolescence curve of compute is not what I think, right? It's not what some historic short (seller) thinks. It's what are the buyers, the most sophisticated companies in the world are willing to pay for today. And when they come to me and they put in a contract for a 5-year deal or 6-year deal, in what world do I not think that they who are the consumers of this understand that there are new, more powerful chips coming out? Of course they do. They understand it, but they also understand what their various use cases are. And they're saying to themselves, "I'm going to buy this because I'm going to need it today. I'm going to need it in 3 years, and I'm going to need it in 5 years"... My opinions around depreciation are informed by the only entities that get to vote in my world, which are the folks that are paying for the compute over time."

Rittenhouse Research

161,828 просмотров • 7 месяцев назад

CoreWeave cofounder brannin mcbee says older GPUs and servers still have strong demand, even as new generations ramp up: "Our pricing on Ampere went up during 2025, and our pricing on Hopper stayed within 10% or so of where we started the year." "Why is that happening? I think it's largely driven by inference and the idea that—not every workload needs the latest, greatest GPUs or servers." "There are different types of workloads that need different compute, and the market is efficiently pairing the right GPUs and servers with the workloads that need them." "We see incredibly robust demand all the way back to the Ampere generation, which I believe is six years old at this point."
1:37

CoreWeave cofounder brannin mcbee says older GPUs and servers still have strong demand, even as new generations ramp up: "Our pricing on Ampere went up during 2025, and our pricing on Hopper stayed within 10% or so of where we started the year." "Why is that happening? I think it's largely driven by inference and the idea that—not every workload needs the latest, greatest GPUs or servers." "There are different types of workloads that need different compute, and the market is efficiently pairing the right GPUs and servers with the workloads that need them." "We see incredibly robust demand all the way back to the Ampere generation, which I believe is six years old at this point."

TBPN

31,256 просмотров • 4 месяцев назад

$CRWV CEO says “older generations of GPUs are going to have a longer useful life than anyone anticipated” as CoreWeave signs a fully priced A100 contract through 2029 for hardware first launched in 2020. That means a 2020-era GPU can keep monetizing already-powered facilities that newer systems may not physically fit into while frontier workloads cascade to newer architectures and inference shifts down the stack. $NVDA CUDA strengthens that model by continuously optimizing Ampere, Hopper and Blackwell so the same workloads require less compute over time without forcing a hardware replacement.
0:25

$CRWV CEO says “older generations of GPUs are going to have a longer useful life than anyone anticipated” as CoreWeave signs a fully priced A100 contract through 2029 for hardware first launched in 2020. That means a 2020-era GPU can keep monetizing already-powered facilities that newer systems may not physically fit into while frontier workloads cascade to newer architectures and inference shifts down the stack. $NVDA CUDA strengthens that model by continuously optimizing Ampere, Hopper and Blackwell so the same workloads require less compute over time without forcing a hardware replacement.

Shay Boloor

135,057 просмотров • 6 дней назад

$AMZN CEO Andy Jassy describes AI adoption as very barbelled, with most existing enterprise workloads still not using inference though he expects that to change significantly over time. "I think it's actually kind of useful to look at at least our view of what we see in the demand and adoption curve right now, which is we see this adoption curve in AI right now is very barbelled." "There is on one end of the barbell, the AI labs are consuming gobs and gobs of compute. And there are a few runaway successful generative AI applications like Claude Code and ChatGPT." "And on the other end of the barbell are enterprises who are getting real value from AI in cost avoidance and productivity. And these are things like automating customer service or business process automation or fraud." "In the middle of the barbell is all of the current enterprise production workloads, some of which are using inference in a pervasive way, but most of which aren't. And that is going to change very significantly over time." "And that will be, in my opinion, that will be the largest absolute segment, the existing production workloads in the enterprise and new businesses and workloads that startups build too."
1:10

$AMZN CEO Andy Jassy describes AI adoption as very barbelled, with most existing enterprise workloads still not using inference though he expects that to change significantly over time. "I think it's actually kind of useful to look at at least our view of what we see in the demand and adoption curve right now, which is we see this adoption curve in AI right now is very barbelled." "There is on one end of the barbell, the AI labs are consuming gobs and gobs of compute. And there are a few runaway successful generative AI applications like Claude Code and ChatGPT." "And on the other end of the barbell are enterprises who are getting real value from AI in cost avoidance and productivity. And these are things like automating customer service or business process automation or fraud." "In the middle of the barbell is all of the current enterprise production workloads, some of which are using inference in a pervasive way, but most of which aren't. And that is going to change very significantly over time." "And that will be, in my opinion, that will be the largest absolute segment, the existing production workloads in the enterprise and new businesses and workloads that startups build too."

Fireside Alpha

29,242 просмотров • 19 дней назад

Cerebras just had the biggest IPO of the year. Founder Andrew Feldman says the 3 most important things he had to convince investors of while doing the roadshow were that demand for inference is going to 1,000,000x, the GPU isn't the only way to do compute, and that the CUDA moat is overstated. What he said: "Jensen said some time ago on Brad Gerstner's podcast that the demand for inference will grow by a 1,000,000x, and nobody believed him. And at the same time, you saw Sam Altman displaying real vision and going out and trying to lock up huge amounts of compute, memory, data centers, and power, because he saw it too." "[We tried] to share what that means — what exponential demand means. And that we're still so early, and yet the demand for AI compute is overwhelming." "The other thing is that there are lots of ways to do this. The GPU isn't the only way. You've got TPUs, Trainium, and us. There are lots of different ways to build a solution here." "And finally — the notion that CUDA is this grand lock-in is overplayed. Gemini 3, which is an excellent model, was trained on TPUs with no CUDA. The Anthropic models were trained on Trainium with no CUDA. Some of the best models, some of the most interesting things are being done without CUDA. And that lock-in might be overplayed." $CBRS
1:46

Cerebras just had the biggest IPO of the year. Founder Andrew Feldman says the 3 most important things he had to convince investors of while doing the roadshow were that demand for inference is going to 1,000,000x, the GPU isn't the only way to do compute, and that the CUDA moat is overstated. What he said: "Jensen said some time ago on Brad Gerstner's podcast that the demand for inference will grow by a 1,000,000x, and nobody believed him. And at the same time, you saw Sam Altman displaying real vision and going out and trying to lock up huge amounts of compute, memory, data centers, and power, because he saw it too." "[We tried] to share what that means — what exponential demand means. And that we're still so early, and yet the demand for AI compute is overwhelming." "The other thing is that there are lots of ways to do this. The GPU isn't the only way. You've got TPUs, Trainium, and us. There are lots of different ways to build a solution here." "And finally — the notion that CUDA is this grand lock-in is overplayed. Gemini 3, which is an excellent model, was trained on TPUs with no CUDA. The Anthropic models were trained on Trainium with no CUDA. Some of the best models, some of the most interesting things are being done without CUDA. And that lock-in might be overplayed." $CBRS

TBPN

36,474 просмотров • 3 месяцев назад

Speaking of spot on. Gavin Gavin Baker was asked in early Jan'25 his prediction for best performing asset class of the year (as @jason references in latest The All-In Podcast): "I think the companies that make high bandwidth memory. Going back to "we're going to run out of compute," it's actually a pretty shocking stat: high bandwidth memory is a bigger part of NVIDIA's COGS on GPUs than Taiwan Semi is. And today there's two companies that can make it, SK Hynix and Micron. We'll see if Samsung gets their act together. But HBM, it's in NVIDIA GPUs, AMD GPUs, Amazon Trainiums. And particularly in a world of test-time compute and inference being so important, high bandwidth memory is arguably more important than it ever was. And it's been sold out for the last two years. So high bandwidth memory would be my pick."
0:52

Speaking of spot on. Gavin Gavin Baker was asked in early Jan'25 his prediction for best performing asset class of the year (as @jason references in latest The All-In Podcast): "I think the companies that make high bandwidth memory. Going back to "we're going to run out of compute," it's actually a pretty shocking stat: high bandwidth memory is a bigger part of NVIDIA's COGS on GPUs than Taiwan Semi is. And today there's two companies that can make it, SK Hynix and Micron. We'll see if Samsung gets their act together. But HBM, it's in NVIDIA GPUs, AMD GPUs, Amazon Trainiums. And particularly in a world of test-time compute and inference being so important, high bandwidth memory is arguably more important than it ever was. And it's been sold out for the last two years. So high bandwidth memory would be my pick."

Fireside Alpha

66,227 просмотров • 1 месяц назад

David Friedberg: Michael Burry’s Datacenter Math is Wrong “I actually think Michael Burberry's got this wrong.” “What Michael Burry is saying is that all of these hyperscalers have extended their depreciation schedule or the useful life of their data centers by roughly 2x, which cuts the operating costs in half when they report it in earnings. And so it's making their earnings inflate.” “So he's claiming they're cooking the books. Google first made this change in Q1 of 2021, where they said the servers are now going from 3 to 4 years. Separately in 2021, Google took networking equipment from 3 to 5 years. And then in 2023, they took it from 5 to 6 years.” “And so this is a result of this effort where they went in and did an analysis. So what happened?” “What happened in the data centers is that the data centers transitioned from being primarily data storage and data transfer systems, where you would use hard drives and RAM and memory to store data and then transmit it back out, to being data processing centers because of the AI boom.” “So as AI became more important in the data center, more of the dollars that are going into data centers were allocated towards chips from data storage, which initially was hard drives.” “And then suddenly, when you put these processors in to process the data to do AI, the majority of the spend and the majority of the energy is going towards the processors.” “I made some calls and I checked around with some other friends, and everyone says the same thing: that these 7-8 year old TPUs and GPUs that are sitting in the data centers are still being used and they're being used at 100% utilization.” “So that actually justifies and validates the depreciation schedule being much longer versus shorter.”
1:33

David Friedberg: Michael Burry’s Datacenter Math is Wrong “I actually think Michael Burberry's got this wrong.” “What Michael Burry is saying is that all of these hyperscalers have extended their depreciation schedule or the useful life of their data centers by roughly 2x, which cuts the operating costs in half when they report it in earnings. And so it's making their earnings inflate.” “So he's claiming they're cooking the books. Google first made this change in Q1 of 2021, where they said the servers are now going from 3 to 4 years. Separately in 2021, Google took networking equipment from 3 to 5 years. And then in 2023, they took it from 5 to 6 years.” “And so this is a result of this effort where they went in and did an analysis. So what happened?” “What happened in the data centers is that the data centers transitioned from being primarily data storage and data transfer systems, where you would use hard drives and RAM and memory to store data and then transmit it back out, to being data processing centers because of the AI boom.” “So as AI became more important in the data center, more of the dollars that are going into data centers were allocated towards chips from data storage, which initially was hard drives.” “And then suddenly, when you put these processors in to process the data to do AI, the majority of the spend and the majority of the energy is going towards the processors.” “I made some calls and I checked around with some other friends, and everyone says the same thing: that these 7-8 year old TPUs and GPUs that are sitting in the data centers are still being used and they're being used at 100% utilization.” “So that actually justifies and validates the depreciation schedule being much longer versus shorter.”

The All-In Podcast

304,297 просмотров • 9 месяцев назад

Marc Andreessen says AI is hyper democratizing. “Is the future of the world going to be one or a small number of companies, or for that matter, governments or super AIs that kind of own and control everything?” “Or is this a technology that's going to diffuse out and be in everybody's hands?” “This is the time in which AI is actually hyper democratizing. It has spread to more individuals, both in the country and around the world in the shortest period of time of any new technology I think in history.” “The best AIs in the world are in the consumer products.” “This technology is going to be in everybody's hands. And everybody is going to be able to use it to optimize the things that they do, have it be a thought partner, assistant for building companies.” Marc Andreessen 🇺🇸
1:34

Marc Andreessen says AI is hyper democratizing. “Is the future of the world going to be one or a small number of companies, or for that matter, governments or super AIs that kind of own and control everything?” “Or is this a technology that's going to diffuse out and be in everybody's hands?” “This is the time in which AI is actually hyper democratizing. It has spread to more individuals, both in the country and around the world in the shortest period of time of any new technology I think in history.” “The best AIs in the world are in the consumer products.” “This technology is going to be in everybody's hands. And everybody is going to be able to use it to optimize the things that they do, have it be a thought partner, assistant for building companies.” Marc Andreessen 🇺🇸

a16z

79,935 просмотров • 9 месяцев назад

IN NEWS: Baseten raises a $150M series D round. Tuhin Srivastava (Founder & CEO, Baseten) on the future of inference: “I think the token price goes down and inference should get cheaper over time. And that really just means there is going to be more inference.” “Every time we lower prices or optimize models to make it cheaper, four months later customers are spending more anyway.” “Inference prices will go down, but if the world is run by AI in 10 years, there is going to be a lot of inference. It better be cheap.”
0:52

IN NEWS: Baseten raises a $150M series D round. Tuhin Srivastava (Founder & CEO, Baseten) on the future of inference: “I think the token price goes down and inference should get cheaper over time. And that really just means there is going to be more inference.” “Every time we lower prices or optimize models to make it cheaper, four months later customers are spending more anyway.” “Inference prices will go down, but if the world is run by AI in 10 years, there is going to be a lot of inference. It better be cheap.”

TBPN

17,057 просмотров • 11 месяцев назад

Gavin Baker: “I think Anthropic is worth $3 trillion today.” Gavin laid out the case on the most recent All-In Podcast: – They will end 2026 with over $100B in revenue – They will likely end 2028 with $200-$300B in revenue – At that scale, Anthropic will be very profitable because compute will be inference-dominated – Anthropic reportedly has 85% gross margins on inference – The market has already absorbed Anthropic’s valuation, it will just shift from private to public @jason: “I'm sorry, did you say Anthropic is worth $3 trillion?” Gavin: “Yeah, I think that is roughly where it would probably trade as a public company.” Travis Kalanick: “Wow. Oh man. Holy sh*t.” Gavin: “They're going to end this year with well over $100 billion (in revenue). So what's the '28 number? Is it $200 billion? Is it $300 billion?” “It's probably not going to trade at 10 times that number, and it will be very profitable at that scale because it'll be inference-dominated and people are reporting they have 85% gross margins on inference.” “But in terms of the market absorbing this, like the market's already absorbed it. It's just shifting from private to public.” “And so in the scale of global capital markets, these seem like really big numbers, but you're just moving from the private markets to the public markets, which are even bigger.”
0:55

Gavin Baker: “I think Anthropic is worth $3 trillion today.” Gavin laid out the case on the most recent All-In Podcast: – They will end 2026 with over $100B in revenue – They will likely end 2028 with $200-$300B in revenue – At that scale, Anthropic will be very profitable because compute will be inference-dominated – Anthropic reportedly has 85% gross margins on inference – The market has already absorbed Anthropic’s valuation, it will just shift from private to public @jason: “I'm sorry, did you say Anthropic is worth $3 trillion?” Gavin: “Yeah, I think that is roughly where it would probably trade as a public company.” Travis Kalanick: “Wow. Oh man. Holy sh*t.” Gavin: “They're going to end this year with well over $100 billion (in revenue). So what's the '28 number? Is it $200 billion? Is it $300 billion?” “It's probably not going to trade at 10 times that number, and it will be very profitable at that scale because it'll be inference-dominated and people are reporting they have 85% gross margins on inference.” “But in terms of the market absorbing this, like the market's already absorbed it. It's just shifting from private to public.” “And so in the scale of global capital markets, these seem like really big numbers, but you're just moving from the private markets to the public markets, which are even bigger.”

The All-In Podcast

85,013 просмотров • 1 месяц назад

Godfather of AI Yann LeCun reveals the humanoid robot industry's secret: the companies building the robots have no idea how to make them smart enough to be useful. "There is a lot of expectations that robotics is going to be a revolution, essentially, in a big market, eventually." "But the secret of the industry is that none of the companies at the moment that are building robots, like humanoid robots, none of those companies has any idea how to make those robots smart enough to be useful." "There's a big push for AI to make progress towards... understanding the physical world. That's what is prompting this investment." "There's nothing wrong with investing in LLMs. LLMs are useful. They're just addressing a completely different set of applications, really." LeCun is not describing someone else's problem. He has spent 15 years on world models, raised more than $1 billion for AMI Labs to build that missing intelligence layer, and in the same interview he calls 2026 the year of the world model and puts consumer robots 3 to 5 years out. Every humanoid robot demo you have seen is a body waiting for a brain. The winner in robotics will not be the company that shipped the body; it will be the one that ships the intelligence. - Yann LeCun (Yann LeCun), Turing Award winner and founder of AMI Labs, on Bloomberg Television.
1:01

Godfather of AI Yann LeCun reveals the humanoid robot industry's secret: the companies building the robots have no idea how to make them smart enough to be useful. "There is a lot of expectations that robotics is going to be a revolution, essentially, in a big market, eventually." "But the secret of the industry is that none of the companies at the moment that are building robots, like humanoid robots, none of those companies has any idea how to make those robots smart enough to be useful." "There's a big push for AI to make progress towards... understanding the physical world. That's what is prompting this investment." "There's nothing wrong with investing in LLMs. LLMs are useful. They're just addressing a completely different set of applications, really." LeCun is not describing someone else's problem. He has spent 15 years on world models, raised more than $1 billion for AMI Labs to build that missing intelligence layer, and in the same interview he calls 2026 the year of the world model and puts consumer robots 3 to 5 years out. Every humanoid robot demo you have seen is a body waiting for a brain. The winner in robotics will not be the company that shipped the body; it will be the one that ships the intelligence. - Yann LeCun (Yann LeCun), Turing Award winner and founder of AMI Labs, on Bloomberg Television.

Karl Mehta

44,400 просмотров • 1 месяц назад

The Way Forward & A New Chapter For AI powered by GamerHash AI & $GHX. Welcome to the future of bleeding edge connectivity and cloud infrastructure ⚡ GamerHash AI is the answer to the billion dollar question on how to offset the GPU scarcity, enable exponential growth of AI through utilizing consumer-level GPUs and establishing a powerful sharing economy by sourcing a global network of gamers. The synergy of our GPU #DePIN and blockchain for #AI #inference is the technology of the future that already works today. See you in the network!
1:01

The Way Forward & A New Chapter For AI powered by GamerHash AI & $GHX. Welcome to the future of bleeding edge connectivity and cloud infrastructure ⚡ GamerHash AI is the answer to the billion dollar question on how to offset the GPU scarcity, enable exponential growth of AI through utilizing consumer-level GPUs and establishing a powerful sharing economy by sourcing a global network of gamers. The synergy of our GPU #DePIN and blockchain for #AI #inference is the technology of the future that already works today. See you in the network!

GamerHash AI

1,866,164 просмотров • 1 год назад

The creator of High Bandwidth Memory (HBM) put a number on the AI build that should stop every infra investor cold. A cluster of a million GPUs runs at roughly 10-20% utilization (Save this). Kim Jung-ho spent thirty years building what feeds the GPU, and his claim is that the GPU is barely working. Here is what is actually happening. Every time a model generates output, the data has to be read out of memory, computed, and written back. The read and the write swallow almost the entire cycle. While that data moves, the GPU does nothing. It sits there, fully powered, fully paid for, waiting. By Kim's estimate the memory is doing only about 30 percent of the work it needs to do. The processor idles the rest. So a million installed GPUs run at 10 to 20 percent. You are not compute constrained. You are memory constrained, and the expensive part is standing around. Adding more GPUs does not fix this. It gives you more processors starving for the same data. Here is the part that decides the next decade. Memory can grow. When a cell cannot shrink any further, you stack it into a high-rise, layer on layer. A GPU cannot be stacked. It runs too hot and needs a cooler bolted to its back, so the one move that rescues memory is closed to the processor. The thing that can keep stacking compounds. The thing that cannot plateaus. The marginal dollar in an AI build now buys more by fixing the memory path than by bolting on another idle GPU. Which is why the companies that control memory bandwidth and supply are not suppliers to the AI trade. They are the AI trade.
0:50

The creator of High Bandwidth Memory (HBM) put a number on the AI build that should stop every infra investor cold. A cluster of a million GPUs runs at roughly 10-20% utilization (Save this). Kim Jung-ho spent thirty years building what feeds the GPU, and his claim is that the GPU is barely working. Here is what is actually happening. Every time a model generates output, the data has to be read out of memory, computed, and written back. The read and the write swallow almost the entire cycle. While that data moves, the GPU does nothing. It sits there, fully powered, fully paid for, waiting. By Kim's estimate the memory is doing only about 30 percent of the work it needs to do. The processor idles the rest. So a million installed GPUs run at 10 to 20 percent. You are not compute constrained. You are memory constrained, and the expensive part is standing around. Adding more GPUs does not fix this. It gives you more processors starving for the same data. Here is the part that decides the next decade. Memory can grow. When a cell cannot shrink any further, you stack it into a high-rise, layer on layer. A GPU cannot be stacked. It runs too hot and needs a cooler bolted to its back, so the one move that rescues memory is closed to the processor. The thing that can keep stacking compounds. The thing that cannot plateaus. The marginal dollar in an AI build now buys more by fixing the memory path than by bolting on another idle GPU. Which is why the companies that control memory bandwidth and supply are not suppliers to the AI trade. They are the AI trade.

Fireside Alpha

38,370 просмотров • 1 месяц назад

Bessent: "These 2008, 2009, 2010 financial rules were too tight. They hamstrung the American financial system. It was time for a change. We're gonna be safe, smart, and sound in terms of our deregulation. But we have to take the financial system out of this straitjacket. This substantial increase in private credit which is outside of the regulated banking system -- that tells me that the regulated system is too constrained."
0:50

Bessent: "These 2008, 2009, 2010 financial rules were too tight. They hamstrung the American financial system. It was time for a change. We're gonna be safe, smart, and sound in terms of our deregulation. But we have to take the financial system out of this straitjacket. This substantial increase in private credit which is outside of the regulated banking system -- that tells me that the regulated system is too constrained."

Aaron Rupar

706,619 просмотров • 8 месяцев назад

NVIDIA just convinced the biggest credit firms on earth to bet $500 billion that Michael Burry is wrong The bet is over ONE number: how long does a GPU live? Jensen thinks it's longer than most and a new asset class w/ a secondary market The hyperscalers write off GPUs over 5 to 6 years (Amazon uses 5, Meta 5.5, Google 6) On the other hand, Burry's famous bear thesis says the real life is 2 to 3 years becuase the newer models make the older models effectively worthless. If true, this would mean the industry is overstating profits by roughly $176 billion from 2026 to 2028 Here's why this one number matters so much. On every $100B of GPU capex: 1. 3 year life = $33B a year of depreciation expense 2. 6 year life = $17B a year 3. 8 year life with resale value = $10B a year The gap between Burry's world and Jensen's world is $20B+ of profit per year on every $100B spent, and the industry now spends well over $500B a year! So who's right and is this new $500B AI infra fund a risk on the market? Let's look at what the market is paying for it right now: A100s are 6 year old chips, fully depreciated, worth roughly zero on the books, and their rental price is UP about 25% this year (see the chart I reposted below) A supposedly dead asset from Burry has been climbing all year Azure ran V100s for 7.5 years before retiring them, Google still runs 7 to 8 year old TPUs in production. Old GPUs don't seem to die, they get demoted from training to inference and then continue earning cash That's what Jensen showed the credit desks, and it changes everything, because an asset with real resale value can be financed more similar to something like an aircraft Airlines lease planes instead of buying them because lenders trust the residual value. If GPUs work the same way, customers don't need equity to buy chips anymore, credit does it, and the pool of NVIDIA buyers expands massively It looks like Jensen is going to be right about this (which makes sense, he is the King of GPUs) and $500 billion of the smartest credit money on earth too his side too This is extremely bullish for NVIDIA of course, but also the hyperscalers and even neoclouds who are the main buyers/sellers of GPUs. The longer their assets retain value, the more they will buy and resell If you want to see how I'm investing in this market you can track my entire portfolio and research in real-time, alongside 4 other analysts, inside Milk Road PRO. It's currently just $1 to try it out (insane price just to check it out) and you can learn more and join here: Good luck out there!
1:59

NVIDIA just convinced the biggest credit firms on earth to bet $500 billion that Michael Burry is wrong The bet is over ONE number: how long does a GPU live? Jensen thinks it's longer than most and a new asset class w/ a secondary market The hyperscalers write off GPUs over 5 to 6 years (Amazon uses 5, Meta 5.5, Google 6) On the other hand, Burry's famous bear thesis says the real life is 2 to 3 years becuase the newer models make the older models effectively worthless. If true, this would mean the industry is overstating profits by roughly $176 billion from 2026 to 2028 Here's why this one number matters so much. On every $100B of GPU capex: 1. 3 year life = $33B a year of depreciation expense 2. 6 year life = $17B a year 3. 8 year life with resale value = $10B a year The gap between Burry's world and Jensen's world is $20B+ of profit per year on every $100B spent, and the industry now spends well over $500B a year! So who's right and is this new $500B AI infra fund a risk on the market? Let's look at what the market is paying for it right now: A100s are 6 year old chips, fully depreciated, worth roughly zero on the books, and their rental price is UP about 25% this year (see the chart I reposted below) A supposedly dead asset from Burry has been climbing all year Azure ran V100s for 7.5 years before retiring them, Google still runs 7 to 8 year old TPUs in production. Old GPUs don't seem to die, they get demoted from training to inference and then continue earning cash That's what Jensen showed the credit desks, and it changes everything, because an asset with real resale value can be financed more similar to something like an aircraft Airlines lease planes instead of buying them because lenders trust the residual value. If GPUs work the same way, customers don't need equity to buy chips anymore, credit does it, and the pool of NVIDIA buyers expands massively It looks like Jensen is going to be right about this (which makes sense, he is the King of GPUs) and $500 billion of the smartest credit money on earth too his side too This is extremely bullish for NVIDIA of course, but also the hyperscalers and even neoclouds who are the main buyers/sellers of GPUs. The longer their assets retain value, the more they will buy and resell If you want to see how I'm investing in this market you can track my entire portfolio and research in real-time, alongside 4 other analysts, inside Milk Road PRO. It's currently just $1 to try it out (insane price just to check it out) and you can learn more and join here: Good luck out there!

Kyle Reidhead | Milk Road

268,175 просмотров • 11 дней назад

Gavin Baker: "if you're a foundation model company and you do not have unique data and internet scale distribution, you are the fastest depreciating asset in human history - I think most of these companies are zeros, and there's like 10 of them" this is him explaining why the 100 largest buildings in the world will be data centers, why there are only three places left in America you can legally build one, and what he says is setting up the first real memory cycle since 1996 "in our lifetime there will be data centers that are bigger than Manhattan" "the two rate limiting factors for AI are no longer compute and GPUs - it's power, finding places to plug the GPUs in" "there's only three places in the United States you can put one of these - it's got to be near multiple nuclear reactors that are unregulated within 50 miles of each other" "if Apple doubles the memory in the iPhone and Google does the same for Android at the same time, you could have the first real memory cycle since 1996 - and in 1996 Micron was a 50 bagger" "the only hope most of these companies have is for Amazon to acquire them the way Microsoft did with Inflection" bookmark & watch the full conversation ↓
11:48

Gavin Baker: "if you're a foundation model company and you do not have unique data and internet scale distribution, you are the fastest depreciating asset in human history - I think most of these companies are zeros, and there's like 10 of them" this is him explaining why the 100 largest buildings in the world will be data centers, why there are only three places left in America you can legally build one, and what he says is setting up the first real memory cycle since 1996 "in our lifetime there will be data centers that are bigger than Manhattan" "the two rate limiting factors for AI are no longer compute and GPUs - it's power, finding places to plug the GPUs in" "there's only three places in the United States you can put one of these - it's got to be near multiple nuclear reactors that are unregulated within 50 miles of each other" "if Apple doubles the memory in the iPhone and Google does the same for Android at the same time, you could have the first real memory cycle since 1996 - and in 1996 Micron was a 50 bagger" "the only hope most of these companies have is for Amazon to acquire them the way Microsoft did with Inflection" bookmark & watch the full conversation ↓

Сarm1ne

274,201 просмотров • 19 дней назад

In just 3 minutes, Ken Griffin (Citadel) & Larry Fink (BlackRock) explain the current state of government overspending, AI hype, & 2026 $600B data center CapEx "The world needs a savior, & the hope is that AI is the savior that we need for productivity." "The the area of recklessness is the spending of governments around the world, who are all, with little exception, all spending well beyond their means." Big issue: "Will AI create the productivity acceleration that is honestly just hoped for in Washington & in the halls of government around the world as a ways to overcome the profligate spending that we're currently engaged in?" Citadel BlackRock World Economic Forum . . . Ken Griffin "Let's take a step back and and talk about where we are right here right now. The the area of recklessness is the spending of governments around the world, who are all, with little exception, all spending well beyond their means. That's the recklessness of this moment in history. This is not a parallel to the 1920s in terms of the recklessness of the of the private capital markets. It's a story of the recklessness of government spending. Within the private sector there's a huge question as to where AI will take us. And I, I was carefully taking notes and listening to what Larry has to say, or to what Madame Lagarde has to say, because this is one of the big issues of our moment. Will AI create the productivity acceleration that is honestly just hoped for in Washington and in the halls of government around the world as a ways to overcome the profligate spending that we're currently engaged in? The world, the world needs a savior, and the hope is that AI is the savior that we need for productivity. And the challenge with this is it is, it may or may not be. We just don't know yet. Now there's a tremendous amount of hype around AI, and in some sense the large AI companies need to create that hype to raise the tens — or actually 100, hundreds of billions, right — of billions of dollars of investment that are going into the field. Like you wouldn't be able to raise hundreds of billions of dollars. We'll spend — and Larry can probably correct me on this — but roughly $600 billion this year in cap ex for data centers in the United States. Larry Fink "I think it could be larger." Andrew Ross Sorkin "But does that mean that it's getting hyped up too much, or it's just the the hype is required as a sales mechanism?" Larry Fink "First of all, so much of the data centers are being built for cloud, right? And and the big issue is gonna be in terms of monetization of of of the spend. The data centers are being built for AI requires more advanced chips. The question is what is the lifetime of that chip. If we have new technological changes in the lifetime of the chip in one year, then that spend is gonna be really a bad spend. If the lifetime as they expect it to be is 4 or 5 years and then those chips can be used for cloud, then then I think these investments are gonna prove to be good investments. So I think it's it's gonna be — you know, if the speed of technology changes and all these investments now, they're gonna be — it's gonna be a challenge. But I agree with Ken. I think we don't know enough, but I'm personally very optimistic on how AI is gonna affect the world economy."
3:10

In just 3 minutes, Ken Griffin (Citadel) & Larry Fink (BlackRock) explain the current state of government overspending, AI hype, & 2026 $600B data center CapEx "The world needs a savior, & the hope is that AI is the savior that we need for productivity." "The the area of recklessness is the spending of governments around the world, who are all, with little exception, all spending well beyond their means." Big issue: "Will AI create the productivity acceleration that is honestly just hoped for in Washington & in the halls of government around the world as a ways to overcome the profligate spending that we're currently engaged in?" Citadel BlackRock World Economic Forum . . . Ken Griffin "Let's take a step back and and talk about where we are right here right now. The the area of recklessness is the spending of governments around the world, who are all, with little exception, all spending well beyond their means. That's the recklessness of this moment in history. This is not a parallel to the 1920s in terms of the recklessness of the of the private capital markets. It's a story of the recklessness of government spending. Within the private sector there's a huge question as to where AI will take us. And I, I was carefully taking notes and listening to what Larry has to say, or to what Madame Lagarde has to say, because this is one of the big issues of our moment. Will AI create the productivity acceleration that is honestly just hoped for in Washington and in the halls of government around the world as a ways to overcome the profligate spending that we're currently engaged in? The world, the world needs a savior, and the hope is that AI is the savior that we need for productivity. And the challenge with this is it is, it may or may not be. We just don't know yet. Now there's a tremendous amount of hype around AI, and in some sense the large AI companies need to create that hype to raise the tens — or actually 100, hundreds of billions, right — of billions of dollars of investment that are going into the field. Like you wouldn't be able to raise hundreds of billions of dollars. We'll spend — and Larry can probably correct me on this — but roughly $600 billion this year in cap ex for data centers in the United States. Larry Fink "I think it could be larger." Andrew Ross Sorkin "But does that mean that it's getting hyped up too much, or it's just the the hype is required as a sales mechanism?" Larry Fink "First of all, so much of the data centers are being built for cloud, right? And and the big issue is gonna be in terms of monetization of of of the spend. The data centers are being built for AI requires more advanced chips. The question is what is the lifetime of that chip. If we have new technological changes in the lifetime of the chip in one year, then that spend is gonna be really a bad spend. If the lifetime as they expect it to be is 4 or 5 years and then those chips can be used for cloud, then then I think these investments are gonna prove to be good investments. So I think it's it's gonna be — you know, if the speed of technology changes and all these investments now, they're gonna be — it's gonna be a challenge. But I agree with Ken. I think we don't know enough, but I'm personally very optimistic on how AI is gonna affect the world economy."

Molly O’Shea

33,210 просмотров • 6 месяцев назад

Gavin Baker says AI will make video games 90% cheaper to make "In the short term, AI is going to be amazing for the video game platforms. I think AI's impact on video games has been very misunderstood. AI world models are going to lower the cost of developing a game by 90%." "If it used to cost $300 million to make a AAA title like Call of Duty, or $200 million or $100 million or whatever the number is, it goes down 90%." "This might end up being really bad for companies that make video games, because you're going to have a lot more competition. But if you were a video game platform, and we all know what those platforms are, this is going to be really good for you, because there's going to be an explosion of content."
1:07

Gavin Baker says AI will make video games 90% cheaper to make "In the short term, AI is going to be amazing for the video game platforms. I think AI's impact on video games has been very misunderstood. AI world models are going to lower the cost of developing a game by 90%." "If it used to cost $300 million to make a AAA title like Call of Duty, or $200 million or $100 million or whatever the number is, it goes down 90%." "This might end up being really bad for companies that make video games, because you're going to have a lot more competition. But if you were a video game platform, and we all know what those platforms are, this is going to be really good for you, because there's going to be an explosion of content."

dnap

218,282 просмотров • 1 месяц назад