正在加载视频...

视频加载失败

返回首页

SF Compute CEO evan conrad explains why GPU clusters are bad collateral for lenders: "This article was more of a prediction of a trend that is already happening. I probably should’ve said, currently, most startups cannot buy compute. It is only getting harder and harder and harder." "If you...

MTSlive's profile picture

MTS

572,777 subscribers

17,071 次观看 • 3 天前 •via X (Twitter)

科学技术财经
LIVE

Anya Rossi• Live Now

Private livecam show

9 条评论

Kushal Mungee 的头像
Kushal Mungee3 天前

@evanjconrad Interesting how @sfcompute addresses this by not owning the hardware itself but basically operating everything around it, so that the lender can claim "operational chips" as collateral rather than just the hardware itself

brexton 的头像
brexton3 天前

@evanjconrad Inject this into my veins

Carmen Li 的头像
Carmen Li2 天前

GPUs aren’t inherently bad collateral. Opaque, unverified, and illiquid collateral is difficult to finance. Better data won’t eliminate credit risk. But it can make risk measurable, transparent, and more manageable. The answer isn’t to declare GPU financing broken. It’s to build the market infrastructure. We’re still in the very early innings. 🚀

Carmen Li 的头像
Carmen Li2 天前

4) Third, correlated credit risk. This is where I agree most. If neoclouds default during a compute downturn, GPU prices may fall exactly when lenders need to liquidate. That’s real wrong-way risk. But aviation, shipping, energy, and real estate financing face similar challenges.

Carmen Li 的头像
Carmen Li2 天前

@evanjconrad 1) I agree with @evanjconrad on correlated risk in GPU financing. But I disagree that GPUs are inherently bad collateral. We’re mixing up 3 different problems: Asset valuation risk Recovery/liquidation risk Borrower credit risk Let’s break them down. 🧵

Carmen Li 的头像
Carmen Li2 天前

5) We don’t conclude aircraft or ships are inherently unfinanceable. We build valuation standards, collateral monitoring, secondary markets, and appropriate financing structures. AI compute needs the same financial infrastructure. That’s what we’ve been building at @silicondata.

Carmen Li 的头像
Carmen Li2 天前

3) Second, liquidation. A functioning GPU cluster may be worth more than disconnected servers. Evan is right. But saying nobody wants to buy the hardware is too broad. There IS a secondary GPU market. @computeexchange The questions are: At what price? How quickly? At what redeployment cost? Liquidity isn’t binary. It has a price.

Carmen Li 的头像
Carmen Li2 天前

@evanjconrad 2) First, valuation. What’s an H100 or B200 worth today? What will it be worth in 12, 24, or 36 months? Without independent pricing, residual value benchmarks, and performance verification, lenders are underwriting with incomplete information. That’s a market data problem.

Kuldeep Pisda 的头像
Kuldeep Pisda2 天前

@evanjconrad correlated risk is why the contract matters more than the box, a take-or-pay offtake from a solid buyer is what made solar farms financeable

相关视频

Jeff Dean says he's bullish on specialized hardware because a handful of workloads will be a lot of the world's compute "I'm pretty bullish on more and more specialized hardware, because I think that's the way you really get much more efficient systems." "And we now have workloads where there's like a handful of workloads that are going to be a lot of the compute in the world." "And so if you think about that, that is crying out for specialization." "The problem with specialization is if what you want to do changes in the future, then the thing you've lovingly crafted in hardware for maybe two years is no longer perhaps as relevant."
0:35

Jeff Dean says he's bullish on specialized hardware because a handful of workloads will be a lot of the world's compute "I'm pretty bullish on more and more specialized hardware, because I think that's the way you really get much more efficient systems." "And we now have workloads where there's like a handful of workloads that are going to be a lot of the compute in the world." "And so if you think about that, that is crying out for specialization." "The problem with specialization is if what you want to do changes in the future, then the thing you've lovingly crafted in hardware for maybe two years is no longer perhaps as relevant."

Fireside Alpha

37,976 次观看 • 1 个月前

SF Compute CEO evan conrad advises AI founders to raise 10X more capital than planned, potentially $1 billion, just to secure compute: "I would raise as much money as I can. I would take your round and I would add 10X to it. And if I was an investor, I would buy compute instead." "I would either try to find a big investment grade normal company that I can partner with and use their balance sheet to co-sign on the lease with me, or I would try to raise the biggest round I possibly can, so I’m above the bar." "If you’re trying to raise a round, I think you should try and raise like a billion dollars. I know. That’s the problem. This is why the article was called End of Startups. I don’t think many people can raise a billion dollar round." "If you don’t fix the debt problem, most of the portfolios of various venture firms are gonna consolidate or get wiped out, or they will convert into basically GPU clouds." San Francisco Compute
1:29

SF Compute CEO evan conrad advises AI founders to raise 10X more capital than planned, potentially $1 billion, just to secure compute: "I would raise as much money as I can. I would take your round and I would add 10X to it. And if I was an investor, I would buy compute instead." "I would either try to find a big investment grade normal company that I can partner with and use their balance sheet to co-sign on the lease with me, or I would try to raise the biggest round I possibly can, so I’m above the bar." "If you’re trying to raise a round, I think you should try and raise like a billion dollars. I know. That’s the problem. This is why the article was called End of Startups. I don’t think many people can raise a billion dollar round." "If you don’t fix the debt problem, most of the portfolios of various venture firms are gonna consolidate or get wiped out, or they will convert into basically GPU clouds." San Francisco Compute

MTS

17,895 次观看 • 3 天前

Gabbit Robotics co-founder Nicole Mattero on why the next AI winners won't fight over your screen, they'll fight to get closer to you: She kicks things off with what she admits is a hot take: "My hot take is that hardware is the only moat going forward because anyone's going to be able to spin up a model and make it really good and really cheap." Her logic is simple. Once a strong model is cheap and easy for anyone to build, the model alone can't set you apart. So what actually protects you? "Where does that leave you? Basically companies that have tactile modes such as hardware are the only ones that actually survive." Hardware, though, is only one piece of her argument. The second piece is data: "The models that win in the long run are the ones where private data is their [moat]." From there, Mattero ties the two together. Hardware gives a company one moat. The ability to gather unique personal data gives it another. Put both in place and the edge starts to multiply: "When you have a company that has hardware as a [moat], but then B is able to collect really unique personal data, that also becomes a [moat]." And that, she says, is the reason so many foundation model companies are reaching beyond the screen and into the physical world: "That's why I think a lot of these foundation models are like, let's get something in the home or on the person. It gives them an edge on both fronts."
0:45

Gabbit Robotics co-founder Nicole Mattero on why the next AI winners won't fight over your screen, they'll fight to get closer to you: She kicks things off with what she admits is a hot take: "My hot take is that hardware is the only moat going forward because anyone's going to be able to spin up a model and make it really good and really cheap." Her logic is simple. Once a strong model is cheap and easy for anyone to build, the model alone can't set you apart. So what actually protects you? "Where does that leave you? Basically companies that have tactile modes such as hardware are the only ones that actually survive." Hardware, though, is only one piece of her argument. The second piece is data: "The models that win in the long run are the ones where private data is their [moat]." From there, Mattero ties the two together. Hardware gives a company one moat. The ability to gather unique personal data gives it another. Put both in place and the edge starts to multiply: "When you have a company that has hardware as a [moat], but then B is able to collect really unique personal data, that also becomes a [moat]." And that, she says, is the reason so many foundation model companies are reaching beyond the screen and into the physical world: "That's why I think a lot of these foundation models are like, let's get something in the home or on the person. It gives them an edge on both fronts."

Big Brain AI

19,342 次观看 • 5 个月前

Gavin Baker says the compute market is now in contango per Silicon Data and his portfolio companies are shocked at how much more they have to pay for clusters in 2027 “You know, Silicon Data, which I do think is, my firm is an investor in them, but I do think they’re the most authoritative data source. Carmen Li, the founder, stood up at GTC and said that the compute market is now in contango. It’s not just that the spot market is tight. If you want a cluster a year from now, you have to pay a lot of money." "And then this is maybe another discontinuity between public and private markets. But just the number of companies we have in our portfolio who are like, okay, we got a cluster of 4,000 Blackwells, and we got it in May or June ‘26, and we wanna bring on another cluster of 4,000 in September 2027, and they’re shocked by how much more they have to pay in the future." "And it’s not like everybody is unaware of this. Everybody’s aware of the supply that’s coming on. But even with all of that supply coming on, it is hard for the companies in our portfolio to find compute even well into ‘27. So for, you know, all the commodity experts who are focused on backwardation and what that implied, I got news for you. We’re not in backwardation anymore.”
1:24

Gavin Baker says the compute market is now in contango per Silicon Data and his portfolio companies are shocked at how much more they have to pay for clusters in 2027 “You know, Silicon Data, which I do think is, my firm is an investor in them, but I do think they’re the most authoritative data source. Carmen Li, the founder, stood up at GTC and said that the compute market is now in contango. It’s not just that the spot market is tight. If you want a cluster a year from now, you have to pay a lot of money." "And then this is maybe another discontinuity between public and private markets. But just the number of companies we have in our portfolio who are like, okay, we got a cluster of 4,000 Blackwells, and we got it in May or June ‘26, and we wanna bring on another cluster of 4,000 in September 2027, and they’re shocked by how much more they have to pay in the future." "And it’s not like everybody is unaware of this. Everybody’s aware of the supply that’s coming on. But even with all of that supply coming on, it is hard for the companies in our portfolio to find compute even well into ‘27. So for, you know, all the commodity experts who are focused on backwardation and what that implied, I got news for you. We’re not in backwardation anymore.”

Fireside Alpha

100,301 次观看 • 3 天前

Security is boring until it isn’t. The phone in your pocket is a highly secure hardware module. It proves the compute is real and keeps workloads confidential. That is why mobile devices can power trustless, decentralized compute without fake hardware or prying eyes. Want to put your phone to work? Get started on Acurast today.
1:59

Security is boring until it isn’t. The phone in your pocket is a highly secure hardware module. It proves the compute is real and keeps workloads confidential. That is why mobile devices can power trustless, decentralized compute without fake hardware or prying eyes. Want to put your phone to work? Get started on Acurast today.

Acurast

27,001 次观看 • 10 个月前

Citrini says the real preclinical bottleneck is monkeys, which take four years to breed and are harder to scale than memory supply “There’s a huge shortage of monkeys and the the you can’t really, that’s probably the most important thing for pre-clinical testing." "And I know a lot of people are going to be like morally opposed to this, but I’ll just say one thing, which is for the, you know, the medicine is closer than ever before to a world where it really is, I say closer than ever before, that doesn’t mean this is going to happen tomorrow or within the next two years or what, but closer to a world where in silica or on on a computer you can simulate, say, like toxicology studies, right, and and reduce the need for animal testing." "I know that sounds counterintuitive because it’s, okay, well, why are you like investing in this company that that does animal testing? Well, the reason for that is in order to get closer to that world, what you need is more data, right? Like you need, you need more experimental data from animal testing, and monkeys are irreplaceable, right?" "Unless you want to do them in humans. If you want to get as close as possible to humans, obviously the answer is monkeys. And it’s it’s harder to expand the capacity of monkeys than it is to increase the supply of memory, right? Like the the the time it takes, I think it’s four years to to breed these these monkeys.”
1:16

Citrini says the real preclinical bottleneck is monkeys, which take four years to breed and are harder to scale than memory supply “There’s a huge shortage of monkeys and the the you can’t really, that’s probably the most important thing for pre-clinical testing." "And I know a lot of people are going to be like morally opposed to this, but I’ll just say one thing, which is for the, you know, the medicine is closer than ever before to a world where it really is, I say closer than ever before, that doesn’t mean this is going to happen tomorrow or within the next two years or what, but closer to a world where in silica or on on a computer you can simulate, say, like toxicology studies, right, and and reduce the need for animal testing." "I know that sounds counterintuitive because it’s, okay, well, why are you like investing in this company that that does animal testing? Well, the reason for that is in order to get closer to that world, what you need is more data, right? Like you need, you need more experimental data from animal testing, and monkeys are irreplaceable, right?" "Unless you want to do them in humans. If you want to get as close as possible to humans, obviously the answer is monkeys. And it’s it’s harder to expand the capacity of monkeys than it is to increase the supply of memory, right? Like the the the time it takes, I think it’s four years to to breed these these monkeys.”

Fireside Alpha

43,715 次观看 • 5 天前

Excited to introduce you to Seb, the coding agent for hardware. Seb is trained for low level systems in highly-regulated environments. Thank you to the engineers and companies that already trust Seb with their hardware. The feedback from hundreds of Firmware, Embedded, and Hardware Engineers has been invaluable. If you are building new hardware or updating a legacy system, Seb is here to help. Get started today at And as a thank you from the Root Access team -- the first 100 engineers to use promo code RAGE50 get 50% off their first month of Pro!
0:57

Excited to introduce you to Seb, the coding agent for hardware. Seb is trained for low level systems in highly-regulated environments. Thank you to the engineers and companies that already trust Seb with their hardware. The feedback from hundreds of Firmware, Embedded, and Hardware Engineers has been invaluable. If you are building new hardware or updating a legacy system, Seb is here to help. Get started today at And as a thank you from the Root Access team -- the first 100 engineers to use promo code RAGE50 get 50% off their first month of Pro!

Ryan Eppley

55,202 次观看 • 5 个月前

Ben Thompson says Nvidia is already cutting prices. The cut just doesn't show up in their margins: "Everyone's coming for them and they're still charging however much money for a chip. But they're not maintaining their margins, because who's buying? This circular financing people talk about, Nvidia's providing a 25% backstop. If you actually ascribe a value to Nvidia taking equity in the Neoclouds, they guarantee they're going to buy all their compute through 2030. And why do they do that? So the entity in question can get a lower cost of capital, so they can buy more GPUs. They get a lower cost of capital because Nvidia assumed the risk. Taking risk has a price. There's also a world where a ton of compute comes to market, the hyperscalers don't need this compute, and Nvidia's paying for compute that no one wants. That is a diminution of Nvidia's profitability if you look at their business holistically. What that is, is a price cut. The price just didn't show up in the margins."
1:29

Ben Thompson says Nvidia is already cutting prices. The cut just doesn't show up in their margins: "Everyone's coming for them and they're still charging however much money for a chip. But they're not maintaining their margins, because who's buying? This circular financing people talk about, Nvidia's providing a 25% backstop. If you actually ascribe a value to Nvidia taking equity in the Neoclouds, they guarantee they're going to buy all their compute through 2030. And why do they do that? So the entity in question can get a lower cost of capital, so they can buy more GPUs. They get a lower cost of capital because Nvidia assumed the risk. Taking risk has a price. There's also a world where a ton of compute comes to market, the hyperscalers don't need this compute, and Nvidia's paying for compute that no one wants. That is a diminution of Nvidia's profitability if you look at their business holistically. What that is, is a price cut. The price just didn't show up in the margins."

Patrick OShaughnessy

160,252 次观看 • 2 个月前

Noah spends 40% of his time on compute It's fascinating to hear how he thinks about securing it when Instinct is growing 10% a day "What does it mean when the amount of compute that you need access to is now doubling effectively every week? Do we buy 2x of what we have right now? Well, we're gonna consume that in a week. So then do you buy 5x? We're gonna consume that in less than three weeks. Do you buy 10x? If you can even stomach what it's like to buy 10x ahead, you're gonna consume that a couple weeks from now. It's not like a business where when you double the number of users, you can buy 2x more resources to power it. The resource has a lead time of several months. You can't just go out tomorrow and buy compute, because honestly, you get taxed like 3 or 4x. Let's say it slows down to 5 to 8%. But 5 to 8% compounding day over day for three, four months, which is the lead time to bring compute online, that's 100 million users. So then do you buy compute for 100 million users? If you're wrong, you're very wrong. You get charged 3-4x."
1:27

Noah spends 40% of his time on compute It's fascinating to hear how he thinks about securing it when Instinct is growing 10% a day "What does it mean when the amount of compute that you need access to is now doubling effectively every week? Do we buy 2x of what we have right now? Well, we're gonna consume that in a week. So then do you buy 5x? We're gonna consume that in less than three weeks. Do you buy 10x? If you can even stomach what it's like to buy 10x ahead, you're gonna consume that a couple weeks from now. It's not like a business where when you double the number of users, you can buy 2x more resources to power it. The resource has a lead time of several months. You can't just go out tomorrow and buy compute, because honestly, you get taxed like 3 or 4x. Let's say it slows down to 5 to 8%. But 5 to 8% compounding day over day for three, four months, which is the lead time to bring compute online, that's 100 million users. So then do you buy compute for 100 million users? If you're wrong, you're very wrong. You get charged 3-4x."

Patrick OShaughnessy

103,942 次观看 • 1 个月前

Elon Musk: Now we’re going to look at the computer for the Cybercab. The computer will be somewhat overspecified, but I think there’s a major opportunity here. Similar to Amazon Web Services, if the car is driving for 50 hours a week, that leaves more than 100 hours where the compute is sitting there unused. You could use that idle compute for distributed inference. If you have a fleet of 100 million vehicles, each with roughly a kilowatt of efficient inference compute, that’s about 100 GW of compute. That’s a massive amount of compute. And if it’s already there, you might as well use it. I think that makes a lot of sense. The autonomous future is here.
0:53

Elon Musk: Now we’re going to look at the computer for the Cybercab. The computer will be somewhat overspecified, but I think there’s a major opportunity here. Similar to Amazon Web Services, if the car is driving for 50 hours a week, that leaves more than 100 hours where the compute is sitting there unused. You could use that idle compute for distributed inference. If you have a fleet of 100 million vehicles, each with roughly a kilowatt of efficient inference compute, that’s about 100 GW of compute. That’s a massive amount of compute. And if it’s already there, you might as well use it. I think that makes a lot of sense. The autonomous future is here.

Tesla Owners Silicon Valley

60,009 次观看 • 1 个月前

WAIT… I can’t believe Sam Altman really said this out loud about local AI: “A lot of people want local inference. And they assume that the problem is solved now because we are going to have open source models. And so we're not going to need all the stuff in the data center. But from a global allocation perspective, actually, it's quite terrible to have everybody have a bunch of the world's limited supply of DRAM sitting in their own computer. And, being used 5% of the time.” this is such a bizarre thing to say in the open. I don’t get why people buying hardware for their own use is framed as a competition with data centers spending billions to buy hardware at scale. if anyone should be complaining about how the world’s memory is allocated, it must be the average consumers facing the skyrocketing hardware prices. it just doesn’t make any sense.
0:44

WAIT… I can’t believe Sam Altman really said this out loud about local AI: “A lot of people want local inference. And they assume that the problem is solved now because we are going to have open source models. And so we're not going to need all the stuff in the data center. But from a global allocation perspective, actually, it's quite terrible to have everybody have a bunch of the world's limited supply of DRAM sitting in their own computer. And, being used 5% of the time.” this is such a bizarre thing to say in the open. I don’t get why people buying hardware for their own use is framed as a competition with data centers spending billions to buy hardware at scale. if anyone should be complaining about how the world’s memory is allocated, it must be the average consumers facing the skyrocketing hardware prices. it just doesn’t make any sense.

ℏεsam

271,641 次观看 • 1 天前

Mark Zuckerberg says he is not an AI researcher -- so here is the one number he obsesses over to win the superintelligence race: "I'm not an AI researcher. So it's not like I'm in there telling them what research ideas to do." "The things that I can do as CEO of the company are: I can make sure that we have the very best people and talent density. I can make sure that we have by far the highest compute per researcher, and that we do whatever it takes to go build out all that capacity." "We're building this Prometheus cluster, which is going to be the first gigawatt-plus single cluster, a contiguous cluster for training, coming online next year." "We're building a cluster in Louisiana that is going to be 5 gigawatt, and we're building several more clusters like that that are just going to be multiple gigawatts." "We're talking about many hundreds of billions of dollars of capital. So you both need to have a good business model that can support it, and you need to believe in it." The tell: the richest lab in the world is not optimizing headcount. It is optimizing compute-per-researcher, and betting hundreds of billions that whoever owns the biggest contiguous cluster owns the frontier. Talent gets you into the room. Watts decide who leaves with the model.
1:06

Mark Zuckerberg says he is not an AI researcher -- so here is the one number he obsesses over to win the superintelligence race: "I'm not an AI researcher. So it's not like I'm in there telling them what research ideas to do." "The things that I can do as CEO of the company are: I can make sure that we have the very best people and talent density. I can make sure that we have by far the highest compute per researcher, and that we do whatever it takes to go build out all that capacity." "We're building this Prometheus cluster, which is going to be the first gigawatt-plus single cluster, a contiguous cluster for training, coming online next year." "We're building a cluster in Louisiana that is going to be 5 gigawatt, and we're building several more clusters like that that are just going to be multiple gigawatts." "We're talking about many hundreds of billions of dollars of capital. So you both need to have a good business model that can support it, and you need to believe in it." The tell: the richest lab in the world is not optimizing headcount. It is optimizing compute-per-researcher, and betting hundreds of billions that whoever owns the biggest contiguous cluster owns the frontier. Talent gets you into the room. Watts decide who leaves with the model.

Karl Mehta

255,725 次观看 • 3 个月前

When I sold my software company in 2010 and became an angel investor, one word could smother a pitch on the spot: “Hardware”. A hardware component made you close to unfundable. The VCs called you a hybrid company, and it wasn't a compliment. The refrain was: “Hardware is hard, hardware is slow, hardware eats cash and stretches timelines. We want pure software”. That was the era. It took two shocks to break the spell. First, COVID broke supply chains, and soon everyone in Washington was doing the math on where things actually get made. For example, the US hasn’t produced tungsten domestically since 2015 . (Aside: this is changing, cc Taylor Sulik ⛏️ ) Rare earth magnets: almost all bought from countries that may or may not be on our side. The ability to make critical materials at home was a national capability that we’d spent decades letting rot. That lit a fire under a lot of folks. Then AI showed up and knocked the legs out from under pure software entirely. The nightmare scenario for a lot of SaaS companies is now the next model dropping and a customer rebuilding your product in-house. So the same VCs who said "we don't want any hardware in this" are now asking founders where the hardware is. Which brings me to moats. A moat has always been a rebuild-time question: how long would it take a well-funded competitor to rebuild what you have? Software used to buy you years. AI is shrinking that number toward months, and a moat measured in months is more like a wooden fence. The physical world doesn't compress that way, since time is baked into its advantages. A factory takes three to five years to stand up regardless of how smart your tools get. Then permits, construction, supply chains, testing data… Another part is that AI trains on the written-down world, and high-moat physical knowledge lives largely offline… public AI models have not yet ingested it, and no prompt can reproduce the underlying empirical record. Anyway, hardware is still damn hard. And that's exactly the point.
1:02

When I sold my software company in 2010 and became an angel investor, one word could smother a pitch on the spot: “Hardware”. A hardware component made you close to unfundable. The VCs called you a hybrid company, and it wasn't a compliment. The refrain was: “Hardware is hard, hardware is slow, hardware eats cash and stretches timelines. We want pure software”. That was the era. It took two shocks to break the spell. First, COVID broke supply chains, and soon everyone in Washington was doing the math on where things actually get made. For example, the US hasn’t produced tungsten domestically since 2015 . (Aside: this is changing, cc Taylor Sulik ⛏️ ) Rare earth magnets: almost all bought from countries that may or may not be on our side. The ability to make critical materials at home was a national capability that we’d spent decades letting rot. That lit a fire under a lot of folks. Then AI showed up and knocked the legs out from under pure software entirely. The nightmare scenario for a lot of SaaS companies is now the next model dropping and a customer rebuilding your product in-house. So the same VCs who said "we don't want any hardware in this" are now asking founders where the hardware is. Which brings me to moats. A moat has always been a rebuild-time question: how long would it take a well-funded competitor to rebuild what you have? Software used to buy you years. AI is shrinking that number toward months, and a moat measured in months is more like a wooden fence. The physical world doesn't compress that way, since time is baked into its advantages. A factory takes three to five years to stand up regardless of how smart your tools get. Then permits, construction, supply chains, testing data… Another part is that AI trains on the written-down world, and high-moat physical knowledge lives largely offline… public AI models have not yet ingested it, and no prompt can reproduce the underlying empirical record. Anyway, hardware is still damn hard. And that's exactly the point.

Adam Rossi

11,932 次观看 • 3 个月前

🎥WATCH: NVIDIA CEO EXPLAINS WHY AI NEEDS 1,000X MORE COMPUTE "The amount of computation necessary for generative AI is like thousand times more." "Then you multiply that by the number of people who now want to use it by a hundred times. Which is the reason why GPU consumption is going through the roof."
0:58

🎥WATCH: NVIDIA CEO EXPLAINS WHY AI NEEDS 1,000X MORE COMPUTE "The amount of computation necessary for generative AI is like thousand times more." "Then you multiply that by the number of people who now want to use it by a hundred times. Which is the reason why GPU consumption is going through the roof."

Coin Bureau

50,688 次观看 • 3 个月前

SF Compute CEO evan conrad on his contrarian thesis that OpenAI and Anthropic won't eat the world, everyone will become OpenAI and Anthropic instead: "Our impression is not that OpenAI and Anthropic will eat the world, it's that everyone will look like OpenAI and Anthropic. That's simply the way that companies work now." "Every single company in the world that is of meaningful scale will look like a frontier lab. There is this belief that the large labs have won it all, and they've got this compounding advantage. The default assumption of that is assuming nobody else will do anything." "The bottleneck was there's 100 people, and they know how to train the models. Or you thought, the problem is compute. The problem of compute is credit. And OpenAI and Anthropic are not unique to having credit. In fact, they don't have as good credit as other people." "If you assume the problem is not the researchers, that that problem's basically gone away the moment you got to auto research, then what you should assume is that pretty much every other company is going to swap the way they work to look like a frontier lab. It won't be OpenAI and Anthropic take the whole world. It's that everyone's OpenAI and Anthropic now." San Francisco Compute
1:32

SF Compute CEO evan conrad on his contrarian thesis that OpenAI and Anthropic won't eat the world, everyone will become OpenAI and Anthropic instead: "Our impression is not that OpenAI and Anthropic will eat the world, it's that everyone will look like OpenAI and Anthropic. That's simply the way that companies work now." "Every single company in the world that is of meaningful scale will look like a frontier lab. There is this belief that the large labs have won it all, and they've got this compounding advantage. The default assumption of that is assuming nobody else will do anything." "The bottleneck was there's 100 people, and they know how to train the models. Or you thought, the problem is compute. The problem of compute is credit. And OpenAI and Anthropic are not unique to having credit. In fact, they don't have as good credit as other people." "If you assume the problem is not the researchers, that that problem's basically gone away the moment you got to auto research, then what you should assume is that pretty much every other company is going to swap the way they work to look like a frontier lab. It won't be OpenAI and Anthropic take the whole world. It's that everyone's OpenAI and Anthropic now." San Francisco Compute

MTS

135,653 次观看 • 2 个月前

Adaption Labs CEO Sara Hooker says data centers in space are "pretty bonkers", and explains what the current consensus gets wrong: "Most co-located hardware is pretty much for training. I think that's why you care [about Space], otherwise you can distribute. Inference compute, which is where everything's moving towards, you can spread that compute more easily. You can have multiple data centers [on Earth]." "So if you care about space, you probably only care about training compute. I think people underestimate the amount of failures that happen, and you don't want to get your training job interrupted." "The real issue is that GPUs still have failure rates. The 2% of GPUs that are just considered done every year, that's really your cost. It's how quickly you can replace those [in space] and what that looks like."
2:29

Adaption Labs CEO Sara Hooker says data centers in space are "pretty bonkers", and explains what the current consensus gets wrong: "Most co-located hardware is pretty much for training. I think that's why you care [about Space], otherwise you can distribute. Inference compute, which is where everything's moving towards, you can spread that compute more easily. You can have multiple data centers [on Earth]." "So if you care about space, you probably only care about training compute. I think people underestimate the amount of failures that happen, and you don't want to get your training job interrupted." "The real issue is that GPUs still have failure rates. The 2% of GPUs that are just considered done every year, that's really your cost. It's how quickly you can replace those [in space] and what that looks like."

TBPN

37,463 次观看 • 8 个月前

"Hardware is hard." - Elon Musk Thrive Capital Partner Philip Clark on why investing in hardware has never been more attractive: "Hardware businesses can be enormously durable once you get them off the ground." "It's really hard to imagine who is possibly going to disrupt SpaceX at this point." "It's harder because there is often an operational aspect to hardware investing, where I don't just have to build a product and press a 'ship the binary' button on a computer." "I actually have to have a manufacturing line and a deployment line that gets this out into the real world." " When you get over the hump of these really, really hard industries and problems—you actually have built something at scale that is very hard to quickly replicate."
1:33

"Hardware is hard." - Elon Musk Thrive Capital Partner Philip Clark on why investing in hardware has never been more attractive: "Hardware businesses can be enormously durable once you get them off the ground." "It's really hard to imagine who is possibly going to disrupt SpaceX at this point." "It's harder because there is often an operational aspect to hardware investing, where I don't just have to build a product and press a 'ship the binary' button on a computer." "I actually have to have a manufacturing line and a deployment line that gets this out into the real world." " When you get over the hump of these really, really hard industries and problems—you actually have built something at scale that is very hard to quickly replicate."

sourcery

11,398 次观看 • 4 个月前

.Peter Steinberger 🦞 says Clawdbot represents a "liberation of data that big tech probably doesn't really want" because it effectively neutralizes its walled gardens. "The demand for the old models has to change. When you are a company, and you want to access Gmail, the amount of red tape is so large that startups have to buy other startups that have the license for Gmail." "But if you run it locally, you work around all of that. I've built plenty of CLIs where I literally pointed Codex at the website and said, 'Build me a CLI.'" "It's the liberation of data that big tech probably doesn't really want. Even the WhatsApp integration is a hack."
3:59

.Peter Steinberger 🦞 says Clawdbot represents a "liberation of data that big tech probably doesn't really want" because it effectively neutralizes its walled gardens. "The demand for the old models has to change. When you are a company, and you want to access Gmail, the amount of red tape is so large that startups have to buy other startups that have the license for Gmail." "But if you run it locally, you work around all of that. I've built plenty of CLIs where I literally pointed Codex at the website and said, 'Build me a CLI.'" "It's the liberation of data that big tech probably doesn't really want. Even the WhatsApp integration is a hack."

TBPN

21,117 次观看 • 9 个月前