The Wisdom Wall

“Once you invest in something, you just start to adopt the, uh, the mouthpiece for it and make a lot of arguments that are central to that company's success. And I think that's happening now that we have a large number of VCs in the military space.”

“If you're gonna sell something in a 120 countries around the world and try and tell one country they can't have it. There's no, there's no effing way that's gonna work. Like, and especially these goods have digital attributes.”

“protecting the U.S. by trying to somehow injure or, or disable or hinder a competitor that's in a different country in the long run will make us weaker. I, I believe that fundamentally.”

“And so now there's a path where investors, GPs get rich, where the outcome of the founders and their companies is irrelevant. Not to say they are aligned in opposite directions, but they're no longer aligned. And now, as a rational actor, you could say, why not maximize the guaranteed portion of the comp, the two…”

“The part that is, the part that is overlooked is when companies become more productive, Using artificial intelligence. It is likely that it manifests itself into either better earnings, or better growth, or both. And when that happens, the next email from the CEO is likely not a layoff announcement.”

“What's weird to me is that the government By trying to protect small companies and not allowing big companies to buy small companies. So big companies get bigger. I think it's got the reverse effect where now small companies have one less chance of winning. And thus the public market is even less willing to hold a…”

“If you're a founder and you want to compete in this, you know, in the model game, you know, and you don't have their backing, there is no, no chance.”

“the math you get to before you account for things like, um, bad GPUs, bad satellites, right, these will all be things that happen, but the math you get to is it's about five billion dollars per gigawatt Of CapEx to put these in space. Right. For comparison, terrestrially, talk about the switch gears, the generators,…”

“if I was starting a company today, I, I, I'm really excited about these companies that are going into companies and getting extremely hands-on and doing effect effectively professional services, um, with AI because we've saturated all the evals, um, and you need to get proximate to the problems.”

“the thing that I, I feel, uh, you know, is happening in enterprises, you, you hear these 95% of projects fail, But, like, you know, like, that's, that's, that's actually what you want. Like, you, you, like, when you are actually experimenting with new technology, if, if, if all of your projects are failing, that means…”

“The biggest problem is that you bake a model, and that's where it's learned everything it needs to learn, and then you freeze it, and then you launch it, and then maybe you give it some context, but that's it. It's frozen. So therein lies the problem, uh, that, you know, we need, we need an AI that really can sort of…”

“Unfortunately, it's something closer to, uh, log of intelligence equals log of compute, but we may figure out better scaling laws, and we may figure out how to beat this.”

“everything is a commodity, right? Compute, storage. I remember everybody saying, wow, how can there be a margin? Except at scale, nothing is a commodity.”

“And so I think the next generation of SaaS applications will have to sort of, if you are high ARPU, low usage, then you have a little bit of a problem. But if you are, we are the exact opposite. We are low ARPU, high usage.”

“What GitHub did in the first, I don't know, 15 years of its existence or 10 years of its existence, it was basically done in the last year, just because coding is no longer a tool. It's more a substitute for wages somewhere.”

“whereas in the enterprise, one is it's not winner take all. And two, it is going to be a lot more friendly for agentic interaction. So it's not like, for example, the per seat versus consumption. The reality is agents are the new seats.”

“And so thinking, post-training, pre-training, we now have three scaling laws, not one.”

“wafer costs are getting, wafer costs are getting higher, which means that unless you do co-design at an extreme scale, you're just not going to be able to deliver the x-factor growth, number one, number, and so you, you know, unless you, unless you're working on six, seven, eight chips a year.”

“So even if they gave it to you for free, you, you, you only have two gigawatts to work with, your opportunity cost is so insanely high, you would always choose the best perf per watt.”

“I do think you end up with a higher fitness level for a community that's behaving that way. Overall, you may end up with a lot less chance of a breakout monopolist, like we've had in many of the sectors in American technology.”

“Amazon had this monopoly retail business they could use to subsidize AWS, gain share for a decade, and then begin to take price. That would be a rational strategy for OpenAI to follow. So you take the profitable consumer business, you use it to subsidize, you know, the, the, the market sharing other applications that…”

“What that tells me is the model layers being increasingly commoditized. And that there's not going to be a lot of intrinsic value in, in, in that intelligence layer, that operating layer. You're going to have to build applications that guys like Bill Gurley and, and you and I are using every day. And that's where the…”

“I guess, by the way, those apps had inherent virality, as you know. I mean, you're kind of the expert of that. This doesn't. This has no virality to it.”

“I like to say that tokens trump tariffs.”

“if productivity for the next decade or so was about two and a half to three and a half percent per year, we could achieve substantial reductions in this key ratio Of debt to GDP.”

“One of the big thing, problems I think they have is having, you survive to this point, and having succeeded, let's say they have revenue of 50 to a hundred million dollars, they feel like they need to protect something, and it puts them on the back foot, not the front foot. It makes them conservative, and I like your…”

“there does seem to be in China, and I, and part three of this research walks through this, a deprioritization of market cap of successful companies.”

“And so by, by denying your exports to a market, You increase the incentive they have to develop their own technology.”

“anyone that tells you when you're ready to go public that you have to wait on the markets to be in a particular place. I would tell them to shut the fuck up and like, like, like take your company out. Like you can't control that thing. That's an external factor.”

“Repeat entrepreneurs in the enterprise space are a golden ticket. Like, like it is so, there are so many learned experiences about go, mostly I think about go to market that are repeatable, that you can line them up and do it again.”

“If those second layer and the third layer are both negative gross margin, the consumer's buying compute from the hyperscaler at a price that's lower than they would if they bought it directly, um, because it's being subsidized by two players in the middle. And on top of that, if this is true, you're triple counting…”

“Out of the thirteen billion dollars of capex required, you know, if you make an assumption about 25 75 or 30 70 equity to debt, that probably requires like a three billion dollar equity check.”

“And so I would always ask my analysts when they would bring me a power company, I would say, why is that better than NVIDIA? Right? It's a derivative, it's a derivative NVIDIA. It's the exact same bet. So why not just buy more NVIDIA?”

“Like you can have the best computers and chips and biggest cluster in the world. But if you don't get the architecture right around your reasoning model and the other guy does, they're probably going to win.”

“NVIDIA's moat in, in inference is actually A lot smaller on software, um, but it's a lot bigger on, hey, they just have the best hardware.”

“You can create data out of thin air almost, right? Um, in certain domains, right? And so this is the whole, the debate around scaling laws is, um, how can we create data?”

“why is Mark Zuckerberg building a two gigawatt data center in Louisiana? Why is, why is Amazon building these multi gigawatt data centers? Why is Google, why is Microsoft building multiple gigawatt data centers? Plus buying billions and billions of dollars of fiber to connect them together because they think, Hey, I…”

“So, the, the whole paradigm of training, you know, pre-training is, is, is not slowing down. It's just, it's logarithmically more expensive each, for each generation, for each incremental improvement.”

“When I do this with O-one, right, because it's doing that thinking phase of 10,000... it spends a lot of memory on generating this KV cache and reading this KV cache constantly. Now the maximum batch size, i.e. concurrent users I can have, is a fraction of that. One-fourth to one-fifth the number of users can currently…”

“It's just impossible. It's kind of like piracy, right? I mean, you can sort of all kinds of terms of use, but it's impossible to control distillation.”

“the network effects are in the app layer. So why would I want to spend a lot on some model capability with the network effects are all on the app layer?”

“Now, the reason I hate these things is they create misalignment. So you have a bunch of investors on a cap chart and now they have, um, differing incentives. The, the later stage investors that wrote this, uh, compounding ratchet term I actually want the IPO to be as low as possible because they'll, they'll get more…”

“if you, if you want it sent moving to electric and getting rid of the big gas guzzlers, you know, tax what's negative rather than trying to pay people for what you think they want. It just creates more perverse behavior.”

“every founder and every company and every board will tell you if we raise money, we're not going to spend it. We're going to stay frugal. That never happens. And instead of focusing on, you might ask a founder, what are the three things that matter most this year that are critical for your success? If you only focused…”

“If you're aiming for traditional venture like returns, then I don't think there are a hundred growth companies, right? That you can go put in a fund of five or ten billion dollars equally weight them and get a four to five X over any reasonable period of time.”

“their revenue per visit is going to fall if they complete the transaction. Because someone is spending marketing dollars to take that customer to their website and run the transaction with the hopes that they'll come directly back to that website. So they'll spend 50 to a hundred percent of first transaction. There's…”

“the real, The real answer is the companies that don't deploy these things are going to go out of business. And so I think margins get competed away in many, many cases. I think it's ridiculous to imagine, oh, every company goes to 60%.”

“What people don't realize is that you can have, ah, three different ISAs, CPU ISAs, they all have their own C compilers, you could take, you could take software and compile down to that ISA. That's not possible in accelerated computing. That's not possible in parallel computing. The company who comes up with the…”

“Somebody who's really, really good at building GPUs have no clue how to be an accelerated computing company.”

“If I have an AI partner where I can simply say, who, who did I send that email to? Like, that's really, really powerful. And I think the switching costs are insurmountable if someone gets to that place first.”

“And if you're going to be founder friendly and write big checks, guess what? You're going to be supportive of founder secondary. And you're going to be in for, you're going to be supportive of broad based employee secondary. When you do those things, you are taking away probably the strongest motivating factor that…”

“You can't go raise another five hundred million and not have pressure from all your employees for employee secondary For spending more money on more projects, et cetera, and the NPV on those other activities will be lower, and the incentive your employees have to stay with you once they sell 10 or twenty million…”

“power is the primitive for AI, and that's why China is heads down in this race, and they're taking our technology, right, and they're, you know, they're improving it”

“I hear people say, um, oh my God, SAS is super cheap at six times revenue. And I'm like, you know, I've worked, I've seen industries mature where, where six ain't a stopping point, you know? Um, and of course, you got to move on to, to, to net income and, and earnings and cashflow and, and get out of this revenue…”

“I've often said that I, that the, the federal government should Have a new test for a monopoly, and I, and this, this is very self-centered bias to someone who's been a venture capitalist working with small companies for a long time, but for over the course of the years, there have been many cases where, um, I've been…”

“Two, as I understand it, there is a substantial double taxation risk, and if this is not the case, it's the thing I'd love for someone to educate me on, but the way the money makes it from the Mag-Seven to the investors, I think is two-step. One, the money is transferred to the startup, As part of this massive license…”

“If you're a closed model company, the question first is, okay, I'm going to have to To compete and keep up with this frontier level competition. That's hard enough. It takes a lot of resources, but number two, Mark's going to give it away. You know, meta is going to give it away for free. Now you have to develop a…”

“having crazy money in the market is not necessarily consistent with creating positive return for your venture capital portfolio. It, it, it can start to trend the other way. If you end up with hyper competition, if you end up where every single player has four hundred million dollars, which I think is true in the, um,…”

“The, the problem I have with the first one, um, is simply that, you know, my entire time in, in technology, like, these tools are competitive weapons. But if I get armed with them and my competitor gets armed with them, it doesn't necessarily increase. It doesn't create like free net income. You might die if you don't…”

“I, I often hear people on a board say something like, well, good thing we, we didn't go public, a good thing we aren't public. But, but they're sitting on the board of a private company that's at a hundred and fifty million in revenue and growing 10%. Like, you're, you just have your head in, in a hole in the ground…”

“when the public markets gives you a sign, right, like it did Meta or Facebook in the fall of twenty-twenty-two when the stock goes to 90 dollars a share, right, there's nowhere, there's nowhere to hide. You confront the brutal truth. You confront the reality. Public market investors You know, it's the collective wisdom…”

“there's two criteria as one is, can you get to a billion of revenue? Open AI would be check, check, check the box. But the second one is, do you have a sort of a break even or a path to break even?”

“you're going to have higher burn rates. And so you're going to know less about your unit economics, just by a natural fact, because you get further away when you're operating that way. And you can't raise four hundred million dollars and not have a high burn rate. Like, there's no point in it.”

“the value is not in the model. Right. Just like the value is not in storage. Right. You could say storage is a part of the AWS cloud, but that there's not a lot of value in that thing unto itself. The value is in the enterprise relationship. The value is in right. The, the, the, uh, the, the pen number of services that…”

“when I meet a company and see them using AI in a way that feels like ultra compelling from us improvement of their own strategic business position, it's almost always a more traditional AI model that's running a very particular optimization problem. It's not an LLM application.”

“I, I have always believed, and I'm not the only one there, there's, there's a assortment of other people in our industry, that being public is great for companies. It raises the bar in terms of their performance. It causes them to, to, uh, to, I think, achieve more than they would otherwise.”

“In fact, like because of a lack of liquidity premium, you're probably worth less than that public company and structurally, Your cap chart's more rigid, and so you have less flexibility. You can't do any acquisition. I mean, there's all kind of reasons why that's a worse place to be”

“Are you going to build the fleet to average or peak? And by the way, you lose whatever your answer is. If you build it to average, you're not going to be able to serve your customers during the peak at all. Like they're only going to be disappointed. And if you build it to peak, you're going to have a bunch of very,…”

“a lot of the incremental work that's being done on AI projects are Stitching together external databases with the LLM and the LLM gets relegated to being an interpreter of the human or to being the interpreter of the data back to the human, but is not involved necessarily in the data store and is partially involved,…”

“Apart from the people who are on the very frontier, if you're on the frontier and you have something totally different, it seems to me that that's a place where that is defensible. But if you're not on the frontier, man, it seems that these are going to be really fast depreciating assets that are going to be…”

“here's where we get into questions of alignment with shareholders, because you could theoretically, um, have a management team that leads a company, you know, over a year to a 20% decline and pick up 80% of the value of their equity award. Whereas shareholders are down 20. You, as a large public owner of shares, you'd…”

“And for a tech company, oftentimes stock-based compensation happens to be the single largest expense component of, of a tech company. And yet somehow inexplicably, even though it's the largest single expense component of that company, uh, it is not considered to be even an expense in some companies and in some…”

“when you don't, when you, when you proforma something out, right, when you stop treating it like cash, right, it leads to a misallocation of resources because you're effectively saying this doesn't exist, and when you, you know, what, what happens when the cost of something disappears? You get a lot more of it.”

“if you dilute free cash flow by 30 to 35% every single year for 20 years, guess what? Your dilution is 35%, you know. It's not a half of, it's not 50 basis points, it's not one percent, it's not two percent, it's literally, you know, a third of the company.”

“The switching cost is not with the LLM. I agree with you on this. The switching cost is really around where your data resides, right? We've heard this talked about data gravity.”

“You're not going to get to the same revenue per visit with a, with a transaction model that you do with the ad model, because you already mentioned Cactale TV, people will pay using marketing math 4050, 60% of first purchase. With a transactional integration, they want to pay five percent. So you have a 10 x reduction…”

“So like we're, we're increasingly not limited by the architecture of the technology. And now we're just limited by like, can Jensen make these things cheaper? Um, can, can the, you know, can the engineers at open AI have more efficient model algorithms?”

“I do think this memory issue is a big, big deal, at least on the consumer side. So none of the major contenders today can remember who you are because it would require them. And I mean, remember over five years, 10 years to really become a personal assistant kind of thing that, that her represented. And it's because…”

“the largest companies in our world have some of the highest growth rates, and I think that's unprecedented.”

“One of the easy defaults you go to in the middle is, oh, I'll just go to my insiders and ask for a bridge. And I will tell you, at least all the data I've seen, the success stories coming off a bridge are, are few to none. That's why we always refer to them as piers rather than bridges.”

“It's actually really bullish for compute and hardware, because if the frontier models are capturing less of the margin, then you're gonna spend more on compute. So the better open source does, the better it is for compute providers.”

“I won't be surprised if you see this happen for other forms of sort of quantitative knowledge work, just because it happens to have the properties that code has. It's testable. You know if it worked or not. Um, it's very RL friendly.”

“chat is a great way of expressing your intent. It's a good way of communicating with the machine, but it's not a great output. Um, where in many cases, what you want back is an artifact.”

“If you were to pre-mortem why a company like OpenAI does not achieve its mission, it's probably focus because of the sheer number of opportunities that become possible when you approach AGI, right?”

“I think writing, actually, um, is very important, and it's not because the AI can't write. You know, AI will become amazing at writing just like any other domains, but because I think the skill of writing forces you to be very clear of what you have to say. And even though prompt engineering is obviously going to go…”

“when I look at sort of in the cloud infrastructure business, one of the key things you have to do is have two things. One is an efficient, like in this context, in a very efficient token factory, and then high utilization. That's, that's it. There are two simple things that you need to achieve. And in order to have…”

“When you have the biggest workload there is running on your cloud, that means not only are we going to learn faster on what it means to operate with scale, that means your cost structure is going to come down faster than anything else. And guess what? That'll make us price competitive.”

“And, uh, and I think circularity ultimately will be tested by demand. Because all this will work as long as there is demand for the final output of it. And up to now, that has been the case.”

“overall, it is going to be true that there is a real marginal cost to software this time around. It was there in the cloud era too. When we were doing, you know, CD-ROMs, there wasn't much of a marginal cost. You know, with the cloud, there was. And this time around, it's a lot more. And so, therefore, the business…”

“search was pretty magical, uh, in terms of its ad unit, uh, and its cost economics, because there was the index, which was a fixed cost that you could then amortize in a much more efficient way. Uh, whereas this one, you know, each chat, uh, to your point, you have to burn a lot more GPU cycles, uh, both with the…”

“if you look at the history of the picks like alternatives, I mentioned in the UK and China and India, the startups that do financial innovation scale up way more aggressively and successfully on those rails that are cheaper and faster.”

“You're supposed to create a startup before the market grows. You're not supposed to come up as a startup when the market's a trillion dollars large. You know, this fallacy, and all VCs know this, this fallacy that a large market, if you could just take a few percent market share, you could be a giant company. That's…”

“my hunch is some of the, um, Uh, enterprise deployments that don't actually work out likely don't have the scaffolding or infrastructure for these agents to interact with as well. Um, a lot of the, like, really successful deployments that we've made, a lot of what our FDs end up doing with some of these customers is to…”

“One of the reasons that happens is the provincial leaders compete with each other. So the provinces are very competitive with one another, not in the same way the states are really. Um, one of the reasons this is true is if you run a province and do well, you put yourself in really good standings to move up in the…”

“if we have this view, the only reason China's competitive or winning is because they're, they're stealing or they're subsidized. I think what that view does is it allows us to delude ourselves into believing we don't need to get better ourselves.”

“You just have a much more natural environment to have this kind of hyper competition that we talked about in EVs or solar, like having that many open source providers gives you that. I would say even maybe more because of the way the models can help one another, at least in the, EV case. Like, you can't take someone…”

“the way to beat them is not to, you know, try to cut them off at the knees. We don't need to make it easy on them. But the United States needs to accelerate our race, and I think by, if we focus too much on how do we slow down China, we take the eye off the ball on how to accelerate our own race”

“if you're not, if you, if you're not confident, you're going to win on offense, you want to play defense. And so for any large tech company, commoditizing a, a potential threat Is, is actually quite valuable.”

“I think that makes it harder if you're in the, the lab game. And you don't have a consumer product and you don't own an application that you can drive high gross margins.”

“I've sort of stepped back from the hysterics and you say, we don't have a loan to value problem as a country. Uh, we, we, we have a spending problem.”

“The model that you use today is the worst AI model that you'll ever use for the rest of your life”

“if the model can kind of do it, then it's going to be great at it in a few months.”

“in an AI world, you need to do that because the, the IMLs are so new. We don't, you don't, you don't top down know all of the capabilities. You see them kind of coming through the mist and we're better at finding all of the opportunities when we have a lot of smart people Thinking about what they can do with the model…”

“Although I would say that we're starting to see network effects, right, on the data side. We're starting to see switching costs with permanent memory, as you and I have talked, you know, starting to see.”

“I just think that these super, super large private companies, if you're not willing to submit yourself to sort of the sunshine and the, and the ray of light of the public markets, you're going to get it through a regulatory agency. So pick your poison and be careful that if you think you can live in the public market…”

“You can't just cut two trillion dollars in a single year. That would be an 800 basis point headwind to GDP. It would throw us into a recession, if not depression, like, like state.”

“So I think what the market was saying is, okay, three hundred billion, half of that will show up as increased prices and taxes on consumers and on businesses. Half of it will get eaten by the, the producer of the product in China, in India, wherever, you know, that's not that big a headwind to the U S economy.”

“tariffs lead to increased inflation. They lead to reduced innovation. Putting a protective blanket over our companies in our country lead to them being less globally competitive. And then fourth, which you mentioned, you know, it leads to retaliation.”

“Um, there's also, if you're using the word resilient, you know, that can include other countries that are allies. It doesn't have to be on the shores of the U.S., and I know people who in the first Trump administration were told to, to diversify away from China, and they built, you know, supply chains in Vietnam, and…”

“The thought that they couldn't, you know, be educated or can't innovate is gonna lead you to a lot of really bad policies.”

“the reflexivity already starts to take place, everything you just described. And, and once it does, some of it is certainly self-reinforcing, so you start decommitting capex, you start decommitting, uh, you know, basic capacity on different manufacturing lines, planes, whatever. Right. And it will become…”

“doing stuff you're good at and letting other people do stuff they're good at is a win, win, win, win, win, and you start backing that up, and you're gonna get lose, lose, lose. I, I'm certain of it.”

“So it is a fast depreciating asset. The second you're off the frontier.”

“building a chip is one thing, but building many chips that connect together, cooling them appropriately, networking them together, making sure that it's reliable at that scale is, is a whole host of problems that semiconductor companies don't have the engineers for.”

“if I just replace, like, six servers with one, I've basically invented power out of thin air, right? I mean, like, you know, in effect, because these old servers, which are six plus years old, or even, you know, they can, they can just be deprecated and put, so with CapEx of new servers, I can replace these old…”

“In fact, there's more inference in training than there is updating the model weights, because you have to generate hundreds of possibilities And then, oh, you only train on a couple of them, right?”

“So I think of AI is the, um, lean for knowledge work.”

“And there is nothing, there is nothing wrong with a down-round IPO. It's simply a price at a moment in time, and if the market in 20 and 21 was overheated, And prices were really high in the private markets. That's just a fact of life.”

“It's clear in these other countries that letting the government do this one part and getting it right leads to all kinds of fintech innovation, because you can transfer money quickly.”

“There are companies that could be sold for a 102 hundred million dollars that results in life-changing outcomes for founders and early employees. If you raised a hundred million dollars too early, that exit path is no longer on the table. You can't sell the business for a hundred million dollars and make any money…”

“valuations represent discounted future expectations. And so, you've, you, you may feel like you've won a prize, but you've really increased everyone's expectation for what this company can achieve. And in order to raise that up around from here, it's way harder than it would have been otherwise.”

“if the cost Of a five percent error or a 10% error is super high, it's going to take AI a long time to get there. I would argue programming is that way. You can't have five or 10%.”

“hypergrowth tends to delay what you learned in microeconomics class. You know, I, I remember when I was a PC analyst and there were five public PC companies all growing a hundred percent. And so in, in moments of hypergrowth, you will have margins that may or may not be durable. Um, and you'll have a number of…”

“people used to think that pre-training was, was hard, and inference was easy. Now everything is hard.”

“there must be a concept of fast thinking, and slow thinking, and, and reasoning, and reflection, and iteration, and simulation, and all that, um, and that now it's coming in.”

“Serial processing requires every transistor to be excellent. Parallel processing requires Lots and lots of transistors to be more cost effective. I'd rather have 10 times more transistors, 20% slower. Than 10 times less transistors, 20% faster.”

“the data center is now the unit of computing. To me, when I think about a computer, I'm not thinking about that chip. I'm thinking about this thing. That's my mental model, and all the software, and all the orchestration, all the machinery that's inside.”

“if the hyperscalers become part of the customer set for the nuclear startups, that may be like 10 X better than selling just to utilities alone. Like you may have brought innovators to the table. On the purchasing side that may be more open-minded, that may be more understanding, may be more willing to share risk,…”

“the reason to make it IP free, Is to get everyone working on the same platform, to remove single source dependency risk, and to get the utilities confidence that this thing's gonna be successful, and gonna go forward, and to get everyone behind it. So do the Linux, you know, of SMRs.”

“there's no way to value a private company. So, so you start with this problem that, that how would you go about saying what it's worth?”

“Something happened over the past, I'd say, Uh, 15 years where some of the smartest companies in Silicon Valley have developed a new strategic play where they use open source as a defensive weapon rather than as an offensive weapon.”

“usually the partner that's involved in the company, the one that's on the board or that led the investment, they're usually pretty, Um, they, they tend to be overly confident. And that's often balanced by the, uh, the other partners back at home. So when you circle around the table and have these discussions, they're…”

“Greedy investors will find you if your company is growing really fast in his great margins. Look at Samsara, ok? Samsara went public, and everybody said, too small, not enough coverage, stock's not working. That, that company has been a grand slam home run in the public markets, ok?”

“when I talk to our entrepreneurs that are using these models, they might design with the cutting edge model, but they all back off to the affordable models on implementation, inference, and runtime. They all do.”

“when you support massive secondaries, you're taking the number one pressure, um, out of the system that used to lead, uh, founders to, to want to go public because their employees are like, I need liquidity, I need liquidity. So you, you, you, you do a release valve, and you take that away”

“We've also talked about the fact that when you move away from coding, the efficacy drops a bit in terms of the productivity gain you might get.”

“if you drive down the price of the cost of compute, then the reflexivity is people will consume a lot more of it. Now, this is also known as the Jevons paradox, right? As price goes, goes down, we demand more of it. The aggregate amount of Consumed, of, of, of the compute consumed actually goes up, right?”

“There's this relative valuation game, which is how bubbles are built. Um, and because you, you, you adjust up to, so you, you re-rate to a new norm, right?”

“there's an argument from Black Shoals that a single RSU is worth about two and a half or three options. So if you could switch From these options to an RSU, this investor thought dilution might drop from, say, four percent to 1.2% or something like that, and there would be, quote, less dilution. And so that change…”

“one thing that drives durability, going back to our competitive advantage period, is switching costs. If I start relying on one of these things as my memory, and, and I don't have a way to pull that out and jump to something else, I'm stuck. Like, I am hook, lock, stuck. Which is very, very positive for the person that…”

“if you're like one click or two clicks removed from Sandhill Road and Sandhill's equally to probably blame at certain points, you're sort of like, oh, tech enabled, everything should all get a six or 10 X revenue multiple when the underlying economics of this business, you know, could be literally a hundred X…”

“I think that was the exactly right economic decision at the time. There was sort of a market share war. Um, you were, you were acquiring customers that maybe had a, a sort of a five or 10 year You know, kind of lifetime value curve, uh, associated with them, maybe even more. Um, and so of course you want to gobble up…”

“Coding is still working better than most other use cases. And it's because you basically have a workflow where you're in a text interface that is just linear, where most of the, the sort of, um, knowledge of that, of the, of the field is all public and open source for the most part and available for training runs. And,…”

“I think the number one thing a founder can do is to as quickly as possible, get in touch with your, with what your real actual valuation is. And then ask yourself, what, what do I need to do structurally to give this company a fighting chance, knowing that that's reality?”

“And when you back into the numbers, you get to something like five megawatts of capacity per Starship launch.”

“It's useful to start with the most zero sum trade off when you do your planning. So I think starting working backwards from GPUs, um, is usually pretty, pretty good idea.”

“It really comes down to your company. What data does your company have that's special? That your competitors don't have. Can you leverage that? And can you build AI that really understands that data? Because that's not a commodity. There's not an AI out there that understands all your business processes in your…”

“what we tell people is that I think the winners are yet to be identified and, and so experiment with more vendors, do shorter term contracts.”

“If the price of compute per like unit of intelligence or whatever, however you want to think about it. Fell by a factor of a hundred tomorrow, you would see usage go up by much more than a hundred. And there'd be a lot of things that people would love to do with that compute that just make no economic sense at the…”

“You know, we talk about the Moore's law improvement on one end, but the software improvements are much more exponential than that.”

“The reality is, like, the standard, like, operating procedures, like the SOPs, are largely in people's heads. And so, unless you have that tiger team, like, mix of, like, technical and, like, you know, subject matter experts, Really hard, like, to get something out the ground.”

“And evals also, oftentimes, need to come up bottom up. Right? Because all of these things are kind of in people's heads, in the actual operator's heads. Like, it's actually very hard to have a top-down mandate of, like, you got, like, this is how the evals should look. A lot of it needs the bottom-up adoption.”

“as a product builder, there's a latency, there's a real latency trade-off that you have to deal with where, you know, your user might not be happy waiting 10 minutes for, like, the best answer in the world. It might be more okay with the substandard answer and, like, no wait at all.”

“Reinforcement fine tuning introduces like RL or reinforcement learning to this loop. Way more complex, way more finicky, but an order of magnitude more powerful.”

“So you go back to GPT-IV, you know, and you're compressing, you know, the entire internet. But now we really don't need to do it because we've trained them to use tools like the internet, right? They're true reasoning engines.”

“I think part of the pitch to get everyone there is that it's open. Because everyone that they're pulling over are coming from closed places, and so if you're really passionate about the work you're doing, and you're passionate about where this is going to go, and there's only, there's only one company that can fund…”

“in the internet age, all the startups began with Oracle and Sun. And eventually they all moved to Linux and MySQL. And so there was a, there was a, we got to win it all cost phase. And then there was a phase where you started worrying about cost and optimization. And so one day, one day we'll likely, you know, make…”

“One of my biggest lessons looking at private market investors versus public market investors is the appetite for institutions in the public market for assets that are perceived to have significant downside, i.e., like, 70 or 80% that are marked to market, I have found is just really low.”

“So our advice being to entrepreneurs that if you are growing over 25%, you are profitable. Time to think about whether you should be public. But that doesn't necessarily mean going public. As you well know, there's a difference between being IPO ready and going IPO. But we think certainly putting all the steps into…”

“the biggest problem was there just weren't that many people connected to a high speed internet. So all the things we dreamed of occurring were just inefficient to happen at that time. Uh, but what's probably even more surprising, Bill, is how dramatically we underestimated the long term. Over the next 20 to 25 years,…”

“First, it's very hard to know which variables are dependent in a, in a complex multi-variable system, and, and, and there may be one you don't know about that it flips, and the whole system takes on a different shape, and this makes these things very hard to predict, which is why I don't like talking about macro for…”

“Google had a Piece of technology called Kubernetes. It was orchestration that would allow you to move a workload. If that became a standard from one, you know, one large server vendor to another, right? It basically created, um, ease of distribution. So you could run on multiple clouds. They went to the Linux…”

“If you get memory, the switching costs explode.”

“The processes for spending that amount of money are sticky and slow. They're not fast. Like, you can't back out. Like, you pre-commit, and you go build.”

“We can't teach it what good art is. Because we have no way to functionally prove what good art is. We can teach it to write really good software. We can teach it how to do mathematical proofs. We can teach it how to engineer systems, because there are, while there are trade-offs, and this is not like, it's not just a…”

“Yes, the queries are expensive, but they're nothing close to the human, right? And so each level of productivity gain I get, um, each level of capabilities jump is a whole new class of tasks that it can do And, and therefore I can charge for that. Right. So this is the whole like axes of yes, I spend a lot more to get…”

“So, you know, if you think that infrastructure expenses are going to grow at 30% a year, then I think you have to believe that the underlying inference revenues, right, both on the consumer side and the enterprise side are going to grow somewhere in that range as well.”

“Sometimes it's okay to fast follow, and it worked out, but you shouldn't do things out of envy. That was one of the hardest lessons I think we've learned. Uh, do it because you have permission to do this, and you can do it better.”

“Because I always, as I always say, from sort of ancient sort of Greece to modern Silicon Valley, there's only one thing that brings civilizations, countries, and companies down, which is hubris.”

“I think it is one of the most savvy competitive Weapons you can bring to the field. I'm sure they were worried that open AI would create proprietary data connectors and they want everyone to use this one, uh, very similar to what Google did with Kubernetes.”

“Historically, Mathematical algorithms that do some type of fit, and this is a very sophisticated form of that, but when you take the variables up to a certain level, they, it stops adding value. You just get too close to the fit.”

“The venture markets have transitioned from a high margin cottage industry to a institutionalized lower margin industry, and that has a lot of implications.”

“NVIDIA's competitive advantage is strongest where the size of the system is largest, which is another way of saying what Renee said. It's flipping it on its head. It's not to, not to say it's, it's, it's weak on the edge, but it's super powerful when you put a whole bunch of them together. That's when the networking…”

“you can train a model across a distributed site and it may just take you a, you know, a month longer because you have to move traffic around. And so instead of taking three months, it takes you four months, but you can't really run a model across a distributed site. Cause that inference is in like real time thing.”

“A lot of people don't even realize that it takes AI to curate data to teach an AI. And that AI alone is pretty complicated.”

“If you built it on this architecture, Without any consideration, it will run on this architecture, ok? You could, you could still go and optimize it for other architectures, but at the very minimum, since it's already been architect, you know, built on NVIDIA, it will run on NVIDIA.”

“A model is an essential ingredient for artificial intelligence. It's necessary, but not sufficient.”

“It is, it is, it is, I believe, uh, wrong-minded to be, uh, closed versus open. It should be closed and open.”

“airbnb and Disney and the cruise lines and the airlines start seeing downticks, now you know that the consumer is really getting pressured. So there is no doubt.”

“But remember, historically, rate cuts that are preceding a recession or fears about a recession are oftentimes sold, right? It doesn't cause the market to go up. It actually causes the market to go down.”

“when prices are high in venture, you know, you don't just stop doing venture. Right? But you do less. And you wait for the really great stuff. And when prices are high in the public markets, you don't necessarily step out of the markets altogether, but you just do less. And that's how we think about it here. It's less…”

“as you move into things that they've been moving into, crypto, um, defense tech, like, you're getting into heavily regulated industries, so you're gonna bump up against this more, and you're gonna think about policy more”

“he pointed out that if you go to the past and start to notice one to three companies just materially blowing out numbers above everyone's, uh, estimates over several quarters. There may be something happening that you should pay attention to.”

“one thing that seems blaringly obvious is this, is this teacher element, and Saul Kahn, who's present in some of the, the demos this week, um, has been leaning into this heavily. He's someone we should definitely pay attention to, um, but they look real, and we talked about the models are infinitely patient, which…”

“If you want to, you know, see four different hotel choices in Taiwan, I don't think you want that read to you. Um, I think that will be tedious versus looking at it on the screen.”

“In fact, I was talking to a large data company this week who's in the business of serving models, and they were lamenting the fact that the pricing umbrella was set by open AI and there is no price, right? So that it's a much lower margin business than the traditional software business.”

“no single VC's gonna stand up and make a company go public. That's not gonna happen.”

“And I think this is one of the big psychic or behavioral hurdles to these companies getting out and getting into the public markets.”

“The capital intensity of that undertaking is very different than starting a software company, right? And so, um, the risk reward to both the founders and the risk reward to the investors changes a lot.”

“And so, if Washington wants to get serious, it needs to be an integrated, right, national policy. You can't fix AI without also taking on, um, our future energy needs.”

“if, if we can't convince somebody that a dollar worth of stock is worth a dollar, Then you actually shouldn't grant it. You should just give a dollar of cash because they'll actually value it, you know, greater.”

“when I look at what I see going on in the startup world, they might start with one of these, you know, really well-known service models that's proprietary, but the minute they start thinking about production, they become very cost focused and on the inference side, and they'll just play these things off of one another…”

“your gross margin will ultimately determine your operating margin. Um, there's almost no way that, that you can, you can kind of make those two, um, you know, kind of get out of sync. Um, and so if you're, if you're subsidizing something or, you know, or, or in just such a commodity business and you're, you know, 40,…”

“there are some of these functions that are so high leverage that if you just nail it, um, you know, the, the deal desk legal team who's, who's negotiating the contract terms. I mean, these things are like, they literally are, are trajectory defining in your cashflow.”

“one has a 90% retention rate or something. Um, which means, which means, um, that sales and marketing is required forever to refuel the customer base because of the churn rate of that product. And, and it means you'll never actually converge on, you'll never actually converge on, on cashflow at the level of a company…”

“a lot of times the product market fit of the first product will get you, especially a really good product market fit will get you to a hundred million or two hundred million or some, some really large level. Um, but then start to Peter out and then you have to figure out other growth drivers.”

“when I look at those healthcare companies, I feel like there are two big buckets of needs. One is like Pure R&D. It's like, you know, you're seeing like a massive amount of data and like you have super smart scientists who are trying to, you know, combine, test out things, you know, so that's one bucket. A second…”

“I've often said that the public markets are the buyers. Of private companies, and you got to know them. You got to know them. You got to study them. You got to know what they want to buy.”

“when a startup would get started in early years, you might have as much as, say, 10% dilution from options. You might hire three executives during that year, and it's a big grant, but then as you move towards being public, people would typically, and these are Very gross rules of thumb steer towards, you know, a three…”