Compute

“Compute is the biggest constraint.”

“There were users who were spending a shit you not like 2000 dollars of compute a month, 4000 dollars of compute a month. There's someone on the platform who'd spent in five months was like 26,000 dollars.”

“It takes the exact same amount of compute. All else equal to make an open source token as a, you know, frontier token for a comparably sized model. Now there's a lot of nuances there, but that's broadly true. It's just a question of what are the margins that a…”

“You're just simply not going to be the low cost provider. If you're not vertically integrated, if you don't own your own compute over the very long. Long-term.”

“It doesn't look like we're going to suddenly end up in a regime where like you could train super intelligence with a really cheap recipe, as opposed to it being more of an iterative thing where like the cost keeps going down, the capabilities keep going up. Th…”

“Where does that leave frontier model companies? They're the most at risk. Why? Because they're not investment grade. They're in the worst position in terms of their balance sheets. They're not in a position to then get access to the compute. They need to be su…”

“Turns out there is like some big hobbling that we have unwittingly, like some, something we were doing wrong that just made these systems way less efficient than they could be. And when somebody figures out how to remove that, like maybe the current Compute is…”

“The quality is bad, and you put analytics on it, or AI on top of it, you're not going to get good results. And even if you hope to get good results, you'll end up spending a lot of money on compute.”

“TSMC, Decreased their rate of growth in twenty-twenty-three and twenty-twenty-four and twenty-twenty-five. So, like, we're, our shortage of compute is going to get worse in the next few years.”

“So, today when we say there's not enough compute, it's not like all the money that the companies are putting in today manifests in compute tomorrow. No, it all manifests in compute in twenty-twenty-eight and twenty-twenty-nine.”

“If energy is what you need for compute and is the bottleneck, And there's unlimited demand for intelligence. Then it stands to reason that if we have a lot less energy, then we will have a lot less intelligence or a lot less tokens or a lot more expensive toke…”

“I know some labs have slowed down on their hiring of researchers unless they're above a very, very high bar because the cost isn't the research or it's the compute associated with the person.”

“I think in the next three, four years we won't be paying for intelligence, we'll be paying for compute through our nose.”

“Essentially, you have the contracted base of installed compute trading at a massive discount to the current spot market. And as those contracts roll off, and Compute gets repriced higher, and Spock can decline, and Compute will still get repriced higher. You k…”

“Anthropic, if they had been as aggressive on compute as OpenAI had been, they would have run away with it.”

“I mean, this is, like, getting used today. Like, there's high demand for it, so when you see people, like, make these bets on compute, like, Meta or SpaceX AI, and they're off on the timing, right, or maybe off Totally. If depending on your opinion, they can t…”

“Because when you think about a data center, the power consumption, it's roughly 60% of the power is being used to compute and 40% is being used to cool.”

“I think that the price of compute is going to go up and that will provide an advantage to The models that have the most lucrative algorithms that are able to produce the most intelligence per watt or the most intelligence per token or per GPU.”

“We did under do it.”

“Our biggest de-risks now for upcoming runs are as big as like the entire compute run from 18 months ago or something.”

“Compute, man. Let's imagine that we could build this product. This product that could just do exactly what I said for all your stuff.”

“We went from a, let's call it a A software-centric world to a compute-centric world.”

“Today we're all talking about compute, compute, compute. I think the primacy of memory will become even more important.”

“Our work suggests that 26, 27 natural gas is, is appropriately supplied. But as we get into 28, and you plug in this compute, and you assign gas to very specific assets as they're plugged in as well, and you continue to export LNG as we're planning to do with …”

“Inference is a big, perhaps even the majority on compute.”

“If you're doing a model company, they're probably going to allocate 70, 80% to compute. You know, if you're doing a coding company, maybe 50, 60%, you know, conversely, a lot of the apps companies I'm looking at are, if you look at it, they don't even, they al…”

“You either have compute that sells itself or people who sell it, but you can't do both.”

“We should be thinking about spending as much on the environments as on the compute and the data, because the environments literally shape the, what intelligence can emerge.”

“I do think there's a bit of an attractor state where just like from a business model perspective, every provider sells out all their compute. Right. I think that is just like the world that we're heading towards, where there just is not going to be enough comp…”

“I do think there's a bit of an attractor state where just like from a business model perspective, every provider sells out all their compute. I think that is just like the world that we're heading towards, where there just is not going to be enough compute to …”

“If I abstract, if I step back today there's not enough compute for what the world is demanding. Unequivocally not enough compute, right? You can't buy compute. Compute is costing two to three X or four X more than it used to cost two years ago. There's not eno…”

“And sooner or later, I expect the cost of money, what I'm going to spending on tokens is going to be way north of what I'm spending on compute someday.”

“If we continue this curve within like three years, you know, we're just gonna be so short on all the compute.”

“Atropic, no way did they think they were going to be anywhere close to breakeven. Right? In, in, in, in this part of the curve. And the reason, like I called it accidental profitability that, you know, people have been talking about that because they want to s…”

“Having said that, I think it's also very difficult to get things right because the same three things are accessible to everyone. So really to stand out and then build something amazing in this day and age is going to be you know, exciting nevertheless, right?”

“I think the, they're gonna get more expensive as the models get better the compute needs are gonna, you know, increase, and then you're gonna get better.”

“Compute will need to move to space at some point.”

“This, you know, this crude way of saying it is intelligence is log of compute kind of works.”

“Compute might become a bottleneck again, because previously, like if you want to train a new model, say you want to generate new synthetic data and then, or write a new algorithm, it might take a few weeks. And during that period of time, you don't, you might …”

“I don't know about 24 months time, but I would bet that in five years the average enterprise spends more on compute than headcount.”

“Maybe a true statement is that if anthropic had all the compute, they'd probably be doing Well north of a hundred billion dollars today. Maybe a 150.”

“The number probably goes up if demand's going up, right? Just natural, basic macroeconomics. So if that happens, we're gonna find ourselves in a position where you might be saying, oh, I could replace That person with some level of agentics, but you know, it's…”

“If you buy too much compute, you go out of business. If you buy too little compute, you can't serve your customers and you're not at the frontier.”

“I would say I spend 30 or 40% of my time on compute even today.”

“There's a level of compute for model development that we will not go below, right? So even if it means it's harder to serve customers, or we have to do kind of unnatural things when it comes to that, we want to continue to make that long-term investment in dev…”

“So the answer is, you know, we're constrained kind of across those use cases internally today.”

“We, You know, run workloads on one day in the morning on a chip for inference, and in the afternoon or evening we use it for model development.”

“Look, I don't want to get sued by shareholders for whom it's not legible why I'm giving away billions of dollars of compute at cost of portfolio companies, right? Because that's what we're doing.”

“OpenAI has six hundred billion dollars in spending commitments for compute.”

“Anthropic is so underserved in compute that they're buying, like, in last minute deals and on the smart market and all that kind of stuff, which is bad for their margin, but is good for the providers.”

“I think, like, there's no world in which there's enough compute to, you know, get the amount of value that we want to get out of our limbs in the next five to 10 years.”

“It's possible you look back and see that as the first disconnect from compute equals revenue.”

“Like, nobody has enough spare compute to just make two, you know frontier models at maximum size, right, with different attributes, so that's pretty difficult.”

“Recommendation system space does not need as much compute to create the output that's necessary to succeed.”

“We rent, build, buy, compute, and we resell it with some positive margin, and as long as it's, you know, positive operating margin, and as long as there's scalable demand for intelligence, which I think is true as long as there's problems to solve, like, no on…”

“Still not enough. We're gonna feel the scarcity. We're gonna feel it. We're feeling it already. You can sense it right now on people who are trying to use these agents and just simply cannot, you know, hitting the rate limits. So we're working on behalf of our…”

“I would say that we, in general, are heading to a compute-constrained world.”

“I mean, I think our competitors are not having a good time on compute, let me put it that way.”

“I think if you play this out, the real problem, again, goes back to Anthropic and OpenAI. If I were them, it is a five alarm fire for them. They more than anybody else needs to get their hands on compute. They need to have LAN PowerShell, but otherwise that re…”

“In, in certain domains that are well explored, like coding, for example, yes, there's an increasing amount of compute required to get an incremental gain in some eval that's super saturated.”