Dylan Patel: evaluation

Transcript context

…—the Ironwoods to Anthropic, and you’re saying the big bottleneck right now, this year or next year, I guess going forward forever now, is going to be the logic and memory, the stuff it takes to build these chips. Google has DeepMind, the third prominent AI lab. If this is the big bottleneck, why would they sell it rather than just giving it to DeepMind? This is again a problem of… DeepMind people were like, “This is insane. Why did we do this?” But Google Cloud people and Google executives saw a different thought process. You and I know the compute team at Anthropic. Both of the main people came from Google. They saw this dislocation, they negotiated a deal, and they were able to get access to this compute before Google realized. The chain of events, at least from our data that we found, was in early Q3, over the course of six weeks, we saw capacity on TPUs go up by a significant amount. It went up multiple times in those six weeks. There were multiple requests. Google even had to go to TSMC and explain to them why they needed this increase in capacity because it was so sudden. A lot of that capacity increase was for selling to Anthropic. Because Anthropic saw it before Google. And then Google had Nano Banana and Gemini 3 which caused their user metrics to skyrocket. Then leadership at Google was like, “Oh.” Then they started making the statement that we have to double compute every six months, or whatever the exact number was. They really woke up a lot more, and then they went to TSMC and said, “We want more. We want more.” TSMC replied, “Sorry guys, we’re sold out. We can maybe get 5-10% more for 2026, but really we’re going to work on 2027.” There was this information asymmetry among the labs, in my mind. I don’t know exactly. It’s the narrative I’ve spun myself from seeing all the data in the supply chain on wafer orders and what’s going on with the data centers that Anthropic and Fluidstack signed. It’s pretty clear to me that Google screwed up. You can see this from Google’s Gemini ARR. They had next to nothing in Q1 to Q3—in Q3 a little bit once they started inflecting. But in Q4 they reached $5 billion in revenue on an ARR basis. It’s clear Google didn’t see revenue skyrocket initially. In a sense, Anthropic had a little bit of commitment issues before their ARR exploded, even though they had far more information asymmetry and saw what was coming down the pipe. Google is going to be more conservative than Anthropic and Google had even less ARR. So they were just not willing to do it, and then they realized they should do it. Since then, Google has gotten absurdly AGI-pilled in terms of what they’re doing. They bought an energy company. They’re putting deposits down for turbines. They’re buying a ridiculous percentage of powered land. They’re going to utilities and negotiating long-term agreements. They’re doing this on the data center and power side very aggressively. I think Google woke up towards the end of last year, but it took them some time. How many gigawatts do you think Google will have by the end of next year?…

Stored transcript either side of the excerpt. The highlighted words are the published quote; the surrounding text is unedited source, never generated.