The Vergecast · Tech Strategy & Big Tech · November 2026
Nilay Patel had suggested a rational move for the frontier labs would be to pause training new models. When his co-host objected, he explained why it wouldn't kill the companies, and why they won't do it anyway: each fears a rival's next model.
A rational economic decision for both OpenAI and Anthropic today is to stop training Frontier models and find ways.
Shut down their company. No,
no, no. If you, they have capable models, right? Like Shastra is very capable. Claude is very capable. Google keeps announcing new atoms every single day. Another Gemini Adam has been discovered. I'm going to talk about that later. Gemini 95 Uranium is finally here. Like whatever is happening. The models are very capable. They have not diffused broadly into society or business. You could stop burning money training new models today and say, my only job is to sell more tokens to more industries. We're going to just go, we're going to figure out. Accounting and what and healthcare and real estate, like whatever it is. We're going to get more customers. And because they're nowhere close to using the capability of the models as they exist today, we do not have to bear the cost of training a new model. That would be a totally rational decision for any of these companies to do. But they are…
it's an interesting point. And I think even more so than that, like Ben Thompson was on a podcast a while ago comparing it to the railroads, which is a comparison a lot of people like to make, right? Huge capital expenditure. And the bet is to, and his point was that took forever. And that was a…
Yeah. And again, it is, they all point to there's ever-increasing demand for AI.
Yeah.