TK Tae Kim Tae Kim Best-in-class insights and analysis from THE NVIDIA WAY book author Tae Kim. "Key Context" Substack, covering AI, semiconductors and tech.

“I think what you're going to see in the next six months is that the gap between Anthropic and OpenAI versus Chinese models is going to widen dramatically. Because the only reason, I mean, this is obvious if you've been following everything, the only reason these Chinese models are good is because they're just distilling the American models. … Both these labs have been cutting off the distillation accounts from China.”

TBPN · Startups & Venture · November 2026

“I think what you're going to see in the next six months is that the gap between Anthropic and OpenAI versus Chinese models is going to widen dramatically. Because the only reason, I mean, this is obvious if you've been following everything, the only reason these Chinese models are good is because they're just distilling the American models. … Both these labs have been cutting off the distillation accounts from China.” — Tae Kim, TBPN

Context

Asked whether cutting China's access to Blackwell chips would widen the gap with open models, Tae Kim argues chips aren't the main constraint on Chinese labs. He says the bigger factor is US labs shutting down accounts used to distill their models.

Transcript

TBPN Around 48:50 into the episode
Jordi Hays

They're just like, we want compute. But the government is taking a different policy, and so you can't look at it as a single stance.

John Coogan

…frontier compute available, the DeepSeek, Moonshot, MiniMax, Ant Group decreases over the next year or two. Now, you could have Huawei and SMIC and SME take off. That's certainly a possibility, but at least in the short term, it feels like China is going to be more GPU-poor in the next year or two.

Tae Kim

So I think the bigger problem for China's models is not that. It's that right now we're going through this RSI thing where Anthropic and OpenAI are hitting apart the RD cycle where AI is just going to take off. And what both these labs have done over the last, I think this has been undercovered, is that they've been cutting off the distillation accounts from China. And I think what you're going to see in the next six months is that the models from Anthropic and OpenAI are the gap between Anthropic and OpenAI and versus Chinese models are going to widen dramatically. Because the only reason, I mean, this is obvious if you've been following everything, the only reason these Chinese models are good is because they're just distilling the American models. And

Jordi Hays

they're buying the same data.

John Coogan

Yeah, there's a little bit of a problem.

Tae Kim

Yeah, yeah. It's mainly installation. And Anthropic's been very good at this, kind of, you know, putting out a document that shows what's going on. And both these labs have been clamping down on it. And as RSI takes effect, I think the leadership of the American AI labs are going to really show up.…

Speaker names come from our own diarization. The position is estimated from where the line sits in the episode.

More from TBPN