Erdem Kiciman

Something that merits more discussion amongst those who believe it was a mistake by both the Biden and Trump administrations to allow the export of less powerful GPUs to China: Since the summer of 2025, it has increasingly been *China*, not just DC, blocking Nvidia GPUs in China. Within weeks of Secretary Lutnick saying the goal was to get Chinese developers "addicted" to the American technology stack, Beijing began pressuring its tech giants to stop buying the H20. Then it ordered them to stop buying the RTX Pro 6000D, sent customs teams to the ports to crack down on Nvidia chip imports, and barred foreign AI chips from new state-funded data centers. And as of last week, Reuters reports that roughly 600,000 H200s are sitting in Hong Kong because Beijing won't approve large-scale imports of a chip that is still better than anything China's own industry makes at scale. If allowing the export of these less powerful GPUs was a policy mistake that *advantaged* China, then why is *China* the one blocking them? From my perspective, China understands that if they are dependent on American AI hardware - either on the ground in China or accessed remotely via the wide range of southeast Asian clouds (which include SE Asian availability zones from the largest American hyperscalers) - then we have immense leverage over them in any future conflict. Being able to turn off the clouds instantly is actually the greatest leverage. Yes, Chinese AI labs still want Nvidia chips. Gray market prices for H200 servers have roughly tripled over the past year. But that is exactly the point. Beijing is willing to impose real costs on its own AI labs to avoid dependence on American GPUs, which tells you how Beijing views GPU imports: not as a gift, but as a threat to its own ambitions. Maybe, just maybe, selling China GPUs that trail our own state of the art, so that Chinese developers stay “addicted” to American GPUs and China doesn't build an alternative AI hardware ecosystem that we have no control or influence over *and* that China could share over time with our adversaries, was actually a smart, pro-American national defense policy. And letting them access southeast Asian clouds might be an even smarter policy as so easy to turn clouds off. I believed and still believed it was smart policy. And I say this as someone whose posts here are hawkish enough on China that I won't travel there anymore. And note that both the Biden and Trump administration allowed the export of less powerful GPUs to China for all of these reasons. It was only under the Trump administration that China woke up to the risk of what was actually a pretty smart, pro-national defense policy and then China began restricting GPUs. The lesson of the H20 is that a deliberately crippled chip doesn't keep China on our stack. Beijing moved against the H20 and RTX Pro 6000D as it concluded that domestic Chinese chips had caught up to them. So if China develops an alternative AI hardware ecosystem that it shares with Russia, Iran and North Korea, we will look back and think it would have been even smarter to offer China the B30A in 2025 instead of the H200s we eventually approved case-by-case. Especially given the H200 is not the frontier: with Vera Rubin now in full production, it is two generations behind what American labs are buying. All of this is to say that Jensen has been consistently lobbying for a position that was adapted by both the Biden *and* Trump admins *and* looks smarter every month that goes by as Chinese labs increasingly adapt their MoE architectures to domestic Chinese hardware. DeepSeek reportedly gave Huawei and Cambricon early access to V4 instead of Nvidia and AMD, and a Huawei-led team has since post-trained V4-Pro on Ascend chips. And if America *ever* falls behind China in AI, the China hawks here in America who helped accelerate the creation of this alternative Chinese AI hardware ecosystem will deserve a significant share of the blame.

I’m thrilled to announce that Kai Asset Management is launching two new funds run by two great additions to our team: @NoelSmith and Scott Maidel (@volatility911). This is part of our belief that, following last week’s midterm elections, U.S. markets will be volatile in the next two years. We got a taste of that last week. Thanks to Bloomberg’s @BernardGoyder for covering the story. Read more in our press release: kai-am.com/media

President Trump sat down with Ross Douthat in November 2026 to discuss Republican losses in the midterm elections, the wars in Iran and Ukraine, the economy, and more.

Shopify captured 50% of incremental growth in ecommerce since 2025. In case you’re unfamiliar with numbers, 50% can also be referred to as a “🚤load” or a “💩ton.”

Introducing Reality-Aware Rendering in @Simulon, a completely new way to create with the real world. Place neural assets with your phone and act out the scene. When you stop recording, Simulon renders the take, reading everything in it (what you do, what you say and the place around you), and your assets respond to all of it. The final is regenerated from your recording: a new performance of the scene, not a copy of the footage. Available now on iPhone. Coming to Android soon.

have a discontinued Mirror from Lululemon? I rebuilt mine to connect to Schoology/powerschool/google calendars etc. It also highlights overdue library books. DM me for instructions.

Fun to co-host the GTC Berlin pregame show with @dee_bosa and nice of her to invite me on the postgame show. Deepseek is generating $38 in inference revenue per GPU hour off-peak. That is $100b per gigawatt. 2x this at peak. Seems important.

In 1983, a BBC interviewer asked Richard Feynman why two magnets push each other apart — and he refused to answer it. What he did instead is the best seven minutes on thinking ever filmed. Bookmark & watch today, no matter what.