And what happens when China cuts us off? #
**Tuesday. **Today we’re diving into the possible end of open-weight AI models from China, startup discounts, consciousness, and why being public would solve so very many problems in startup-land today. To work! — Alex
📈 Trending Up #
The global demand for data? …Nvidia’s AI model market share…the War with Iran…enterprise token demand? …vertical AI…hindsight…app-store-level age verification…French robots…dilution at Rivian…
Closed-source Chinese models? Reuters reports that the Chinese government is considering “restricting overseas access to China’s most advanced AI models, including those yet to be released.” That’s not all, the publication adds that the CCP has discussed “the possibility of implementing new measures to restrict who can fund domestic AI startups.”
The United States has similar rules in place, including export controls on certain AI technologies and limits on Chinese investment in domestic technology companies. Regardless, the era of super-cheap, nearly as performant AI models from China could be coming to a close. Such restrictions would be a massive blow to the global open-weight AI industry, which has delivered rapid intelligence gains at basement prices.
Ironically, any move by the Chinese government to limit the export of its best AI models could dramatically boost closed-source American and European AI labs. Imagine the delight of Anthropic, OpenAI, and Mistral if suddenly they weren’t competing with DeepSeek, Alibaba, MiniMax, Z.ai, Moonshot, and the rest? Hell, concerns that open-weight models may undercut the value of SOTA, but expensive American AI models would evaporate overnight.
- Perhaps a business model shift is in the offing, too. The Times reportsthat Alibaba is having a bear of a time wringing sufficient revenue from its AI operations. Which, to be fair, is an uphill journey as I can download 458 different Alibaba AI models from Hugging Face. Can’t do that with GPT-5.6 or Fable! - Why are government restrictions and closed-source monetization linked? If the Chinese government clamps down on the releaseof new AI models from its domestic technology industry, they will not be open by default. And that means that the open-weight AI labs won’t have to compete with every inference provider in the world to sell their models. Less competition, higher prices, more revenue. Bingo.
(I’ve been texting with Reflection AI for a bit, gently poking the company for not having a model out yet that I can play with. Theoretically *the *American *open *AI lab of note, it could snag the narrative by releasing something blistering and at least open-weight around the time that China its doors shut.)
Startup AI discounts: Speaking of lower-cost AI models, American AI labs are offering fat discounts to startups, the Journal reports. These include massive discounts on SpaceXAI’s Cursor subsidiary, or simply millions of dollars’ worth of tokens offered.
The idea here is the same as zero-cost cloud offerings for startups. It’s really what your parents warned you about drugs when you were a kid: The first taste is free! For cloud providers AI labs the exchange is burn today for revenue tomorrow. Just as startups might get hooked on using Azure over AWS early and stay (the commercial version of Proverbs 22:6), they could become long-term OpenAI-shops, or forever Anthropic-pilled as they scale.
- OpenAI made a splash by offering tokens for equity to several Y Combinator cohorts. I believe the term for that was Simple Agreement for Current Tokens. Expect more of the same as startups grow faster than ever; the prizes are bigger, so too will the bids to earn them become.
[📉](https://finance.yahoo.com/news/servicenow-pledges-1-5bn-investment-110000403.html) Trending Down
[📉](https://finance.yahoo.com/news/servicenow-pledges-1-5bn-investment-110000403.html)
Not understanding loops…draining the swamp…Samsung shares after earnings…global Nvidia dominance? …Meta’s market cap? …paying for AI with cash…Russian oil production…
**Our monopoly on consciousness: **Of all the AI labs, Anthropic is the most ooey-gooey, which I appreciate. While most AI labs treat their AI models as little more than balls of math, Anthropic wants to know how Claude feels! A few admissions before we go on:
- I don’t think that humans have a unique claim to consciousness;
- I view consciousness as a spectrum, with humans at one extreme;
- I don’t think that biologyis necessarily a gating function for consciousness; - I found the Ted Chiang story “The Lifecycle of Software Objects” persuasive in empathy terms, and grew up reading Asimov on the subject of sentient robotic ethics.
All that said, Anthropic just dropped a paper detailing an interesting finding in its AI models. Mapping global workspace theory to how Claude thinks, Anthropic details what it calls the “J-space” in its models. In short (we will revisit this topic after I read through the pertinent academic papers; which me luck), much of your brain’s work is occluded from your consciousness. That ‘hidden’ thinking matters ….
Subscribe to Cautious Optimism to unlock the rest.
Become a paying subscriber of Cautious Optimism to get access to this post and other subscriber-only content.