83% vs 39%: Why China Loves AI, America Fears It — and Who's Right

 This week in AI

Last Thursday, a Chinese startup you've probably never heard of did something remarkable. Moonshot AI released Kimi K3, the largest open-weight AI model ever built — 2.8 trillion parameters, which is a fancy way of saying it's enormous. On some benchmarks it outperforms the best models from OpenAI and Anthropic, and it runs at about a third of Anthropic's price (Prof G Media, 2026, 00:00:44). Even more remarkable: unlike American frontier models, which are locked behind corporate walls, Kimi K3's weights are free. Anyone can download it, study it, and build on it.

If that sounds like a technical footnote, it isn't. It's a snapshot of the strangest divide in the world right now: the country most excited about AI is giving it away, while the country most anxious about it is arguing over whether to accept the gift. According to Stanford's 2026 AI Index, 83% of Chinese respondents feel optimistic about AI — versus just 39% of Americans (Stanford HAI, 2026). This post is about that gap: where it comes from, who has the better argument, and what it means for you.

The week AI started moving at a different speed

To understand why emotions are running so high, you need to understand how fast things are moving. It used to take AI labs six to twelve months between major releases. Now it takes weeks. Kimi K3 landed in one of the most crowded release windows the industry has ever seen, arriving within days of several other major models.

Part of the reason is something called recursive self-improvement — AI helping to build the next AI. That is the second huge news item. This used to be science fiction. Then, in February 2026, OpenAI wrote in its release notes that early versions of GPT-5.3-Codex were "instrumental in creating itself," helping debug its own training runs (Communications of the ACM, 2026). Anthropic, the maker of Claude, has warned that AI systems may be approaching the point where they can design their own successors with little human input, and has called on labs — including itself — to prepare for a coordinated slowdown if that happens (Scientific American, 2026).

Speed changes economics too. On the Prof G podcast, analyst Ed Elson laid out the price collapse: OpenAI's top model costs $45 per million output tokens, Claude costs $50 — and DeepSeek's Chinese equivalent costs $0.87, roughly 99% cheaper (Prof G Media, 2026, 00:03:52). Scott Galloway calls this "AI dumping": free Chinese models went from less than a third of global traffic in late 2025 to about two-thirds recently (Prof G Media, 2026, 00:01:19). Whatever you call it, intelligence is getting cheap, fast.

Two countries, two completely different moods

Now here's the puzzle. You'd think the country producing these breakthroughs and the country using them most would feel roughly the same about the technology. They don't — not even close.

Stanford's AI Index found 83% of Chinese respondents optimistic about AI's future, against 39% of Americans (Stanford HAI, 2026). A separate survey discussed on Prof G found that 84% of Chinese respondents were more excited than worried about AI, compared with a small minority of Americans (Prof G Media, 2026, 00:07:49).

Why? Selina Shu, a former Bloomberg China reporter, offered the most convincing explanation: pragmatism. In China, she says, people simply don't see AI as "this machine god." Talk of artificial general intelligence is "pretty much absent from most convos" — for the average Chinese person, AI is a tool you learn, often because your boss tells you to. The American-style doom discourse — "AI will make us extinct... It would be like the Terminator" — barely exists there (Prof G Media, 2026, 00:08:54–00:09:25).

Americans, meanwhile, marinate in a culture where AI is either the apocalypse or a scam, and where the loudest voices are often the most extreme.

And here is the irony, and it's a big one. While Chinese users cheerfully treat AI as a power tool, Washington is seriously debating whether Americans should even be allowed to use Chinese open-source models — models that are free, capable, and inspectable by anyone. Analyst Patrick Moorhead called the debate ironic, noting that "the Chinese seem to be doing fine with their models" (CNBC, 2026). Think about that. The country that fears AI the most is arguing about whether to refuse a free gift from the country that fears it least. If anxiety were a reliable guide to policy, this would be the moment it proved its worth. Instead, it looks a lot like fear eating strategy for breakfast.

So who's right — the optimists or the worriers? To answer that, we need evidence from both sides.

The optimists have receipts: the protein-folding revolution

Optimism about AI isn't just vibes. In 2024, the Nobel Prize in Chemistry went to Demis Hassabis and John Jumper of Google DeepMind, alongside biochemist David Baker, for cracking one of biology's hardest problems: protein folding (Nobel Prize Outreach, 2024).

Here's the sixteen-year-old-friendly version. Proteins are the molecular machines that do almost everything in your body, and what a protein does depends on its 3D shape. For fifty years, figuring out that shape from its chemical sequence was brutally slow — sometimes a single protein took a PhD student years. DeepMind's AlphaFold learned to predict those shapes in minutes, and then did it for essentially every protein known to science, releasing the results free to researchers everywhere. That's not a chatbot writing your essay. That's a genuine acceleration of medicine: faster drug discovery, better understanding of diseases, new enzymes for breaking down plastic.

Erik Brynjolfsson, the Stanford economist who has spent thirty years studying what technology does to jobs, told an interviewer he recently visited DeepMind, where Demis Hassabis said he believes they'll "start curing a majority of diseases within 10 years." Brynjolfsson's reaction: "I hope it's right. That sounds ambitious but your daughter will see that" (Silicon Valley Girl, 2026, 00:51:06). Ambitious, yes. But this is the same lab that already won a Nobel for doing something biologists thought was decades away.

So... will AI take your job? An honest answer

This is the question that actually matters if you're sixteen. And honesty requires admitting the news is genuinely mixed.

Start with the bad. Brynjolfsson's own research paper, "Canaries in the Coal Mine," found that AI has already reduced employment by about 16% for workers under 25 in the most AI-exposed occupations (Silicon Valley Girl, 2026, 00:00:56). Asked directly whether someone should train to become a junior software engineer earning $95K, his answer was blunt: "No. Unfortunately, that's one that's very much in the bullseye... We see that in the data they're disappearing" (Silicon Valley Girl, 2026, 00:16:45). Mid-level marketing manager? Same answer. Paralegal? "Oh my gosh, it's even worse" (00:19:19). Across the US, nearly 55,000 job cuts in 2025 were directly attributed to AI (AIMultiple, 2026). And Brynjolfsson doesn't sugarcoat it: "There are a bunch of jobs, millions of jobs that are going to disappear" — already happening, not someday (Silicon Valley Girl, 2026, 00:19:44).

Now the part the doom headlines skip. The same data shows the least exposed occupations are growing, older workers are gaining employment, and — most interesting of all — people who use AI to augment their work rather than automate it "did significantly better" (Silicon Valley Girl, 2026, 00:01:25). The pattern for most workers isn't replacement; it's transformation. The job title stays, the work inside it changes.

Brynjolfsson's favorite example is the radiologist. In 2017, AI pioneer Geoffrey Hinton famously said we should stop training radiologists because AI could read medical scans. Instead, "we now have more radiologists than ever. There's almost a shortage" (Silicon Valley Girl, 2026, 00:22:04). Why? Two reasons. First, reading images is only one of a radiologist's 26 distinct tasks. Second — and this is the key economic idea — when something gets cheaper, people often want much more of it. If an MRI costs $2,000, you skip it; at $200, you get that sore shoulder checked (00:22:56). Cheaper medical imaging means more scans, more patients helped, and more demand for the humans who do everything around the scan. Economists call this elastic demand, and Brynjolfsson estimates roughly half the economy works this way (00:05:44).

His practical advice is surprisingly concrete. Every project, he says, has three parts: defining the question, executing it, and evaluating the result. AI is getting very good at the middle part. The humans who thrive will be the ones who master the first and third — asking the right questions and judging the answers — becoming, in his words, "kind of like the CEO of a bunch of agents" (Silicon Valley Girl, 2026, 00:13:28). And in a twist nobody saw coming, he argues the liberal arts — philosophy, art, music — are becoming more valuable, because they develop taste and judgment, exactly what machines lack (00:24:39).

One more dose of realism: even the optimists admit the money side is wobbly. Oracle borrowed $43 billion to build data centers while burning negative cash flow, and as Prof G's Ed Elson put it, "bubbles aren't built with equity, they are built with debt" (Prof G Media, 2026, 00:12:47–00:13:25). The technology is real; some of the valuations may not be.

Conclusions and implications for decision-makers

If you've read this far, you're probably not sixteen — you're someone who has to make decisions about AI for an organization. So let's translate everything above into three concrete takeaways.

For governments: get involved intelligently. The evidence in this post points to one clear failure mode: doing the disruption without doing the transition. Brynjolfsson's warning bears repeating — with free trade, "we did the first part... but we didn't do the second part where we helped out the people who were hurt," and the result was a backlash that economists now call a catastrophe (Silicon Valley Girl, 2026, 00:20:40). His prescription is not to freeze old jobs in place, which "hasn't worked for a country" ever (00:20:13), but public investment in education and retraining, done before the displacement peaks, not after. Intelligent involvement means funding the transition, not fighting the technology.

For the private sector — especially in Europe: stay equidistant. European companies have long defaulted to the US ecosystem out of habit. That habit is now expensive. When a Chinese open-weight model delivers most of the capability at a fraction of the price — recall the $50 versus $0.87 per million tokens comparison (Prof G Media, 2026, 00:03:52) — refusing to even understand that ecosystem is not prudence, it's negligence. The strategic position for a European firm is to be as close to the Chinese open-source ecosystem as to the American closed one: benchmark both, build switching capability, and let the price war work for you rather than around you.

And keep an eye on Brussels in 2027. Horizon Europe is the world's largest research and innovation programme, with a budget of €95.5 billion — and it is about to get dramatically bigger. The European Commission has presented plans to almost double Horizon Europe's budget to €175 billion under the next long-term budget for 2028–34 (Science|Business, 2025b). The numbers aren't final: they still have to go through negotiations between the European Parliament and member states, with an agreement envisioned by the end of 2027 — which makes 2027 exactly the year to position yourself, before the first calls open in 2028. There is real reason to hope the evaluation system will tilt toward newer, more concrete players this time: the proposal includes a tripling of the budget for the EU's innovation scheme, the European Innovation Council, with the increase going largely toward start-ups and close-to-market technology rather than purely curiosity-driven research (Chemistry World, 2025; European Commission, 2025). For startups and applied AI programs that found previous framework programmes impenetrable, this is the most favourable structural shift in a generation. The organizations that will win those grants are the ones who start understanding the programme architecture now — not the ones who discover it when the calls are published.

The common thread across all three: agency. Brynjolfsson's closing point applies to institutions as much as individuals — when tools become more powerful, "by definition we have more agency," and the real question is not what AI will do to us but what we want to use it for (Silicon Valley Girl, 2026, 00:52:01).

Cautious optimism is a strategy, not a mood

So who's right — China's 83% or America's 39%?

Honestly, neither, fully. Chinese pragmatism gets the near future right: AI is a tool, and tools reward the people who pick them up. But American anxiety isn't irrational either — Brynjolfsson himself says the next decade could be "the best decade in human history by far" or "one of the worst 10 years ever," depending on choices we make about safety, concentration of wealth, and helping displaced workers (Silicon Valley Girl, 2026, 00:00:28, 00:51:31). His warning from the free-trade era is worth remembering: we did the disruption, skipped the part where we helped the people who got hurt, and got a backlash. "What's happening with AI, I think, is 10 times bigger" (00:21:33).

His closing advice works whether you're sixteen or sixty: stop asking what AI will do to you and start asking what you want to do with it. AI, he says, should stand for "amplifying intention, not artificial intelligence" — it takes your agency and multiplies it, and "if you don't have any, it doesn't do much for you" (Silicon Valley Girl, 2026, 00:28:41).

As he told his interviewer: "If you're not both excited and scared, you're missing at least half the story" (00:50:42). Hold both. That's not sitting on the fence — that's the only position from which you can actually see the whole field.

---

References

AIMultiple. (2026). Top 20+ predictions from experts on AI job loss. https://aimultiple.com/ai-job-loss

Chemistry World. (2025, July 23). European Commission plans to double EU research budget to €175bn for 2028–2034 programme amid concerns over industrial focus. https://www.chemistryworld.com/news/european-commission-proposes-massive-increase-in-horizon-europe-budget-to-175-billion/4021891.article

CNBC. (2026, July 17). China's Moonshot AI unveils Kimi K3 that rivals OpenAI, Anthropic. https://www.cnbc.com/2026/07/17/moonshot-ai-kimi-k3-model-openai-anthropic-china.html

Communications of the ACM. (2026). Is recursive self-improvement really here? https://cacm.acm.org/news/is-recursive-self-improvement-really-here/

European Commission. (2025, July 28). Horizon Europe to continue beyond 2027 with increased budget and new four-pillar structure. Intellectual Property Helpdesk. https://intellectual-property-helpdesk.ec.europa.eu/news-events/news/horizon-europe-continue-beyond-2027-increased-budget-and-new-four-pillar-structure-2025-07-28_en

Nobel Prize Outreach. (2024). The Nobel Prize in Chemistry 2024. https://www.nobelprize.org/prizes/chemistry/2024/summary/

Prof G Media. (2026, July 24). The week [Audio podcast episode transcript]. Prof G Media.

Science|Business. (2025a, July 16). Commission puts forward €175B budget for FP10. https://sciencebusiness.net/news/planning-fp10/commission-puts-forward-eu175b-budget-fp10

Science|Business. (2025b, July 17). Horizon Europe budget to double, but €68B will remain in Competitiveness Fund. https://sciencebusiness.net/news/planning-fp10/horizon-europe-budget-double-eu68b-will-remain-competitiveness-fund

Scientific American. (2026). Anthropic warns AI may soon begin recursive self-improvement. https://www.scientificamerican.com/article/anthropic-warns-ai-may-soon-begin-recursive-self-improvement/

Silicon Valley Girl. (2026). Interview with Erik Brynjolfsson, Stanford economist [Video podcast episode transcript]. Silicon Valley Girl Podcast.

Stanford HAI. (2026). Public opinion. The 2026 AI Index report. Stanford Institute for Human-Centered Artificial Intelligence. https://hai.stanford.edu/ai-index/2026-ai-index-report/public-opinion

VentureBeat. (2026, July). China's Moonshot AI releases Kimi K3, the largest open-source model ever, rivaling top U.S. systems. https://venturebeat.com/technology/chinas-moonshot-ai-releases-kimi-k3-the-largest-open-source-model-ever-rivaling-top-u-s-systems