Eric Schmidt on Superintelligence: Why Free Chinese AI Is Already Winning

Former Google CEO Eric Schmidt tells Norway’s sovereign wealth fund chief Nicolai Tangen that China’s AI models trail the US by about six months, and that the West is already losing because China gives its models away free. The conversation, on the In Good Company podcast from Norges Bank Investment Management, covers how close AGI really is, why agents will flatten companies, why Europe keeps losing the technology race, the “mini Chernobyl” he thinks it will take to get AI treaties, the threat AI poses to democracy, and the personal cost of never switching off.

TLDW

Schmidt calls AI the biggest technology wave he has seen in five decades, pointing to Anthropic’s revenue growth beating Google’s old record by more than twice the speed. He thinks AI can already know everything humans know but real discovery is still five to ten years away, and that agents plus robots will absorb most business workflows while large companies, especially European ones, resist the cultural change. His worst case is a ruinous US and China race that loses control of the systems, and he expects treaties only after an accident. He closes with a blunt prediction that a 10,000-person company probably has a couple of thousand people it no longer needs.

Thoughts

Schmidt’s most important argument redefines what winning the AI race means. His team benchmarks Chinese models against American ones, and he puts China about six months behind, helped by distillation from US models and smuggled chips. Most commentators would call that a comfortable lead. Schmidt says it means the West is “already losing,” because China’s strategy is to make its models free, and if you count users rather than benchmark scores, a poor country that wants modernity will pick the free Chinese model over the expensive Western one. Tangen adds a twist that sharpens the point: after June 12, people outside the US lost access to the best American models, including Anthropic’s Mythos. A strategy that sells the best models at premium prices to a shrinking set of countries is the opposite of how Android or Google search won the world, and Schmidt, of all people, knows it.

The workforce prediction is the bluntest thing he says, and it is aimed straight at the fund’s portfolio companies. Schmidt describes his old job at Google as sitting in meetings 12 hours a day, and calls that coordination a tax that AI platforms can largely remove. A friend who cut 20% of a fast-growing company laid off the “glue people” who moved plans and numbers around, kept salespeople, regulatory staff and builders, and got death threats for it. Schmidt’s guess for any 10,000-person company is a couple of thousand people it does not need. Earlier he describes the tool that makes this possible, Claude Tag sitting in Slack with a CEO able to have their own agent look at what everyone else is doing, and an executive whose AI clone hands an employee named Vinnie his next task every hour. He admits the second one “sounds pretty abusive.” Put together, the flatter company he predicts is also a far more watched one, and he spends less time on that trade-off than it deserves.

His theory of AI regulation is honest to the point of being grim: treaties happen after an accident. Others in the industry have said they need a “mini Chernobyl,” something that scares everyone enough to regulate without harming many people, and Schmidt says the most likely version is a terrorist attack later traced to an AI technique. He also names the problem he does not know how to solve, open-weight models whose guardrails can be tuned out until they help design weapons, which is China’s chosen strategy. He mentions a breakout about a week before the interview where bots attacked a website largely on their own and were hard to stop. The uncomfortable implication is that someone who sits on the US side of AI talks with China expects safety policy to be written in response to harm, not ahead of it.

On democracy he drops his usual optimism. He was raised to think democracy rested on classic liberal values, but AI and social media can now convince large groups to believe false things and vote against their own interests. The mechanism he describes is specific: trust in institutions is falling, people trust friends, and online friends can be faked at scale. His example is personal. A friend told him data centers were killing cattle, a claim he traced to one farmer with no police or veterinary reports, and he builds data centers himself. Being booed at a commencement speech earlier in the year taught him how hard it is for any individual to push back once a meme takes hold, which is a striking admission from someone who spent a decade running the world’s largest information company.

Key Takeaways

Schmidt ranks AI as the largest technology wave of his career, bigger than PC versus Mac and Android versus iPhone, because it touches everything.
He rejects the bubble argument by pointing to Anthropic’s revenue growth, which he says broke Google’s record at more than twice the speed.
Frontier models take about six months to build and two to test, then are antiquated within five months; he expects the next cycle to be measured in days.
AI can already write all the code (“My job is dead”), but real discovery, such as deriving general relativity from 1902 knowledge, looks five to ten years away.
Most business workflows can be expressed as chains of agents, and combined with robots, could replace much human process work.
The fastest gains are in software, cyber and math, which are “scale-free”; biology and other data-poor fields will move slowest.
Europe is not growing, regulates before it innovates, has underfunded universities and a “tall poppy” culture, and lacks its own foundation model.
He proposed to President Macron that companies under 20 employees face only health and safety rules, to encourage many small bets.
European firms can still win selling compliant AI to Europe, as Mistral does, and ASML proves Europe can build a world-critical, vertically integrated company.
China is about six months behind, but free Chinese models will win most of the world’s users, which Schmidt counts as a loss.
He expects AI treaties only after an accident, and calls guardrail-stripped open-weight models an unsolved problem.
AI-powered fake friends make misinformation easy, which is the one area where he is not an optimist.
Founders return to their companies because this is a historic moment and they want a seat at the table; he wants AI regulated for children “this year, not next year.”
His biggest work failure was Google not getting ahead in social media, and his personal flaw is being too slow.
His advice to young people is to build an agentic system for something they care about, or a foundation model for a field they love.

Chapters

1:22 From Mainframes to AI: The Biggest Wave Yet

Schmidt traces his career through the mainframe, PC, internet and social media eras, and says each platform wave is larger than the last. The ocean is rising because of fundamental discoveries plus scale compute: two-nanometer chips, better memory and data centers too big for humans to grasp. He uses Anthropic’s revenue growth to argue this is not a bubble, and describes models that are obsolete within months of release. “We’re living in a piece of history that people will write about for a thousand years,” he says.

6:03 The Underhyped Part: An AI Assistant for Every CEO

Asked what people underrate, Schmidt points to AI assistants for leaders. Two founders he knows ask their assistant every Friday whether anything in the company conflicts with their values, and another executive built an AI version of himself that assigns work to employees. He notes that ten major math problems thought unsolvable were solved in the last month, and that math is the start of every next revolution.

8:48 How Close Is AGI? The 1902 Relativity Test

Schmidt describes a current “event horizon”: AI that can think for hours, use tools and write all the code, but cannot yet discover something genuinely new. The industry test is whether a model trained only on math known in 1902 could derive general relativity, and today it cannot. He outlines recursive self-improvement and two theories of discovery, random reinforcement learning or applying techniques from one field to another, and estimates five to ten years.

11:22 Agents, Flatter Companies and Scale-Free Domains

Schmidt defines an agent as an input, a chain of reasoning and actions, and a summary, and says almost any professional workflow is a series of agents. He predicts most companies will not make the cultural changes needed, especially in Europe, while Anthropic already writes most of its code with its own models. Gains come fastest in software, cyber and math, where learning loops are quick and no new data is needed, and slowest in biology. Friends such as Demis Hassabis think all human disease could be cured in 15 to 20 years.

17:49 Why Europe Keeps Losing

Schmidt says Europe is a great place to live but is not growing, and a democracy that does not grow starts to fray. He blames underfunded universities, cultural angst about wealth and regulators who act before innovators, contrasting the US AI strategy he helped write with Brussels’ list of prohibitions. Europe has no educational or talent gap, but its talent moves to the US, and without its own foundation model, Europeans will get their history and culture from American or Chinese systems.

21:28 The Worst Case and the Mini Chernobyl

Referring to his book Genesis, Tangen asks what worries him most. Schmidt first sets out the upside of solving climate, energy, disease and scarcity, then names the worst case: a ruinous China and West competition in which humans lose control. He describes a recent incident where bots attacked a website on their own, argues aviation and electricity show safety problems get solved, and expects treaties to follow an accident.

25:59 China Six Months Behind, and the Mythos Ban

Schmidt calls Chinese models fast followers that appear to rely on distillation and smuggled chips, and puts them six months behind. Because Chinese models are free, he says the West is already losing on users. On the June 12 cutoff of top US models outside America, including Mythos, he urges patience with an administration that is “still learning where the power is.” For Europe he sees a home market Mistral can win, proposes lighter labor rules for small companies, suggests training in partner countries with cheaper energy, and points to ASML as proof Europe can build world leaders.

31:10 AI and the Threat to Democracy

Schmidt says he is an optimist except on democracy. AI and social media let skilled operators convince people that false things are true, using fake online friends in a world where trust in institutions is falling. He tells the story of a friend convinced that data centers kill cattle, and reflects on being booed during a commencement speech, which ironically got his speech read more widely.

34:38 Founders, Leadership and the Coordination Tax

Schmidt says the myth of the founder is real: professional CEOs like him lack the control that let Steve Jobs remake Apple or Mark Zuckerberg pivot from virtual reality to AI. Founders return because they want a seat at the table for a historic period, and he insists AI must be regulated for children now. Leaders today need raw processing speed, and AI will remove much of the coordination tax inside large companies, leaving them with far fewer staff.

41:13 Failure, Never Switching Off and Advice to Young People

Schmidt names Google’s missed lead in social media as his biggest failure and says, personally, he is too slow. He flies helicopters and planes and races cars, but says he cannot switch off and that it is not healthy. He is running four new companies to create wealth for science, inspired by the single VAX 11/780 that powered his Berkeley work at 22. His advice to young people is to build agents or a foundation model for a field they care about, and not to fear the word superintelligence.

Notable Quotes

“I was a professional programmer. My job is dead, right?”
Eric Schmidt, on AI that can write all the code

“Run as fast as you can. Align it with human values and if bad things happen, fix them immediately.”
Eric Schmidt, summarizing the US AI strategy he helped write, versus Europe’s

“The way treaties happen is there’s an accident.”
Eric Schmidt, on how AI regulation between the US and China will come about

“If they’re six months behind, then we’re already losing because their strategy is that the products are free.”
Eric Schmidt, on Chinese AI models

“You just build fake friends and then you befriend everyone and then everyone believes you.”
Eric Schmidt, on how AI makes misinformation campaigns easy

“My guess is you’ve got a couple of thousand people who you don’t need.”
Eric Schmidt, to anyone running a 10,000-person company

“I want to scale science.”
Eric Schmidt, on why he is building four new companies

Watch the full conversation here.