Eric Schmidt
Related people
Episodes
Notes
Masters of Scale · Eric Schmidt scores US vs. China: who’s winning the AI race?
-
Is the US-China AI Race a Zero-sum Game?
Eric Schmidt tells Bob Safian the AI race is not a zero-sum game, and that China is not an enemy. Competitors can still create a whole new market. The best case he describes is a practical conversation between US and Chinese elites about how to manage the technology. He says that conversation is not possible in the current political environment.
One rule he would put on the list is no surprises. In his Pentagon work, a country that launches a missile notifies the others. They check that the launch matches what was announced, and they train their own systems on it. Those decisions were made 50 years ago so a launch is not a mystery. He wants a similar list of collaborations aimed at human fl
-
Why Does Sending Chinese PhD Students Away Hurt the US?
Eric Schmidt gives the United States the edge on leadership and talent, then says the country is giving that edge away. Taking the smartest people, arresting them, kicking them out, and denying them visas is a bad strategy. The pattern he calls out is graduate students, typically from China and from other countries: the United States gives them a PhD and then makes them leave.
One person who wanted to stay in America now leads China's quantum program. Schmidt says the quantum adversary is China and that person. The country trained him, then kicked him out, and he went on to lead the rival program. Schmidt still scores leadership and talent for the United States. The loss, as he tells it, is
-
Why Did China Choose Open-source AI Models?
Eric Schmidt says China chose open-source and open-weight models because it was behind and needed users quickly. America's large models are mostly closed. Training one model can cost a hundred million to five hundred million dollars, so a closed lab asks why it would give that work away. Open source, as he describes it, builds a platform other people build on. Closed source aims at the end user. Open source has largely won in America as the foundation people build on, while the largest models stay closed.
China was willing to take the risk that others would copy the weights, and he says that bet looks like it is working because the projects are getting funded. Reflection AI has released a 5
-
What Did AI Do to Math in One Month?
Eric Schmidt says mathematicians face Armageddon: everything in their world was overthrown in one month. The world of mathematics had not changed in 50 years since Gödel's hypothesis. Mathematicians have an existential crisis, and the crisis is going to occur because of the frontier models. Everything worries him. One model of what will happen is frontier models dominating achievements in math and chemistry.
Using a cheaper Chinese model is likely to be legal in the U.S. American CEOs are patriotic to a point and very focused on money. Welcome to capitalism. People use small servers with open source weights to solve problems because the token cost is lower. Token costs run about three to fi
-
How Much Power Does American AI Need?
Eric Schmidt says American AI needs somewhere between 60 and 80 gigawatts of power to get through 2030 and 2031. One gigawatt costs about $50 billion and is about one nuclear power plant, and new plants are not going up at a high rate. He leaves open whether the new power can be financed.
China has infinite energy, most of it renewable, a buildout the United States said it would do 10 years ago and failed to do. Interconnection queues do not run seven years, and China does not have the kind of debates. Data centers are going up at massive scale. Even with a much more restricted financial market, China has funded the startups. DeepSeek is seeking funding and just did a huge raise. The fundin
-
How Far Behind Is China in the AI Race?
Eric Schmidt says China is now six months behind in the race to build reasoning and intelligence systems. A year ago or two years ago, China looked a year or two behind. He was part of the group that put chip restrictions on China and assumed the limits would work for about 10 years. Chinese teams still figured out the algorithms and knowledge with essentially one-tenth of the energy and some factors less of performance.
Many studies show the six-month gap. Whether the gap will get slower or bigger is a huge debate in the industry. At the moment the contest is a two-horse race. The race for intelligence is the race of a millennium, and children, grandchildren, and great-grandchildren will a