"Eric Schmidt's Candid Insights: Why Google Lost in AI and What the Future Holds"

It's quite amusing. Eric Schmidt, who served as Google's CEO for a decade, was recently invited to a conference at Stanford University's Computer Science Department. During his speech, he was quite candid and at one point earnestly told the students that the meeting was confidential and that they shouldn't share what he was saying. However, the organizers informed Eric Schmidt that the conference was being live-streamed with cameras... his expression froze. Although Stanford later removed the video from YouTube, many people had already archived it, and there's a full transcript on GitHub.

Here's a summary of the key points he made:

Why is Google currently losing so badly in the AI field? Because Google values employees going home early and work-life balance more than winning the competition. If your employees only come to the office one day a week, how can you compete with OpenAI or Anthropic?

Look at Elon Musk, look at TSMC. These companies succeed because they push their employees hard. You must push your employees to the limit to win. TSMC makes physics PhDs work on the factory floor in their first year. Can you imagine American PhD students working on an assembly line?

I've made many mistakes myself, such as thinking NVIDIA's CUDA was a dumb programming language. But now CUDA is NVIDIA's most formidable moat. All large models run on CUDA, and only NVIDIA's GPUs support CUDA, an unbeatable combination for other chipmakers.

When Microsoft partnered with OpenAI, I couldn't believe it. How could Microsoft outsource its most crucial AI business to such a small company? Again, I was wrong. Look at how slow Apple is in AI. Big companies have become bureaucratic; the real grinders are in startups.

TikTok taught Americans a lesson. Young entrepreneurs here, if you can steal music or whatever, do it quickly—seemingly mocking TikTok for condoning pirated BGM early on. If you succeed, you'll have the money to hire top lawyers to clean up. If you fail, no one will sue you.

OpenAI's Stargate claimed it needed $100 billion, but it might actually require $300 billion. The energy gap is too big. I suggested to the White House that the U.S. should either strengthen ties with Canada, which has abundant hydropower and cheap labor, or cozy up to Arab countries for sovereign investments.

Europe is out of the game. Brussels (EU headquarters) has been destroying opportunities for technological innovation. France might have some hope, but not Germany, and other European countries even less so. India is the most important swing state among U.S. allies, and the U.S. has already lost China.

Open source is great. Most of Google's historical infrastructure has benefited from open source. But honestly, the costs in the AI industry are too high for open source to bear. The French large model Mistral, which I invested in, will switch to a closed-source approach. Not every company is willing or able to be a patsy like Meta.

AI will make the rich richer and the poor poorer, and the same goes for countries. This is a game for strong nations. Countries without technological resources need a ticket to join the supply chain of strong nations, or they'll miss out on the feast.

AI chips belong to high-end manufacturing, have high output value, but are unlikely to drive employment. Few of you may have visited a chip manufacturing plant; it's all mechanized, doesn't need people. People are slow and dirty, so don't expect a manufacturing revival. Apple's relocation of the MacBook production line to Texas isn't because wages are low there; it's because they no longer need to hire on a large scale.

Historically, electricity did not create more productivity than steam engines when first introduced to factories. It was about 30 years later, when distributed power systems restructured workshop layouts and assembly systems emerged, that productivity soared. Current AI is like electricity back then—valuable, but organizational innovation is needed to reap huge rewards. Right now, everyone is just picking the "low-hanging fruit."