Peter Thiel: Artificial Intelligence Will Be Much Less Monopolistic Than the Internet

Peter Thiel, while receiving the Axel Springer Award and in a long interview with Mathias Döpfner, described artificial intelligence as a technology that could be larger than the internet and has the first opportunity to accelerate growth since the 1970s; at the same time, he has long characterized this round as more difficult to harvest a single tax booth than the search era. He summarized publicly: artificial intelligence will be much less monopolistic than the internet. His fixed analogy over the past two years has been that AI in 2024 is like the internet in 1999: technically significant, risky for investment, with Amazon dropping from $113 to $5.5 and taking eight years to return to its starting point.

In the Döpfner program, he reversed the stagnation proposition: progress over the past few decades has concentrated on bits rather than atoms, with the internet contributing about 1 percentage point to GDP annually for ten to fifteen years, but not enough to elevate civilization to the next level; artificial intelligence might be able to. He does not deny extinction risks, stating that 5% to 10% is "already quite bad," while opposing the idea of writing deceleration as a neutral choice—he believes a zero-growth world will not automatically become a peaceful social democracy but will require a world government worse than disease. He referred to the anti-AI sentiment in the U.S. as mainly "what we are doing to ourselves," and it should not be simply blamed on China.

This does not cancel out his statement in "Zero to One" that "competition belongs to losers, and one should create creative monopolies." He has pointed out that Google Search has extracted high profits since 2002 after shaking off Microsoft and Yahoo, while the airline industry creates greater value but struggles to retain it. He described artificial intelligence as having a different structure: multiple labs operating in parallel, open-source weights that can be replicated, and a leading model window measured in months, lacking the Google-style "one entry point that collects 95% of advertising" tax authority. He has invested in both OpenAI and Anthropic-related assets, and has turned Palantir into an operating system for government and enterprises, indicating that the monopoly he desires is more likely to appear in workflows and data permissions rather than recreating a universal search box.

In 2019, he characterized artificial intelligence as leaning towards authoritarianism, with "a few large companies, a few large governments, and a few large computers managing everything." The 2026 perspective has changed to: the technology may be larger, but the market structure is more fragmented. The intermediate variable is that training costs remain high, but reasoning and fine-tuning allow the second place to survive; cloud and chip oligopolies are collecting rents at the lower level, while the application layer has yet to see a cash machine equivalent to search advertising.

In terms of market mechanisms, this marks the "model competition" as a competitive industry and "search advertising" as an unreplicable rent. The buyers are venture capitalists looking to make vertical applications in a multi-model coexistence; the sellers are labs still hoping to use capital expenditure to create the last universal entry point. Funds continue to flow to Anthropic, xAI, cloud, and custom chips, but valuation discounts arise from "not being able to price attention like Google." Beneficiaries are companies that can sell workflows on fragmented models, as well as infrastructure providers selling electricity, cabinets, and GPUs; under pressure are those treating ChatGPT as the next search monopoly. Less monopoly does not mean less spending, but rather that profits are harder to concentrate in one profit statement.

Source: Public Information