Gavin Baker

“Gemini 3 shows that scaling laws for pretraining are intact. This is the most important AI datapoint since the release of o1. This means that Blackwell models are likely to show a significant increase in performance when they come out in the second quarter of 2026. GPT-5 was not evidence of a slowdown in scaling laws.”

Blackwell-trained frontier models in fact shipped earlier than Q2 2026 — OpenAI's GPT-5.3 Codex (February 2026) was trained and served entirely on GB200 NVL72, and xAI's Grok 4.5/4.6 were built on GB200/GB300 — but whether they represent the 'significant increase in performance' Baker predicted is a judgment call that needs a benchmark-based resolution.

“I think Anthropic is worth $3 trillion today... They're going to end this year with well over $100 billion [in revenue].”

As of September 2026, the New York Times reported Anthropic is on track to exceed a $100 billion annualized revenue run rate by the end of 2026 (up from $65 billion in July) — but annualized run rate is not the same as $100 billion of revenue booked during the year, so resolution will depend on which measure is applied.

“My most contrarian belief: America, at some point over the next four years, will print at least one year of greater than 5% real GDP growth. I think productivity is going to go vertical because of AI and deregulation. I think there could be a world where this is the late '90s.”

Window runs through calendar 2028; no year of >5% US real GDP growth has been recorded so far, leaving 2026-2028 to resolve it.

“So what's the '28 number? Is it $200 billion? Is it $300 billion? ... It will be very profitable at that scale because it'll be inference-dominated and people are reporting they have 85% [gross margins on inference].”

Target date is more than two years out; Anthropic was reported at roughly a $65 billion annualized run rate in July 2026 and on track for over $100 billion by year-end 2026.

“On power, the near-term shortage starts to ease in 2027 and 2028 as new sources of energy come online — and orbital compute solves it in the long term.”

Deadline is end of 2028; resolution will turn on whether US data-center power constraints measurably loosen in 2027-28 as new generation comes online.

“The most important thing in the next 3-4 years is data centers in space. In every way, data centers in space, from a first principles perspective, are superior to data centers on earth. In space, you can keep a satellite in the sun 24 hours a day. The sun is 30% more intense, which results in six times more irradiance than on Earth. So you don't need a battery... Space cooling is free. You just put a radiator on the dark side of the satellite.”

Deadline is end of 2029. Baker restated the thesis in May 2026, saying orbital compute's feasibility and economics would be proven 'within the next two years' and that it would begin capturing meaningful market share by the end of the decade.