Boil the Oceans in Centuries: An Astrophysicist Audits Silicon Valley's Math

Grant Kurzweil his bet. Let energy production grow at his rate, year after year, doubling on the schedule he calls the law of accelerating returns. The oceans boil within a few centuries. The observable universe is exhausted in under 4,000 years. Astrophysicist Adam Becker laid out that arithmetic in a recent Machine Learning Street Talk interview, and it is the spine of his new book [More Everything Forever: AI Overlords, Space Empires, and Silicon Valley's Hack to Escape the Limits of Physics](https://www.hachettebookgroup.com/titles/adam-becker/more-everything-forever/9781541619593/) (Basic Books, 2025).

Becker is not a futurist skeptic by trade. He is the author of What Is Real? The Unfinished Quest for the Meaning of Quantum Physics, a book arguing that physicists' choice of interpretation is not a philosophical footnote but a live scientific question. In the new book and the interview with host Tim Scarfe, he argues that the four canonical Silicon Valley futures, the 2045 singularity, the Mars colony, the AI apocalypse, and the mind upload, are not separate bets. They are one bet, expressed four ways, and the bet is the same perpetual-growth curve that Kurzweil has been selling for two decades.

Kurzweil's "law of accelerating returns" is the load-bearing premise of all of them: the claim that the rate of technological progress, and the compute and energy that drive it, will keep doubling on a Moore's-Law-style curve. Becker says the math is cherry-picked. Pick the right slices of computing history and you can fit an exponential to almost anything, but no exponential runs forever. Physics caps it. The sun will not deliver more than a known fraction of incident energy to Earth's surface, thermodynamics will not let you recycle waste heat, and the universe has a finite mass-energy budget. You can hide the ceiling for a few decades by switching substrates (vacuum tubes to silicon to whatever comes next), but you do not erase it.

Mind uploading requires either emulating a human brain at biological resolution, which the perpetual-energy thought experiment Becker describes would chew through the observable universe on a 4,000-year clock, or pretending the brain is a kind of software that can be lifted off the substrate, which his book argues is a category error. Mars colonies require living in a place where the dirt is perchlorate-laden poison, the atmosphere is too thin to breathe, and the day an asteroid the size of Mount Everest hit the Yucatán was, on most measures, a more hospitable afternoon. Orbital data centers face the same thermal budget, with the added cost of launching every watt of cooling.

Becker does not say AI is a bubble. He says AI is a real technology whose current valuation rests on a physical premise the universe will not honor. The GPU clusters, training runs, and frontier-lab buildouts of 2024 and 2025 treat Kurzweil's curve as fact, and if the exponential does not keep paying, the bet stops being a growth story and starts being a refinancing story, with the people who signed the checks told last. The counterargument he cannot dismiss is that a century of compute doublings is, on its own, a remarkable track record, and that the static-physics assumption baked into his universe-exhaust math is itself a model.

The reframe that runs through the interview, and that Dave Karpf's Substack review and the Kirkus star review both pick up, is that the public AI debate is not really a fight between optimists and pessimists. It is a fight inside one story. The boosters, Yudkowsky, Bostrom, the longtermist cluster, warn that AI will end the world if we are not careful, and in Becker's telling they are not grifters. They are sincere. They are also making the same compute-can-buy-intelligence bet the billionaires making the GPUs are making. The doomer narrative and the booster narrative both require an intelligence explosion, both require it soon, and both require the same physics. The argument the two camps are having is which utopia to build, not whether the underlying exponential will hold.

Regulate the whole tech industry, including the parts that are not yet AI. Tax billionaires out of existence, or at least down to a level where they cannot bankroll their own private physics. Take social problems seriously, starting with climate, disease, and the ordinary work of running a civilization, instead of treating them as preludes to a post-physical future. That is Becker's prescription, and it is the part of the book he wants the reader to walk away with. The futurist bundle is not wrong because the future will be small. It is wrong because the future will be political, and the bundle is a way to avoid that argument.

Karpf's Substack review calls the book a corrective, less a manifesto than a "we should be talking about this" intervention. The Kirkus review reads it as a polemic. Becker would probably take either label, and add that the labeling is exactly the kind of fight he wants the reader to stop having.