MIT Technology Review published a piece this week tracing the slowdown of the major computing scaling laws — Moore's, Dennard's, Koomey's, Kryder's and Keck's — and the industry's pivot toward brute-force data center construction. Six of the biggest hyperscalers are on track to spend more than $700 billion this year on computing infrastructure, roughly double the record set in 2025. The article's argument is blunt: the physical limits of computing and communications sit far below anything running in a data center today.
The laws that stopped cooperating
Each of those laws described a different kind of free lunch. Moore's Law delivered more transistors. Dennard's Law kept power density flat while they shrank. Koomey's Law made computation more energy-efficient, Kryder's Law made storage cheaper, and Keck's Law did the same for bandwidth. According to the article, Moore's has slowed, Dennard's has ended, Koomey's has begun to flag, Kryder's has slowed to a crawl and Keck's is flattening.
📊 Market Data Snapshot
That's not a temporary dip. It's the end of an era where each generation of hardware handed operators a cost reduction without anyone having to negotiate a power purchase agreement. The replacement strategy is capital: build bigger, build more, and pay for it.
Feynman's prize, and what it actually showed
The article reaches back to Richard Feynman's December 29, 1959 lecture at Caltech, 'There's Plenty of Room at the Bottom,' delivered at an American Physical Society meeting. Feynman put up two $1,000 prizes for miniaturization challenges. William McLellan won the first about a year later with a quarter-milligram electric motor assembled by hand. Tom Newman won the second in 1985, using electron-beam lithography to shrink a page of text by a factor of 25,000.
The nostalgia is the point. Electron-beam lithography is the same family of technology now pressing against physical limits in semiconductor fabrication. The room at the bottom is nearly full.
The capex bill and the crypto question
There's no crypto asset in the facts, and this isn't a crypto story on its face. But the financing structure matters for anyone holding risk assets. A $700 billion infrastructure cycle is funded by cash flow and debt, and both of those compete with everything else for capital. When the marginal cost of compute is rising instead of falling, the companies that own cheap power and vertically integrated supply chains win, and everyone who built a model on deflationary hardware loses.
For Bitcoin, the second-order read is a mild tailwind to the scarce-supply thesis and a headwind to the idea that crypto is where the next wave of capital lands first. The AI buildout is the magnet. Crypto is a secondary beneficiary at best in the near term, and altcoins that leaned on cheap-compute narratives have a problem.
Mining economics get stickier
The mining sector is the one part of crypto with direct exposure to semiconductor scaling. If next-generation ASICs can't be relied on to rescue margins, the hashprice floor is stickier than most models assume, and miners with the cheapest power contracts separate from the ones banking on hardware efficiency. The article doesn't make that argument. It doesn't have to. The mechanics follow.
What's unresolved is the $700 billion figure itself. The article describes the hyperscalers as on track to spend it this year but doesn't specify the year in the source material, which means it functions as guidance rather than money already out the door. Capex guidance is not capex. If any of the six pulls back, the brute-force narrative reverses quickly, and high-beta crypto assets would feel it first. Watch the next round of earnings calls for the first sign of a cut.
