United States – October 19, 2026 -- Six of the largest hyperscalers are on track to spend more than $700 billion on computing infrastructure this year, about double the record set in 2025, as gains from shrinking transistors fade and performance is bought by building bigger systems.
Brute-force scaling replaces transistor shrinkage as the main source of progress
The spending goes into new data center sites faster than the electrical grid can feed them. Progress now comes mostly from adding chips, memory, power and cooling rather than from smaller transistors. In AI, training compute has doubled roughly every six months since 2010, far faster than Moore's Law allows, and little of that growth comes from shrinking.
Five scaling laws that drove the information age are slowing or have ended
Moore's Law, which tracks transistor density, has slowed. Dennard scaling ended around 2004, when chips could no longer shed the heat of running every transistor flat out. Clock speeds, which had passed three gigahertz, stopped climbing, and Intel canceled its promised four-gigahertz processor. The industry moved to multiple cores per chip.
Koomey's Law, which tracks computation per joule, has begun to flag. Kryder's Law for magnetic storage density has slowed to a crawl, and Keck's Law for the capacity of a single optical fiber is flattening. Each new chip generation arrives more slowly and pays back less in speed and power than the one before.
Transistor density has risen 10 orders of magnitude since the first transistor
More than 100 million transistors now fit in a square millimeter, and 100 billion or more fit on a chip. Computational speed has grown 12 orders of magnitude. Energy per operation has fallen 17 orders of magnitude, from about 100 joules for an early vacuum-tube machine to about a femtojoule for a single logic-gate switch.
Gordon Moore's 1965 observation concerned cost, not physics. At that time, 50 components per chip minimized the manufacturing cost of each one. Too few wasted the chip, and too many multiplied errors and collapsed yields.
Physics sets limits far below today's hardware
The scaling laws are observations about engineered systems, not fundamental physical laws. The limits set by quantum mechanics, general relativity and information theory lie far below anything in a data center today. MIT physicist Seth Lloyd calculated the ceiling in a 2000 Nature paper, "Ultimate physical limits to computation." He imagined a kilogram of matter, about the mass of a laptop, packed into one liter.
Under E = mc2, that kilogram holds about 9 x 10^16 joules. That is roughly what the entire United States draws from its electrical grid every two days. Speed in such a machine is governed by this energy.
Some of these limits may be out of practical reach. Proof-of-concept systems that approach them could still push stalled technology paradigms forward while adding to fundamental science.