Strip away every opinion published about artificial intelligence this year, every prediction, every panic, every conference keynote, and one number remains that cannot be argued with.

​

​

​

​

​

​

That is what Google, Amazon, Microsoft and Meta alone plan to spend on capital expenditure in 2026,  up 77% from last year's already record $410 billion. Goldman Sachs puts total hyperscaler capex from 2025 to 2027 at $1.15 trillion, more than double everything spent in the three years before. For scale: $725 billion is roughly what the United States government spends each year on Medicare.

​

​

Words versus money

Regular readers will recognise the discipline from the Seven Percent Signal. Sentiment surveys told us 60% of businesses expected the economy to worsen; hiring intentions told us what they were actually prepared to bet. When words and money diverge, follow the money.

​

The same test applies here, at a thousand times the scale. Commentary about AI is words. Capital expenditure is money, the least reversible, most scrutinised money a public company spends, signed off by boards, disclosed to markets, and punished savagely when it disappoints. The four richest companies on earth have just made the largest coordinated capital bet in corporate history that machine capability keeps compounding.

​

​

​

​

​

Three implications for everyone else

For anyone running a business several sizes smaller than a hyperscaler - which is everyone - the forecast has three consequences.

​

01 The disruption is pre-paid

This capacity gets built, powered and sold whether or not the hype cycle cools, because the money is already committed. Data centres under construction do not pause for sentiment. Plan for AI capability arriving on schedule, not receding the prudent assumption is now that the technology keeps improving, because three-quarters of a trillion dollars says it will.

​

02 The cost curve only bends one way

Compute this expensive to build must be sold, and the builders need adoption at scale to justify the spend. That means the capability is being pushed into the market at falling prices, every quarter, whatever AI can do for your cost line, it does more of it for less. Businesses that locked in their assumptions about "what AI can do" in 2024 are already pricing off stale information.

​

03 The twelve-month question is a capital question

Whether AI reaches a given business model is no longer a technology debate to have opinions about. It is a capital-allocation certainty looking for its next target. The right question is not if but through which door, your cost line, or your revenue line. The four-question filter for answering that is in The AI Resilience Test.

​

​

What would change the signal

A signal is only useful if you know what would invalidate it. Watch for two things: hyperscalers cutting capex guidance mid-year (the single most honest indicator that returns are disappointing), and the gap between capex growth and cloud-revenue growth widening for more than two consecutive quarters. Neither has happened. Until one does, the bet stands, and so does the forecast.

​

​

​

 

 

 

 

​

​

​

​

​

​

​