Four companies — Amazon, Microsoft, Alphabet and Meta — plan to spend a combined $725 billion on capital expenditure in 2026. Almost all of it goes to one thing: the data centres, chips and power needed to build artificial intelligence.
That's a steep jump from the $410 billion the same four spent in 2025 — itself a record. This is the largest coordinated building spree in the history of technology, and analysts already expect the combined figure to top $1 trillion in 2027.
The number is hard to picture until you hold it against a whole economy. $725 billion in a single year, from four companies, is more than the entire annual economic output of most countries on Earth — approaching the yearly GDP of Switzerland, and larger than that of Sweden, Belgium or Argentina.
The money is split unevenly:
- Amazon — ~$200 billion, the most aggressive of the four
- Microsoft — ~$190 billion, with $25bn of that blamed on rising memory-chip prices
- Alphabet (Google) — ~$180 billion, funding both its cloud and its Gemini models
- Meta — ~$125 billion, after raising guidance on AI demand
Those are each company's own guidance; the $725 billion headline is an analyst tally that runs a little above their sum.
What makes the number remarkable isn't just its size — it's the bet behind it. None of these companies yet earns AI revenue close to what they're spending. Microsoft's AI business runs at roughly a $37 billion annual rate; the capex dwarfs it. Each firm is wagering that falling behind on computing power is more dangerous than overspending on it.
For investors, that tension is the whole story. When these companies report earnings, markets no longer reward the spending — they punish any hint that the payoff is slipping further away. Meta's stock fell 6% the day it raised its AI budget.
The scale is almost abstract: three-quarters of a trillion dollars, in one year, on machines most people will never see, to build something whose return nobody can yet measure.
That is what conviction looks like with a balance sheet attached.