$650 billion.
That is Bridgewater’s estimate for what Alphabet, Amazon, Meta, and Microsoft alone will spend on AI infrastructure in 2026. One year. Four companies. Up from $410 billion in 2025.
Most headlines treat that number as proof that AI is real and the investment is justified. Ray Dalio, speaking at the Forbes Iconoclast Summit in New York, read it differently. He said it is what a race looks like when nobody knows who the winner will be and nobody can afford to slow down.
The logic of the AI arms race is straightforward and genuinely difficult to escape if you are one of these companies. If you spend and your competitors do not, you win. If you do not spend and your competitors do, you lose. The rational response for every individual player is to spend aggressively, which means the collective outcome is an industry-wide commitment of capital that may significantly exceed what the market can eventually justify.
Dalio noted that paper wealth in AI-related assets has grown far faster than the actual money supply available to support those valuations. When conditions change, whether through debt pressure, rising interest rates, fund redemptions, or a single geopolitical shock to chip supply chains, the holders of that paper wealth face a moment of reckoning. They need to convert it into money. And the money supply was never large enough to absorb it at the prices currently quoted.
This is the same structure that preceded every major speculative correction. Railway mania in the 1840s. Radio stocks in the 1920s. Dot-com in 2000. The numbers are always different. The mechanism is always the same. Real technology, genuine excitement, capital commitments that outrun the underlying economics, and eventually, a forced reconciliation between paper wealth and actual money.
$650 billion is not a reassuring number. It is a countdown.