Amazon And Google Are Winning The Ai Capex Race — But What’s The Prize?

Sometimes, it tin look for illustration nan AI manufacture is racing to spot who tin walk nan astir money connected information centers. Whoever builds nan astir information centers will person nan astir compute, nan reasoning goes, and frankincense beryllium capable to build nan champion AI products, which will guarantee triumph successful nan years to come. There are limits to this measurement of reasoning — traditionally, businesses yet win by making more money and spending less — but it’s proven remarkably persuasive for ample tech companies.

If that is nan game, Amazon does look to beryllium winning.

The institution announced in its net connected Thursday that it projects $200 cardinal successful superior expenditures passim 2026, crossed “AI, chips, robotics, and debased world orbit satellites.” That’s up from nan $131.8 cardinal successful capex successful 2025. It’s tempting to property nan full capex fund to AI. But dissimilar astir of its competitors, Amazon has a important beingness plant, immoderate of which is being converted for usage by costly robots, truthful nan non-AI expenses aren’t truthful easy to activity away.

Google is adjacent behind. In its net connected Wednesday, nan institution projected betwixt $175 cardinal and $185 cardinal successful superior expenditures for 2026, up from $91.4 cardinal nan erstwhile year. It’s importantly much than nan institution spent connected fixed assets past year, and importantly much than astir of its competitors are spending.

Meta, which reported past week, projected $115 to $135 cardinal successful capex spending for 2026, while Oracle (once nan poster kid for AI infrastructure) projects a measly $50 billion. Microsoft doesn’t person an charismatic projection for 2026 yet, but the astir caller quarterly fig was $37.5 billion, which pencils retired to astir $150 billion, assuming it keeps up. It’s a notable increase, and 1 that has led to investor unit connected CEO Satya Nadella — but it still puts nan institution successful 3rd place.

From wrong nan tech world, nan logic present is simple. The revolutionary imaginable of AI is going to move high-end compute into nan scarce assets of nan future, and only companies that power their ain proviso will survive. But while Google, Amazon, Microsoft, Meta, Oracle and others are frantically prepping for nan compute godforsaken of nan future, their investors aren’t convinced. Each institution saw its banal value plummet arsenic investors balked astatine nan hundreds of billions of dollars being committed, and companies pinch higher spends tended to driblet more.

Crucially, this isn’t conscionable a problem for companies for illustration Meta that haven’t figured retired their AI merchandise strategy yet. It’s everyone — moreover companies for illustration Microsoft and Amazon pinch a robust unreality business and a straightforward return connected really to make money successful nan AI era. The numbers are simply excessively precocious for investor comfort.

Investor sentiment isn’t everything — and successful this case, it whitethorn not do overmuch to alteration nan industry’s mind. If you judge AI is astir to alteration everything (and nan statement is beautiful compelling astatine this point), you’d beryllium a fool to alteration people conscionable because Wall Street sewage jumpy. But going forward, large tech companies will beryllium nether a batch of unit to downplay really costly their AI ambitions really are.