Big Tech Will Spend $600B on AI in 2026 and Barely Bought Anyone Building It
PitchBook’s Q2 2026 industry analysis quantifies what individual earnings reports have signalled all year: the four largest US hyperscalers are on track to spend roughly $600 billion on AI infrastructure in 2026, making this the largest single-year capital deployment in tech history.
What the headline obscures is a structural shift in how that capital is being deployed.
Building, Not Buying
Acquisition activity by the Big Five — Google, Microsoft, Amazon, Apple, Meta — hit a decade low of just seven deals in 2024. It recovered modestly to 14 in 2025 and sits at 12 YTD in 2026. These are not the acquisition volumes you would expect from companies pouring $600 billion into an emerging technology.
PitchBook’s framing is precise: Microsoft’s roughly 30% stake in OpenAI and Meta’s $14.3 billion Series G into Scale AI are not acquisitions. They are infrastructure guarantees — long-duration, high-dollar commitments that lock in compute supply, model access, or data pipeline relationships for years. They behave like capex on the balance sheet. They are not consolidating ownership of the companies involved.
What $600B Actually Buys
The $600B is almost entirely hardware and real estate: data centre construction, GPU procurement, power infrastructure, and long-term cloud compute contracts. Amazon locked $100 billion in AWS spend through its Anthropic deal. Google committed $40 billion in Anthropic equity plus a five-year $200 billion cloud contract. Microsoft is running $190 billion in AI compute spend through 2026 alone.
At that scale, acquiring a $5 billion or $15 billion AI startup is economically trivial and strategically complicated. The startups that matter — Anthropic, OpenAI, Scale AI — are too large to absorb cleanly, too strategically sensitive to fully control, and in some cases explicitly structured to prevent it. So hyperscalers buy large minority positions and sign decade-long supply deals instead.
The Exit Market Implication
The practical consequence for AI companies is that strategic acqui-hire activity, historically the primary exit path for frontier AI teams, has dried up in favour of partnership structures. The decade-low M&A figures confirm hyperscalers are not looking to consolidate capability through ownership — they are renting it at very high prices through compute and API arrangements.
That dynamic sustains the multi-lab model race but sharply reduces the probability of a major consolidation event. The companies routing $600 billion through data centres have already chosen their partners, and those partners are not for sale.