Home / AI

Big Tech's AI Spending Is Heading for $800 Billion-Plus in 2026

Three quarters in, the largest cloud and AI spenders have already put $556 billion into data centers, and next year's budgets point higher still.

AIOct 30, 20264 min read2 diagrams

Key takeaways

  • Five big spenders are on track for $803B to $828B of capex in 2026, with $556.2B already spent through three quarters.
  • Amazon held its roughly $220B forecast; Microsoft trimmed to about $170B because third-party capacity slipped.
  • Alphabet's CFO said capex will increase significantly in 2027, so the ceiling is not in sight.

The latest round of earnings calls put a number on the AI buildout that would have sounded absurd two years ago: between $803 billion and $828 billion of capital spending in 2026 across five of the biggest buyers. The money is going to chips, buildings and power, and executives are already signalling more in 2027.

What happened

As the third-quarter reports landed in late October, tallies of company guidance put combined 2026 capital expenditure for Amazon, Alphabet, Microsoft, Meta and SpaceX at $803 billion to $828 billion. Year-to-date spending through the first three quarters already stands at $556.2 billion, so the fourth quarter alone is expected to account for roughly $250 billion to $270 billion.

Company by company, Amazon kept its forecast of about $220 billion in cash capex. Microsoft now expects about $170 billion for the calendar year, down from its previous outlook, and attributed the cut to delays in third-party data center capacity rather than weaker demand. Alphabet's finance chief told investors the company expects capex to increase significantly in 2027. Reports on Meta put its 2026 range at $135 billion to $145 billion, and the stock fell sharply on the combination of heavy spending and negative free cash flow in the quarter.

Bar chart of 2026 capex: spent to date, Microsoft, Amazon, and projected full-year range 2026 capital spending, $ billions 170 220 556.2 803 828 Microsoft Amazon Five, YTD Q3 Full year, low Full year, high Dashed bars are projections; Microsoft and Amazon are annual guidance
Guidance and year-to-date totals as of late October 2026. The five-company total includes SpaceX. Source: Hargreaves Lansdown / Business Insider.

Where the money actually goes

The headline number is not a single bet. Most capex is split between short-lived assets, mainly GPUs and custom accelerators that are written off over a few years, and long-lived ones: buildings, cooling, substations and land. The long-lived part matters because it has to be ordered years ahead and depends on contractors, equipment and utility interconnection, none of which scale at the speed of a chip order.

That is exactly where Microsoft hit friction. Its lower figure was not a retreat; the spend slipped because partners building capacity on its behalf ran late. In an era where the bottleneck is steel, transformers and grid connections, a budget is a ceiling, not a guarantee.

Diagram showing capex flowing into chips, buildings and power, with a bottleneck at the grid Capex budget $803B to $828B Chips and servers Short-lived, bought fast Buildings and cooling Long-lived, slower Power and grid Typical choke point Live capacity Timing can slip
Simplified view of how a capex budget becomes usable compute. Illustrative.

Why it matters

At this scale, AI infrastructure is a macroeconomic variable. Spending of roughly $800 billion a year is large enough to move GDP forecasts, utility planning and chip supply chains. It also shifts the burden of proof: investors are no longer asking whether the money is being spent but whether the revenue arrives fast enough to justify it. Meta's share-price reaction shows how quickly patience can thin when free cash flow turns negative.

The financing question

Spending at this pace tests balance sheets that were once thought bottomless. Meta's quarter shows the pressure: reported revenue of roughly $64.6 billion, up 26%, against quarterly capex near $34.9 billion and negative free cash flow of about $2.1 billion. The market sold the stock by nearly 9% on that combination, which is a reminder that investors reward growth in AI revenue but punish growth in AI cost when the two are out of step.

Earlier in the year, tallies of announced plans were closer to $700 billion, and the climb to more than $800 billion happened while the year was already under way. Each earnings season has brought upward revisions, and the exceptions, like Microsoft's trim, stem from supply-side delays rather than reduced ambition. Alphabet's finance chief going out of her way to flag a significant increase in 2027 is a signal that the planning horizon is lengthening, not shortening.

For suppliers, the pattern is simple. Every dollar of capex is some other company's revenue: chipmakers, memory vendors, power-equipment makers, construction firms and utilities. Nvidia's results a few weeks earlier, with $89 billion of quarterly data center revenue, are the mirror image of the capex totals reported here. If one number turns, the other follows.

What to watch

  • Initial 2027 guidance in January, especially whether Alphabet's promised significant increase is matched by Amazon and Microsoft.
  • How much of the budget is financed by debt and leases rather than operating cash flow.
  • Whether third-party capacity delays, like Microsoft's, spread to other buyers.
  • Cloud revenue growth relative to capex growth, the cleanest read on payback.

Sources

  1. AI hyperscalers are expected to finish the year spending a whopping $828 billion on capex — Hargreaves Lansdown / Business Insider
  2. Bloomberg: Alphabet, Microsoft, Meta Earnings This Week Will Judge Whether $800B AI Capex Pays Off — AI Weekly
  3. Hyperscalers Hit $700 Billion in 2026 AI Spending Plans — Yahoo Finance

More in AI