In 2025, Microsoft, Alphabet, Amazon and Meta between them spent around $410 billion in capital expenditure. Their current 2026 plans point to a midpoint of roughly $725 billion, an increase of about 77%.
Much of that money is going into the physical guts of artificial intelligence: data centers, the chips inside them, and the power and cooling needed to keep them running.
The number is so large it resists intuition. So the useful question isn’t “is this a lot?” It obviously is. The useful question is: a lot compared to what, spent by whom, and for what return that anyone can point to yet?
What $410 billion actually looks like
Numbers this size only make sense against a familiar yardstick. In a February estimate, Torsten Slok, chief economist at Apollo, put 2026 capital spending by Amazon, Microsoft, Alphabet, Meta and Oracle at around $646 billion, or roughly 2% of US GDP. The estimate predates the latest increases in company guidance, but its comparisons show the scale.
Apollo said $646 billion was roughly comparable to the annual GDP of a country such as Singapore, Sweden or Argentina. It was also higher than the combined military spending of Germany, France, the UK, Japan, Italy and Canada, and about 70% of projected US defense spending of $917 billion.
A handful of firms are approaching the scale at which their equipment budgets are measured against the budgets of nations.
Who is spending what
The spending isn’t evenly split. On current 2026 guidance, Amazon expects about $200 billion, Microsoft roughly $190 billion, Alphabet between $195 billion and $205 billion, and Meta between $125 billion and $145 billion. Together, the midpoint is $725 billion.
What makes the figures feel almost reckless is how large they loom against sales. In a February estimate based on then-current guidance, CreditSights put 2026 capex at about 54% of Meta’s revenue, 47% of Microsoft’s, 46% of Alphabet’s and 25% of Amazon’s. For Meta, projected capex was equivalent to more than half its projected annual revenue. Companies do not spend at that intensity unless they believe something enormous is at stake with CreditSights calling these “Unprecedented”.
Why the number keeps climbing
The logic driving the escalation is competitive, and it is stated openly. Alphabet CEO Sundar Pichai framed the calculation as an asymmetry of regret: “the risk of underinvesting is dramatically greater than the risk of overinvesting for us here.”
If AI matters as much as these firms think, being caught short of computing capacity could be disastrous, while overspending would merely be expensive. When every competitor reasons the same way, the collective bill grows before the eventual demand is fully proven.
What the 75% caveat means
The headline sums are total capital expenditure, not pure AI spending. These companies build for other reasons too.
CreditSights estimated that roughly 75% of aggregate hyperscaler capex in 2026 would fund AI infrastructure, including GPUs, servers and data centers. Applied only as a rough guide to the big four’s current midpoint, that would be about $540 billion.
Whether the math adds up
The spending is real and rising. The returns are the open question.
On the skeptical side, a preliminary 2025 report from MIT’s Project NANDA found that, despite $30 billion to $40 billion in enterprise generative-AI investment, 95% of organizations in its study were getting no measurable return, while 5% of integrated pilots were extracting millions in value. That is striking, but it is not a settled verdict. The report drew on 52 structured interviews, surveys of 153 senior leaders and a review of more than 300 publicly disclosed initiatives, and it acknowledged selection bias, differing success metrics and a six-month observation period that may understate longer-term results.
The valuation worry runs alongside it. In a 2025 note, Torsten Slok contended that “the difference between the IT bubble in the 1990s and the AI bubble today is that the top 10 companies in the S&P 500 today are more overvalued than they were in the 1990s.” That is one economist’s comparative call, not a consensus, but the timing question is real: the capex and depreciation arrive before anyone can know whether the eventual demand will justify them.
Pichai’s asymmetry is persuasive if AI is the defining platform of the era. The MIT finding and Slok’s warning are persuasive if much of this computing capacity is being built ahead of demand that may not materialize on schedule. What we can say is that a handful of companies are making one of the largest concentrated capital bets in corporate history. Whether it resolves into durable value or a historic misallocation is a question no one can answer today.
