The $700 billion AI bill
Despite lingering doubts, tech giants are accelerating their investments in AI
Questions about a potential bubble surrounding generative artificial intelligence are doing little to curb the ambitions of US tech giants. Quite the opposite: they are preparing to double down on investment. This year, Google, Amazon, Microsoft, and Meta are expected to rack up nearly $700 billion in capital expenditures — better known as capex — an increase of 60% compared with 2025. A level of investment that is “justified and sustainable,” according to Jensen Huang, CEO of Nvidia, the main beneficiary of this spending frenzy.
These colossal sums are primarily being poured into expanding computing power, through the purchase of hundreds of thousands of graphics processing units (GPUs) used to train and run AI models. It is a full-blown arms race to avoid missing the promised technological revolution, one that is beginning to seriously worry financial markets. Not only is the profitability of these investments far from guaranteed, but their scale is now approaching, or even exceeding, the cash generated by Google, Amazon, Microsoft, and Meta.
Four times more than in 2023
In detail, Amazon expects capex of $200 billion. Google is close behind, with projected spending between $175 and $180 billion. Meta is planning investments of $115 to $135 billion. Microsoft, for its part, has not issued formal guidance, but based on the growth rate observed at the end of 2025, its spending could reach $165 billion. These investment levels are unprecedented and unmatched. By comparison, Apple, whose AI infrastructure remains limited, will make do with just $14 billion in capex.
Together, the four companies are expected to spend at least $655 billion in 2026. That is almost four times more than the combined capital expenditures of the 21 largest non-technology U.S. companies — automakers, telecom operators, oil companies, or even Walmart — according to data compiled by Bloomberg. The surge is striking. In 2023, just after the splashy launch of ChatGPT, their capex totaled only $154 billion. It rose to $251 billion the following year, before reaching $412 billion in 2025.
Supporting new uses
To justify such massive spending, these companies point to the immense potential of generative AI. “The most important technology in history,” according to Mark Zuckerberg. In the face of competition, the main risk is not “wasting a few hundred billion dollars,” the Meta founder argues, but “moving too slowly.” From this perspective, the priority is to build the largest possible computing infrastructure, capable of supporting the expected explosion in usage, particularly the rise of AI agents, which are set to automate a growing number of tasks within companies.
The stakes differ from one group to another. At Amazon and Microsoft, capital expenditures are mainly directed toward their respective cloud offerings, AWS and Azure, to meet growing customer demand and provide access to the latest GPUs. At Meta, the proliferation of data centers is meant to support internal needs, as its billions of users increasingly adopt AI-powered features, for example through smart glasses. At Google, the strategy lies somewhere in between: strengthening its cloud services while also feeding its Gemini model.
Underestimated depreciation?
Although all four giants continue to post massive profits, the scale of their investments is starting to raise concerns. Next year, Amazon’s capex is expected to exceed its operating cash flow, while it will come close to doing so for Google, Meta, and Microsoft. In other words, nearly all the cash generated by advertising, e-commerce, or cloud services will be plowed back into data centers and GPUs. To raise liquidity, Meta issued $30 billion in bonds at the end of 2025. Amazon, which has just laid off 30,000 employees, has not ruled out following suit.
Of course, none of these companies is at risk of financial collapse. The real question is how much revenue AI will actually generate. All the more so because GPUs have a limited lifespan, as ever more powerful models come to market. This raises the issue of depreciation, the accounting principle that spreads investment costs over time. Investor Michael Burry accuses these companies of underestimating it, a practice that would artificially — and temporarily — limit the impact of capex on profits.