Just hold on to one number: in 2026, Google plans to spend as much as $190 billion in capital expenditure (CapEx), roughly six times its 2022 level. This is one of the largest single-company investment programs in the history of tech, almost all of it poured into AI. If you want to get to know Google as a company first, see What Kind of Company Is Google.
This piece unpacks three things: where the money goes, what the market is worried about, and why this ledger is deliberately kept a little out of focus.
The one-line framing: Google can absolutely afford this spending; the real bet is whether the returns on it arrive fast enough.
How Big This Spending Is
Lay out the capital expenditure of the past few years, and the curve is frighteningly steep.
| Year | Capital Expenditure |
|---|---|
| 2023 | About $32.3 billion |
| 2024 | About $52.5 billion |
| 2025 | About $91.4 billion |
| 2026 (estimated) | About $180–190 billion |
At the 2026 I/O conference, Google CEO Sundar Pichai put it bluntly: the company’s annual capital expenditure was about $31 billion in 2022, and 2026 is projected to be six times that. In the first quarter of 2026 alone, it spent about $35.7 billion. Nearly doubling for two years running — that acceleration is itself a signal.
Where the Money Goes
On the earnings call, the CFO gave a rough breakdown: roughly 60% to servers, 40% to data-center construction and networking equipment.
Within that 60% on servers sit a mix of in-house TPUs, purchased Nvidia GPUs, and in-house Axion CPUs, but Google did not break out the amounts for each. The 40% on data centers builds the facilities that house these chips, plus power, cooling, and networking. One thing worth calling out on its own is power: this much compute consumes a staggering amount of electricity, so in early 2026 Google spent about $4.75 billion to acquire Intersect Power, a clean-energy and data-center infrastructure company — the goal being to lock down a power source first, so that expansion doesn’t get stuck for lack of electricity. When you have to buy up the power supply in advance, you can see the sheer scale of this infrastructure race.
What the Market Is Worried About
The flip side of colossal investment is colossal doubt, and this part has to be spelled out — you can’t report only the good news.
The core worry is cash. Google’s Q1 operating cash flow was handsome, but after subtracting capital expenditure, free cash flow for the quarter was only about $10.1 billion. The CEO of research firm SemiAnalysis publicly predicted that if full-year capital expenditure really hits $180 billion, Google’s free cash flow next year will likely approach zero — and said the market isn’t ready to face it. Interestingly, he said this in a bullish tone, arguing that whoever dares to spend like this is the top-tier capital allocator. Read this in a balanced way: it’s a third party’s forecast, not Google’s official line, and Q1 actual free cash flow was still positive.
The second worry is depreciation. The data centers built and the chips bought get gradually amortized into expenses over the coming years. Q1 depreciation expense already rose about 45% year over year, and the CFO warned this will keep compressing the bottom line. Some research firms have even slapped Alphabet with a “selective” rather than “high-confidence” rating, on the grounds that capital expenditure eats into free cash flow and that the market questions its discipline on investment returns. On top of that, Google’s own advertising network revenue has declined for several quarters running (down about 4% year over year in Q1), and some analysts believe this is tied to the structural shift in which AI Overviews intercepts traffic and reduces outbound clicks. These are all real headwinds.
The Other Side: The Demand Is Real
Having laid out the doubts, you also have to lay open the other side, or it’s equally lopsided.
What underpins this gamble is visible demand. Google Cloud’s Q1 revenue grew 63% year over year, and Pichai said outright that the company is “compute-constrained” — if it could supply more, cloud revenue would be higher. The most-watched figure is the cloud order backlog, which nearly doubled in a single quarter to $462 billion. But there’s one definitional point you have to understand first: the CFO explained that this surge newly includes the TPU hardware sales contracts Google began offering for the first time — not purely organic growth in cloud services, though most of the backlog is still cloud contracts. Factor in that change in definition, and the number won’t get misread as “demand doubled in a single quarter.”
There are other demand signals too: search advertising, driven by AI Overviews lifting query volume to new highs, grew 19% year over year in Q1; and within cloud, revenue from products built on Gemini grew nearly eightfold year over year (though Google did not disclose the base amount). Google has even begun selling TPU hardware directly to a small number of large customers, effectively extending from “renting out compute” to “selling chips.” After the earnings release, Google’s stock rose about 7%, a contrast to the falling stock prices of other big spenders in the same period, which shows the market’s acceptance of Google’s ledger is relatively high.
Why This Ledger Is Out of Focus
Reading Google’s AI finances comes with a built-in obstacle: it deliberately keeps you from seeing clearly.
Alphabet’s financials carry no standalone line for an “AI business.” AI’s contribution is broken up and hidden inside the growth of search advertising and the growth of cloud. The company is happy to report growth rates (cloud generative-AI product revenue growing nearly eightfold sounds impressive), but it doesn’t report the absolute amounts or the base, so there’s no way for outsiders to know how small the starting point of that “eightfold” was or how large it grew to. This isn’t a practice unique to Google — most of the big players do it — but it does turn “how much AI actually earns for Google” into a question you can only estimate and never calculate precisely. When you read any pretty AI growth rate, remember there’s a base missing behind it.
What Google Hasn’t Explained
Plenty of key details on this topic were left blank by the company, and it’s only honest to flag them:
- The line items within capital expenditure: only “60% servers, 40% data centers and networking” is disclosed; how much went to TPUs, Nvidia GPUs, and CPUs respectively is not broken out.
- AI’s absolute contribution: only growth rates, no base amount, so there’s no way to quantify the revenue AI creates on its own.
- The point at which free cash flow hits zero: the third-party forecast uses “next year,” which is semantically ambiguous as to whether it means 2026 or 2027, and Q1 actuals were still positive.
- 2027 capital expenditure: management said it will be “significantly higher” than 2026, but gave no specific number.
Penna’s Take
Pull this $190 billion ledger back into the wide view, and what it really tests is a question about time.
Google’s logic is clear: if it doesn’t pour money into building compute now, it won’t have a ticket in the future — and the cloud’s $462 billion order backlog proves the demand is real. The market’s worry is just as clear: depreciation and power bills will eat into profit and cash one step ahead, while AI’s returns take time to show. These two things don’t contradict each other; they can be true at the same time, the only difference being which arrives first. So the question to really keep an eye on is concrete: over the next few quarters, can the pace of monetization in cloud and search keep up with the pace of depreciation and capital expenditure. The answer to this question will be revealed quarter by quarter through the earnings reports, and right now no one has the final word.
Further reading: What Kind of Company Is Google, The Cost Moat of In-House TPUs, Where the Google Antitrust Case Stands.