Big Tech Spent $700 Billion on AI This Year. The Money Came From You.

In Q4 2025, Alphabet, Microsoft, Amazon, and Meta spent a combined $12.6 billion on stock buybacks. That's the lowest figure since Q1 2018. In 2021, those same companies returned $149 billion to shareholders in a single year.

The money didn't disappear. It moved.

Amazon, Alphabet, Microsoft, Meta, and Oracle have committed roughly $700 billion in capital expenditure for 2026 — almost entirely for AI infrastructure. That's 2.1% of US GDP. One year. Five companies. Data centers, custom chips, cooling systems, and the electrical substations to power all of it.

Amazon leads at $200 billion. Alphabet committed $175-185 billion. Microsoft pledged $145-150 billion. Meta announced up to $135 billion. Oracle rounded it out at $42 billion.

Here's the problem. Those five companies generated $575 billion in operating cash flow in 2025. They're spending $700 billion. The $125 billion gap is being covered by cash reserves and debt. Meta issued $30 billion in bonds in October. Oracle raised $25 billion. Amazon added $15 billion. These companies are borrowing to build AI infrastructure they haven't figured out how to monetize at the scale they're building it.

Amazon's situation is the starkest. Morgan Stanley projects negative free cash flow of $17 billion for Amazon in 2026. Bank of America's estimate is worse: negative $28 billion. Amazon hasn't bought back a single share since Q2 2022. Every dollar goes into the buildout.

Meta's buybacks dropped from $8.8 billion in Q3-Q4 2024 to $3.3 billion in the same period of 2025. Alphabet's fell from $30.6 billion to $17 billion. The trend is directional and accelerating.

Ohsung Kwon, senior strategist at Wells Fargo, summarized the shift: "Currently, AI development capability and opportunity monetization are considered more important than shareholder returns."

That sentence should make shareholders nervous. Not because the companies are wrong about AI's importance. But because "opportunity monetization" is banker-speak for "we haven't figured out the revenue yet."

Goldman Sachs forecasts AI infrastructure investment will exceed $500 billion this year globally. Morgan Stanley raised its hyperscaler capex estimate to $740 billion, up from $570 billion two months earlier. These aren't gradual adjustments. The estimates keep climbing because the companies keep announcing larger commitments, each one trying to signal they won't be the one that underinvested.

The market noticed. Since January 2026, the combined market capitalization of Apple, Alphabet, Nvidia, Amazon, and Microsoft has dropped by over $1.3 trillion. Investors aren't opposed to AI spending. They're opposed to AI spending without proportional revenue growth. Meta's Q4 revenue grew 21%. Its capex grew 94%. That math doesn't work forever.

The last time tech companies chose infrastructure over buybacks at this scale was the fiber optic buildout of 1999-2001. That infrastructure eventually proved useful — it just bankrupted the companies that built it first. The survivors bought the cables for pennies.

There's a version of this story where $700 billion in AI infrastructure creates the compute layer for a new economy. There's another version where it creates the largest overcapacity event in the history of corporate capital allocation. The companies themselves don't know which version they're in. They're building anyway because the cost of being wrong about building too much feels lower than the cost of being wrong about building too little.

The money has to come from somewhere. For the first time since 2018, it's coming from shareholders.

---

If you work with AI tools daily, I built a set of prompt packs that actually work — tested across Claude, GPT-4, and Gemini. System prompts, code review chains, data extraction templates, and more.

👉 Browse the prompt packs on Polar.sh — individual packs from $5, or get all 5 for $19.