The 10 Biggest AI Capex Spenders of 2026
A ranked snapshot of who's writing the largest checks for the AI infrastructure buildout this fiscal year, and what each commitment actually buys.
A ranked snapshot of who's writing the largest checks for the AI infrastructure buildout this fiscal year, and what each commitment actually buys.
Disclosure: I hold low-single-digit-percent positions in several supply-chain names that benefit from this capex (CRWV, CEG, IREN, VRT, BE), per the house rules. I do not currently hold positions in AMZN, GOOGL, META, MSFT, ORCL, TSM, INTC, SSNLF, 000660.KS, or MU — the names doing the spending in this article. Figures last refreshed 2026-05-15 against /api/top-capex; updated quarterly.
The largest concentrated capital allocation since the post-iPhone era is happening right now, and the dollar figures are the most reliable signal of where it's headed. Aggregate FY2026 capex guidance from the ten companies below sums to north of $880 billion — more than the entire combined S\&P 500 capex line in 2018. Roughly 80% of that flows directly into AI compute infrastructure (data centers, GPUs, electrical, cooling), and the remainder into the chip and memory fabs that feed it.
Amazon's CEO put it on the record in their most recent 10-K: <cite>"We're not investing approximately $200 billion in capex in 2026 on a hunch."</cite> That sentence is the spine of this article. These checks are deliberate, signed at the top, and sized against expected AI demand 24-36 months out.
Here is the ranking, by FY2026 capex guidance (midpoint), with FY2025 actual for context.
| Rank | Company | FY2026 capex (guidance) | FY2025 actual | YoY |
| ---- | -------------------------------------------- | ----------------------- | ------------- | -------- |
| 1 | Amazon ($AMZN) | $200B+ | $131.8B | ↑ \~52% |
| 2 | Alphabet ($GOOGL) | $175–185B | $91.5B | ↑ \~97% |
| 3 | Meta ($META) | $115–135B | $69.7B | ↑ \~79% |
| 4 | Microsoft ($MSFT) | $120B+ | $64.6B | ↑ \~86% |
| 5 | Samsung ($SSNLF) | $73B | $47.5B | ↑ \~54% |
| 6 | TSMC ($TSM) | $52–56B | $29.2B | ↑ \~85% |
| 7 | Oracle ($ORCL) | $50B | $21.2B | ↑ \~136% |
| 8 | CoreWeave ($CRWV) | $30–35B | $10.3B | ↑ \~215% |
| 9 | SK Hynix (000660.KS) | $29B | $21.0B | ↑ \~38% |
| 10 | Micron ($MU) | $20–25B | $15.9B | ↑ \~42% |
Three observations the table makes obvious:
1. Five names own \~76% of the spend. AMZN, GOOGL, META, MSFT, and ORCL combine for north of $675B in FY2026 guidance — by definition the most concentrated AI infrastructure spend in human history. (For context, the entire US Apollo program in inflation-adjusted dollars was about $260B over a decade. The five biggest hyperscalers will spend more than two and a half Apollos in a single year.)
2. The growth rates are nonlinear. Oracle is more than doubling capex year-over-year. CoreWeave is more than tripling. Even Microsoft and Meta — already running at $65–70B in FY25 — are growing capex roughly 80–85% on top of that. The pace itself is the news, not the absolute numbers.
3. The chip fab layer is the smaller half of the trade. Samsung, TSMC, SK Hynix, and Micron combine to roughly $179B — material, but eclipsed by hyperscaler spend nearly 4-to-1. The marginal AI dollar is going into building data centers and the gear that fills them, not into fabs. That has implications for who you'd want to own as a vendor — more on that in the picks-and-shovels piece.
The honorable mentions
A few names sit just below the top 10 but are worth tracking for different reasons:
- Nebius ($NBIS): $16–20B FY26 guidance, neocloud, FY25 capex was $4.1B — that's a \~4.4x year-over-year ramp at the midpoint, the steepest acceleration on the entire list
- NextEra Energy ($NEE): $12–13B FY26 guidance, the renewables play on grid expansion
- Intel ($INTC): $9.1B, much-discussed but small and shrinking — capex actually cut from $14.6B in FY25
- IREN ($IREN): $5.8–6B in GPU capex, much of it earmarked for the Microsoft deal
- Constellation Energy ($CEG): $3.9B in nuclear and grid capex — small in absolute terms, but the highest-leverage power play on this list
What this number doesn't tell you
The table ranks who's spending. It does not rank who's spending well. Oracle more-than-doubling capex YoY can be a brilliant allocation (if AI cloud revenue follows) or a catastrophic one (if it doesn't). The signal investors should care about is not the absolute dollar number but the conversion of capex into return on invested capital — and that takes 2–3 years to show up in financials.
The ranking I'd actually pay for is the one where the dollars below get re-sorted by capital ROI five years from now. Unfortunately, that ranking does not yet exist. We're working on it.
Where the money lands
If you're a public-market investor wondering how to participate, the direct answer is uncomfortable: most of the capex above flows to private equipment vendors, foreign suppliers, and a handful of public infrastructure companies that don't show up in mainstream tech ETFs. The cleanest list of public beneficiaries is in Top Picks-and-Shovels Stocks for the AI Buildout, which is also packaged as a live watchlist you can track in one click.
The mental model worth holding: the names above are the ones writing the checks. The names in that other piece are the ones cashing them.