On the day Oracle fell 12% after announcing its AI investment, semiconductor equipment stocks rose 8%. It was the same AI spending news — so why did one side rally while the other sold off? Who is actually making money from the AI boom?
2026-06-12
The AI Investment Paradox: Those Who Spend vs. Those Who Sell the Tools
Oracle disclosed yesterday that it poured $55.7B into capital expenditures for FY2026 alone — financing $43B of that through debt, with free cash flow (FCF) swinging to a deficit of $23.7B (Reuters / TheStreet, 2026-06-11). The market's -12% reaction is straightforward: companies that spend money building AI infrastructure and companies that sell the tools used to build it have fundamentally different profit structures.
Who Spends and Who Sells
Microsoft, Alphabet, Amazon, Meta, and Oracle — the Big Five hyperscalers — have committed a combined $660B–$690B to AI infrastructure investment in 2026, roughly 36% more than 2025 (CreditSights / Futurum, 2026). Of that, approximately 75%, or around $500B, flows into GPUs, HBM memory, networking, data centers, and power systems (AI Capex Cycle CFA Analysis, alcapitaladvisory.com).
Looking at the supply-chain beneficiaries of that spend makes the picture clear:
| Layer | Representative Stocks | Benefit Mechanism |
|---|---|---|
| Semiconductor equipment | LRCX, AMAT, ASML | Direct expansion of wafer fab equipment (WFE) demand |
| HBM memory production | SK Hynix, Micron | Rising HBM content per AI server |
| Power & cooling infrastructure | Vertiv, Eaton | Surging data center power consumption |
| Cloud platforms | Microsoft, Google | Long-term monetization potential, near-term FCF pressure |
| AI software | Still few players | Monetization models yet to be proven |
Lam Research (LRCX) raised its 2026 WFE outlook to $140B and holds more than 35% share of the etching and deposition equipment market essential for HBM production — with a three-year return of 321% (TradingView/Zacks, 2026). Applied Materials (AMAT) surged +8% after announcing a next-generation AI packaging partnership with Broadcom.
Oracle's Problem Is FCF Structure, Not CapEx Scale
Understanding the investor exodus from Oracle requires examining the debt structure. Microsoft and Alphabet fund a significant portion of similar AI spending from internal cash flows, but Oracle saw its FCF deficit balloon from just $394M in FY2025 to $23.7B in FY2026 — essentially borrowing against future growth to pour money in today.
Across Big Tech, over $100B was raised through bond issuances in 2026 to fund AI CapEx, and investors have begun hedging at record levels via credit-default swaps (Morgan Stanley, 2026). Morgan Stanley projects AI-related global debt issuance could double to $570B in 2026.
So Where Should Investors Position?
The core principle is that "those who sell the shovels" profit before "those who mine the gold." Phase 1 of the AI infrastructure investment cycle (now) is the physical build-out phase, with semiconductor equipment (LRCX/AMAT/ASML), HBM memory (SK Hynix/Micron), and power & cooling infrastructure (Vertiv/Eaton) as the direct beneficiaries.
Hyperscalers like Oracle that are building AI infrastructure are in a "front-loaded capex" phase. Converting that CapEx into revenue requires broader AI service adoption, price discipline, and cloud margin recovery — all stories for 2027 and beyond. The gap between companies generating cash flow today and companies burning cash today is precisely what yesterday's Oracle -12% vs. equipment stocks +8% reflected.