Sector Intelligence

The $1 CapEx Trillion AI Question: Who Pays – and Who Profits?

Sep 11, 2026

By Ieva Bružaitė | Founder, Orion Capital | September 11, 2026

AI CapEx is no longer just a technology story. It is becoming one of the largest capital-allocation experiments in modern corporate history. Microsoft, Amazon, Alphabet and Meta are committing extraordinary amounts of capital to GPUs, servers, networking equipment, data centers and power infrastructure. The scale is changing so quickly that the central question for investors is also changing.

The first phase of the AI trade was about demand.

The next phase may be about return on capital.

For investors, the question is no longer simply how much Big Tech can spend on AI. It is whether those investments can generate enough incremental revenue, cash flow and productivity to justify the capital being deployed. Between the capital deployers and the eventual monetization sits a large group of beneficiaries – from NVIDIA and Broadcom to memory, networking, data-center and power-infrastructure suppliers.

AI CapEx Is Entering a New Phase

CapEx – capital expenditure – represents investment in assets expected to generate value over multiple years. In the AI economy, that increasingly means GPUs and accelerators, CPUs, servers, networking equipment, data centers, cooling systems and the electrical infrastructure required to operate them.

2026 AI Infrastructure Spending: A Comparison

Company	2026 CapEx / run-rate	What the capital is building	Investor question
Amazon	~$220B (raised from $200B)	AWS and AI compute, data centers, networking	Can AWS growth and AI monetization rebuild free cash flow?
Alphabet	$195-205B (raised twice this year, from $175-185B)	Servers, data centers and networking	Can Search, Cloud and AI revenue scale faster than depreciation?
Microsoft	~$175B CY2026 (accounting adjustment from $190B); $41B latest quarter	CPUs/GPUs and cloud/AI capacity	Will constrained capacity convert into durable high-return growth?
Meta	$130-145B expected	AI infrastructure for ads, recommendations and future products	How much incremental ad/productivity return comes from the spend?

Key point: the same AI CapEx that is a cash outflow for hyperscalers becomes revenue opportunity for semiconductor and infrastructure suppliers.

Follow the Money

Hyperscaler CapEx → Semiconductors → Networking → Data Centers → Power & Cooling → Compute → AI Products → Revenue → Free Cash Flow → ROIC

Investors have spent much of the past several years concentrating on the first half of this chain. Every additional dollar committed by hyperscalers to AI infrastructure creates potential revenue somewhere else in the ecosystem. The next analytical challenge is the second half: whether the installed compute produces enough monetizable demand and cash flow.

The Suppliers: Monetizing the Buildout

NVIDIA is the clearest example of an infrastructure beneficiary. In its latest reported quarter, revenue reached $96.2 billion, up 106% year over year, while Data Center revenue reached $89.0 billion, up 117%. Broadcom provides another view of the same cycle through custom AI accelerators and networking. Its Q3 2026 AI semiconductor revenue reached $16.7 billion, up 221% year over year, with management expecting approximately $21.7 billion in Q4.

For infrastructure suppliers, Big Tech’s CapEx can become current revenue. A company purchasing that infrastructure, however, must monetize the asset over several years. That difference is central to the next phase of the AI investment debate.

The Buyers: Four Different Return-on-Capital Tests

Amazon: growth versus cash consumption

AWS revenue increased 37% year over year in Q2 2026 to $42.2 billion, and Amazon said its AI business within AWS had surpassed a $25 billion annual revenue run rate. Yet the capital requirement is enormous. Trailing-12-month purchases of property and equipment reached roughly $173 billion, while free cash flow moved from positive $18.2 billion a year earlier to an outflow of $7.6 billion, with Amazon attributing much of the change to higher AI investment.

Meta: AI improves the core business, but the hurdle rises

Meta’s Q2 2026 revenue increased 28% year over year to $60.8 billion. Quarterly CapEx reached $31.1 billion and free cash flow was $784 million. The company now expects $130-145 billion of 2026 CapEx. AI may strengthen advertising and future products, but investors increasingly need to measure the return generated by that capital.

Microsoft: demand meets capacity constraints

Microsoft’s latest quarterly Microsoft Cloud revenue reached $59.3 billion, up 27%, while commercial remaining performance obligations reached $678 billion, up 84% year over year. The company has continued to describe infrastructure capacity as constrained. This supports the bull case: much of the investment is being deployed into an environment where demand is already substantial rather than purely hypothetical.

Alphabet: scale, servers and monetization

Alphabet has raised its 2026 CapEx guidance twice this year, most recently to $195-205 billion – up from $175-185 billion at the start of the year – with the majority directed toward technical infrastructure and management citing accelerating demand it could not otherwise meet. Its investment case increasingly depends on whether AI can deepen monetization across Search, Cloud and advertising while maintaining attractive margins and returns on invested capital.

The Hidden Second Layer: Commitments Beyond CapEx

The infrastructure race extends beyond reported annual capital expenditure. Reuters has calculated that Microsoft, Meta, Oracle, Amazon and Alphabet together have roughly $1.09 trillion of future data-center lease commitments that have not yet commenced. These commitments are not equivalent to conventional debt, but they show how far into the future today’s infrastructure decisions extend.

Financing is becoming part of the story as well. On September 8, Amazon moved toward its first sterling bond sale as large technology companies increasingly tap debt markets while financing the AI buildout. That does not imply financial distress; it does underline the sheer scale and duration of the capital cycle.

What Happens to Free Cash Flow?

A company can simultaneously report strong revenue growth, strong operating earnings, rapid AI adoption and falling free cash flow. There is no contradiction. If operating cash generation rises by $20 billion while capital expenditure rises by $50 billion, free cash flow can decline despite a growing underlying business.

AI CapEx → AI Revenue → Operating Cash Flow → Free Cash Flow → ROIC

Bull Case Versus Risk Case

Bull case	Risk case
AI becomes a foundational computing layer across the economy.	Compute supply eventually grows faster than monetizable demand.
Current capacity remains constrained and utilization stays high.	Depreciation and financing costs rise faster than incremental profits.
Cloud, advertising and enterprise AI monetization catches up with investment.	Competition lowers AI pricing and compresses returns.
Suppliers continue benefiting from sustained infrastructure demand.	AI succeeds technologically while some capital deployed earns disappointing financial returns.

AI succeeds technologically while some capital deployed earns disappointing financial returns – the quiet middle case worth holding in mind alongside the two poles above.

Orion Capital View

The AI investment debate is moving from demand toward capital efficiency. For the past several years, the strongest question investors could ask was: Who benefits from AI adoption? The next question may be more demanding: Who can convert AI investment into durable free cash flow?

Infrastructure suppliers remain positioned to capture enormous spending while the buildout continues. For hyperscalers, however, the hurdle is rising. Hundreds of billions of dollars of annual CapEx – combined with long-term data-center commitments – mean that revenue growth alone will eventually be insufficient to judge success. Investors will increasingly need to follow utilization, margins, depreciation, free cash flow and return on invested capital.

The first phase rewarded those who supplied the AI infrastructure.

The next phase may reward those who prove they can earn an attractive return on it.

Disclaimer: This publication is independent research and reflects the views of the author. It is provided for informational purposes only and does not constitute investment advice, an offer, solicitation or recommendation to buy or sell any security. Financial figures should be checked against the issuer’s latest filings before publication or investment use.

Sources and Data Notes

About the Author

Ieva Bružaitė is the Founder of Orion Capital, an independent research platform covering AI, semiconductors and long-term investing.

Ieva Bružaitė - Founder of Orion Capital

← Back to Orion Capital