Building the engine before anyone has ordered the car

The hyperscalers are not investing in AI. They are financing a bet that the world reorganises itself around their infrastructure within three years. The capex is committed. The customers are not. This issue examines what that gap looks like from the outside — and where it points.

One Number

The annual AI revenue the five hyperscalers collectively need to generate a 10% return on projected 2026 capex.  ·  JPMorgan Chase/ The Economist

$560bn is not a revenue forecast. It is the minimum the five largest cloud providers need to generate annually just to achieve a 10% return on what they are committing to spend in 2026 alone. The infrastructure is being built. The demand that justifies it is not yet there.

One Argument

The AI capex is committed. The revenue is not.

The AI capex has been growing at 72% annually over three consecutive years, reaching what CreditSights has called “historically unthinkable levels”. The annual capex reaches levels of 45-57% of revenue while for Oracle it is 86%. CreditSights estimates that approximately 75% of that will fund AI-related infrastructure. Global AI compute capacity, according to Stanford Artificial Intelligence Index Report 2026, grew 3.3x per year since 2022, reaching 17.1 million H100 (an NVIDIA data centre graphics processing unit)-equivalents.

Currently the US AI firms have invested about 3-4% of American GDP over four years. This remains below the levels of British railway investment in the 1840s which, according to The Economist, reached 15-20% of GDP and serves as a useful historic benchmark.

The infrastructure buildup continues to accelerate despite the fact that, according to MIT State of AI in Business 2025 Report, 95% of organisations are getting zero return from generative AI investment. Meanwhile, Amazon's projected 2026 capital expenditure of $200bn is enough, according to Morgan Stanley, to push its free cash flow to negative $1.7bn.

The Economist reports that electricity took 40-50 years to generate productivity gains on factory floors after commercial deployment in the 1880s. The expectation of a quick payback from AI is overoptimistic by roughly four decades of precedent.

Hyperscalers argue that the AI is a structurally better business than anything we have seen before. Their free cash flow margins, according to Reuters, average 15% and are significantly higher than the 3.5% reported by the telecoms who built the optical fibre networks in the 1990s.

Order backlogs support the demand story. According to CloudWars, Microsoft, Oracle, AWS and Google Cloud are reporting combined backlogs of $1.63tn with about $800bn coming from enterprises that are expecting the hyperscalers to equip them for success in the AI economy.

The commercial logic is not trivial. But backlogs are commitments to spend - not proof that the infrastructure pays back at the speed the capex assumes.

Every technology wave has required someone to overbuild first. The question is not whether the bet pays off - it is who absorbs the loss if it takes a decade rather than three years.

One Position

Infrastructure, not models, is the position worth taking now. Which AI platform wins the race is unknowable. That power, cooling and data centre capacity will be needed regardless of the winner.

During all the gold rushes in history the providers of picks, shovels, kerosene and working apparel were the ones to benefit first. If even half of the 79% of family offices globally (JPMorgan 2026 Global Family Office Report) who admit that they have zero exposure to AI infrastructure today decide on a 5% allocation the supply and demand situation changes radically. The demand for underlying assets - power, data centres, cooling - is going to dwarf the supply.

This changes if the hyperscalers’ free cash flow margins begin to compress materially before infrastructure demand catches up. The risk is not hypothetical — CNBC reports that combined FCF for the four largest hyperscalers fell from $237bn in 2024 to $200bn in 2025. If that trajectory continues as capex accelerates, the investment thesis compresses with it.

If this was forwarded to you and you’d like to receive Corporate Financier’s Notes every Thursday, you can subscribe here.

Reading on your phone? Tap the three dots or overflow menu in the top corner of your email app and select View in browser for the best experience.

Corporate Financier’s Notes

Published every Thursday. One number, one argument, one position — for senior finance professionals who want the institutional view with a point of view attached.

© 2026 Corporate Financier’s Notes. All rights reserved. You are receiving this email because you subscribed to Corporate Financier’s Notes. If you no longer wish to receive these emails, you may unsubscribe at any time.