The AI Capex Debate Isn't a Bubble Question Anymore. It's a Balance Sheet Question.

The AI Capex Debate Isn't a Bubble Question Anymore. It's a Balance Sheet Question.

When Alphabet hiked its 2026 capital expenditure forecast in late July, shares of Amazon, Meta, and Microsoft fell as a group — on the same week Alphabet's cloud business was reporting 82% growth. That reaction tells you where the AI investment cycle actually stands. The bear case that demand might collapse is effectively over. The burden of proof has shifted to a harder question: can this build be financed, and which companies convert it into revenue and profit most efficiently? J.P. Morgan's midyear answer — stronger AI revenue is making the $5.5 trillion global AI capex cycle economically viable — is an opinion. I deal in facts. So let me run the actual numbers.

The revenue side of the argument now checks out

The strongest signal in the first half of 2026 was that the three hyperscaler clouds accelerated growth simultaneously in the June quarter — not one winner stealing share from the others, but every platform growing faster than it did in March:

Cloud	Q2 2026 revenue	Q2 growth	Q1 growth	Q2 operating income
AWS	$42.2B	+37% (fastest in 18 quarters)	+28%	$16.6B (+63%)
Microsoft Intelligent Cloud	$39.3B	+43% (Azure & cloud)	+40%	$15.9B (+31%)
Google Cloud	$24.8B	+82%	+63%	$8.8B (+212%)

The global cloud infrastructure market hit $143 billion for the quarter, up 43% year over year. That is what investor confidence in the capex boom was waiting for. A bubble thesis requires revenue to start disappointing; instead the entire category accelerated.

The arithmetic that matters most is Google Cloud. Growth went from 63% to 82% between the March and June quarters, but the operating margin went the same direction — to 35.6%, nearly double the 17.8% from a year earlier. Growing that fast while getting more profitable per dollar is the definition of not buying growth with discounts, and it is the single most direct answer to the question of whether AI revenue can justify AI capex. Sales backlog roughly quintupled to $514 billion from $106 billion a year ago — multi-year, already-signed contracts for AI capacity and models, with about half expected to convert to revenue within 24 months. When demand shows up that far ahead of delivery, the revenue side of J.P. Morgan's argument is no longer theoretical. Sundar Pichai attributed the acceleration to "strong demand for AI infrastructure and AI solutions," and the backlog is the proof.

This is what separates Google from the other two in the current phase. AWS is still the market leader at 28% share with a $169 billion annual run rate, but it grew 37% — strong, yet the slowest of the three. Azure delivered a steady 43% on its cloud growth with Microsoft's enterprise install base as ballast. Google Cloud, by contrast, is the smallest of the three at 15% share and roughly a $99 billion run rate, but it is the one accelerating with the steepest margin trajectory. In tech, finding a growing market matters more than defending a moat — and in this cycle, Google is the one on the right side of the transition from capacity build-out to revenue conversion.

Demand is real. The financing is where the risk now lives.

But this is the part that matters, and it is not the part the market sold off on. Every one of these companies is now generating cash, and spending it all — and then borrowing — to fund the build. At Amazon, the scale is stark: trailing-twelve-month capital expenditures of roughly $173 billion against a free cash flow margin near zero, with free cash flow down 186% year over year. Amazon's free cash flow went negative in the first quarter and is forecast to stay negative for all of 2026. Alphabet took its first negative free-cash-flow quarter in the June period, its capex running around $132 billion on a trailing basis. Microsoft, whose trailing capex is roughly $116 billion, is expected to post a negative free-cash-flow quarter for the first time since at least 2001 by year-end. Only Meta remains comfortably cash-positive, because it is the only one of the four that does not yet have a real cloud business to build.

The funding gap is being bridged with debt, which is precisely the pattern I've flagged repeatedly in the supply chain: commitments surge, and leverage follows. J.P. Morgan expects $150 billion of U.S. hyperscaler debt issuance in 2026 plus another $100 billion abroad, part of a projected $2.1 trillion in data center financing over five years. The balance sheets already show it — Alphabet's long-term debt up 111% to $98 billion in the first half, Amazon's up 81% to $119 billion since the start of the year.

Notice that the supply-constraint signal has migrated from order books into prices. Microsoft attributed roughly $25 billion of its 2026 capex to higher component costs, driven specifically by memory supply constraints on AI chips. When a buyer spends $25 billion extra just to pay higher prices for constrained components, that is demand strong enough to validate the build — and simultaneously the warning that the build's economics depend on how long those constraints hold and how much they cost.

So the demand side of J.P. Morgan's thesis survives its facts-check, and the leverage side is now the live risk. This is the crux of the whole trade: operationally, the clouds are generating record operating cash flow, projected to clear $900 billion by 2027. But free cash flow — what's left after the build — is negative or rolling over at three of the four majors. The AI build is not being funded by leftover cash; it's being funded by debt and, implicitly, by the hope that revenue keeps accelerating faster than the spending.

Who actually touches the money

All of this capex has to land somewhere, and that is the supply-chain fact underneath the whole debate. Nvidia captures roughly 40% of hyperscaler capital expenditure and an estimated 85% to 90% of the AI accelerator market by revenue, protected by the CUDA software ecosystem and a one-generation-ahead hardware cadence. The results show it: Nvidia's fiscal first-quarter revenue hit $81.6 billion, up 85% from a year earlier, with data center revenue of $75.2 billion alone. The company reports fiscal second-quarter earnings on August 26, and the number that matters is whether guidance keeps stepping up as the hyperscalers keep spending — and whether the composition of that revenue is tilting toward inference workloads, which is the more durable part of the demand and the part that decides whether this build pays for itself over time.

Nvidia is the pure expression of the capex boom, and it trades like it — a $5.2 trillion market cap at roughly 20.6 times trailing sales. That multiple sits squarely in the range where high-quality secular compounders have always lived, so I am not making a bubble argument. The question is opportunity cost. The boom is being underwritten by customers whose free cash flow has gone negative; every dollar Nvidia collects from them is a dollar that no longer appears on their balance sheets. The asymmetry is worth stating plainly: Nvidia converts at a roughly 47% free cash flow margin with about 89% return on invested capital, while the customers paying it are cash-flow negative. That does not break the demand thesis today. It changes the risk profile of owning the entire trade at the same time.

Where the capital goes now

Company by company, the allocation logic follows the conversion economics, not the growth rates in isolation:

Alphabet/Google Cloud is the structural winner of this phase of the cycle. It is the only hyperscaler growing 80% plus while expanding operating margin, and its $514 billion backlog is the clearest evidence in the industry that revenue is pre-committed ahead of capex. The market sold the stock lower on the capex hike; the conversion economics argue the opposite direction. This is the name that deserves the larger weight in the AI trade on a 2026-2027 horizon.

Amazon/AWS remains the scale leader, and that scale is its defense, but it is spending the most, growing the slowest of the three clouds, and its free cash flow has gone most negative. The risk is not demand. It is that AWS's advantage gets funded with leverage at a time when the rest of the company is no longer throwing off cash to absorb it.

Microsoft/Azure is steady at 40%-plus, and its enterprise distribution is real, but Microsoft flipping to its first negative free-cash-flow quarter since 2001 is a threshold worth treating with respect. The OpenAI partnership gives it demand, and the Copilot installed base gives it a software-layer monetization angle Apple-style — but the cash-flow flag is the signal to watch.

Meta is the outlier: the only hyperscaler without an established cloud business, now positioning to rent out computing capacity it is buying from others. That is the most speculative position in the group — effectively a leveraged bet on being able to resell Nvidia capacity at a margin before it has proven partners.

Nvidia is the right side of the transition, and the long-term thesis — hardware dominance migrating into software monetization — remains intact. The debate is not whether Nvidia stays the beneficiary. It is whether a 20%-of-sales multiple attached to customers now funding their builds with debt still deserves the same weight as the names converting that debt into revenue and margin. My answer on allocation: if you own the hyperscaler that is converting best, you already own a claim on the same demand with less financing risk.

Here is the time horizon on this thesis. We are roughly in the second inning of a build that J.P. Morgan expects to exceed $1.1 trillion across the hyperscalers in 2027, and AI revenue is finally rising fast enough to make that build self-justifying. The bubble conversation is stale. The live question is whether the financing holds. The break condition for this view is specific: two consecutive quarters of decelerating cloud growth at the majors, or a rollover in Google Cloud's margin, and the leverage cuts in the other direction — the same debt that is funding the build becomes the thing that forces spending down. Until that signal fires, the evidence says the capex boom is economically real, and the names converting it into revenue and margin — Google first among them — are where the capital belongs.

The capex debate was whether AI revenue would ever arrive. That debate is settled. The next debate is whether the balance sheets pay for it. That one is just getting started.