Hyperscalers pre-sell data center capacity as AI spending hits $725B

D.A. Davidson's Gil Luria argues the $725 billion hyperscaler AI spending spree is already generating returns — because each new data center is pre-sold to customers before construction begins.

"There's a disconnect between what the companies are saying about return on investment from this AI spend and what investors feel," Luria, Head of Technology Research at D.A. Davidson, said in a July 10 CNBC interview. "What investors see is diminishing cash flows, the lowest levels of cash flow margin they've seen in a long time."

OpenAI and Anthropic's combined run rate surged from less than $20 billion to more than $75 billion in six months, Luria said, citing the demand that underpins the buildout. Microsoft's Q3 FY26 capex totaled $30.88 billion, up 84 percent year-over-year, while operating margin held at 46.3 percent and the AI business reached a $37 billion annual run rate. Amazon posted AWS revenue of $37.587 billion in Q1 2026, up 28 percent — the fastest growth in 15 quarters — at a 37.7 percent operating margin. Google Cloud revenue grew 63 percent to $20.03 billion, though free cash flow fell 47 percent to $10.12 billion.

The five largest hyperscalers — Alphabet, Amazon, Meta, Microsoft and Oracle — have collectively added about $350 billion in debt over five years to fund the buildout, according to Bloomberg-compiled data. Luria said the companies argue that projected returns on AI services, stacked against relatively low interest rates on new debt, make the spending worthwhile. The near-term test will be whether Azure accelerates from 40 percent growth and whether AWS and Google Cloud sustain their recent momentum.

Pre-sold capacity underpins the thesis

Luria's argument rests on a structural shift in how hyperscalers finance data center construction. "When we build a data center, it's already pre-sold," he said. "We know what it's going to cost to build and operate. We're marking that up substantially to our customers, and therefore there's a good return." Microsoft's commercial remaining performance obligations reached $627 billion, while Alphabet's cloud backlog nearly doubled quarter-on-quarter to more than $460 billion — both supporting the claim that contracted demand, not speculation, is driving the capex cycle.

Amazon's custom chip business topped a $20 billion revenue run rate, growing triple digits year-over-year. Anthropic committed to up to 5 gigawatts of Trainium capacity and OpenAI to roughly 2 gigawatts starting in 2027. Amazon guided full-year 2026 capex at roughly $200 billion, while Alphabet plans $175 billion to $185 billion.

Investor skepticism meets accelerating cloud growth

Despite the bullish demand signals, equity markets have taken a cautious view. Only Alphabet's stock has outperformed the S&P 500 this year, while Microsoft and Oracle shares have dropped more than 20 percent. Amazon's free cash flow went negative in the quarter ending March 31, and S&P Global Ratings downgraded Oracle to the lowest investment-grade rating in May, citing growing AI spending.

Luria said both sides can be right. Hyperscalers are spending enormous sums upfront to meet contracted demand, while the revenue and cash flow from those investments will arrive over several years. "This doesn't look bad," he said of the combined debt load. "If they were borrowing an order of magnitude more? That would look bad."

Microsoft shares, trading at roughly 22 times forward earnings, have declined 20 percent year-to-date through July 9. Amazon trades at about 28 times forward earnings, while Alphabet trades at roughly 21 times. The valuation compression reflects the market's uncertainty about when the cash flow inflection arrives — a question that will likely dominate earnings calls beginning later this month.

This article is for informational purposes only and does not constitute investment advice.