Alphabet Raises AI Spend to $205B: The CRE Read on the Capex Floor

On July 22, 2026, Alphabet reported results for the quarter ended June 30. In the earnings release filed with the SEC as an exhibit to its Form 8-K, the company reported revenue of $119.8 billion, up 24 percent year over year, operating income of $40.8 billion at a 34 percent operating margin, and Google Cloud revenue of $24.8 billion, up 82 percent. Reported diluted earnings per share of $9.11 was lifted by a large non-operating gain of roughly $98 billion in other income, so operating income is the cleaner read on the core business. Those are figures Alphabet filed, not PSV estimates. The number that moved the story sits in the cash flow statement: purchases of property and equipment, the capital expenditure line, were $44.9 billion for the quarter, up from $22.4 billion a year earlier, which computes to roughly double. Free cash flow for the quarter was negative $5.9 billion.

On the earnings call, management raised full-year 2026 capital expenditure guidance to a range of $195 billion to $205 billion, up from the prior $180 billion to $190 billion, and said roughly 60 percent of the spend goes to servers and 40 percent to data centers and networking, with 2027 capex expected to increase significantly. That guidance and split are management commentary, not filed line items. The filing also shows how the buildout is being funded: Alphabet raised about $49.6 billion in net proceeds from a stock and mandatory convertible preferred offering, which the company says is for general corporate purposes including capital expenditures to scale AI infrastructure and global compute, and entered an agreement to sell up to $40 billion of additional stock. Secondary reporting noted the shares fell after the print on the capex hike; that market reaction is reported, not a PSV finding.

Every data center lease, every power interconnection queue, every industrial land bank near a substation is a bet that hyperscaler capex keeps climbing. Alphabet just told the market it will spend as much as $205 billion this year, is issuing equity to help fund it, and is willing to run negative free cash flow to secure compute capacity. Strip away the AI headline and the signal for real estate is specific: the most disciplined capital allocator on earth is torching free cash flow and diluting shareholders to lock in physical infrastructure. That is the demand floor under data center rents, under the powered-land premium, and under the industrial and utility corridors that feed these campuses. It is also not Alphabet alone. Reported aggregates put the four largest hyperscalers near $725 billion of combined 2026 capex, up sharply from the prior year, which is the reported context, not a filed figure.

The counter-signal is the one lazy coverage skips: the market sold the stock, and free cash flow went negative. That is investors starting to price the risk that capex outruns monetization. Here is the distinction that should drive an underwriting decision. A twenty-year, triple-net data center lease to an investment-grade hyperscaler is credit risk: it pays as long as the tenant honors the contract, whatever happens to the AI cycle. A speculative powered-land play, a merchant data center built on the expectation that demand keeps absorbing capacity, or an industrial corridor priced to today’s absorption pace is cycle risk: it pays only as long as the capex supercycle continues. Those are different exposures that look identical on a pro forma when demand is rising. The re-rating that made early data center owners look brilliant runs in reverse just as fast if a hyperscaler ever blinks. Know which risk you actually own.

Treat hyperscaler earnings season as a standing market-intelligence input, not a headline to skim. The inputs are primary records and your own footprint: the quarterly 8-K earnings releases and 10-Q cash flow statements of the major hyperscalers, the capex and guidance commentary from each earnings call, and your own map of data center, powered-land, and industrial exposure by market. The job for an AI assistant is a cited capex tracker: pull the filed capital expenditure line and the stated forward guidance each quarter, flag the direction of change and any commentary on capacity constraints, power, or third-party leasing, and tie it to which of your markets and assets sit in the path of that spend, with every number traced to the filing or transcript it came from.

The reviewer is the head of research or the CIO, and the approval gate is explicit: no underwriting assumption, discount rate, or exit-cap change moves on a capex read alone, and nothing goes to the investment committee without a human signing off on what the filing actually supports versus what the call implied. The same run surfaces the questions worth working: which of your data center or power-corridor deals are underwritten to tenant credit versus to the cycle continuing, and which counterparties in your rent roll are exposed if hyperscaler spend decelerates. The tool assembles the record and the map every quarter. The allocation call stays yours.