$17 Billion and 86% of Alphabet's Free Cash Flow: OpenAI and Google Just Put a Balance-Sheet Test on the AI Buildout

The trigger was the Nov. 4, 2026 midterms. Of 41 local ballot measures on data centers across 10 states, voters backed restrictions or rejected pro-development measures in all but one, according to CNBC's count and a Reuters review of Ballotpedia data. A CNBC poll published Oct. 24 found 63% of registered voters opposed an AI data center within 10 miles of their home, against 13% who supported one. The market split on Nov. 4 as Democrats took Congress, then the Nasdaq fell 6.4% on Nov. 5 before chip stocks recovered on Nov. 6 as the pledges landed. The voters had decided; the policy was the response.

The damage showed up in credit first. CoreWeave's five-year credit default swap spread widened to a record 1,267.9 basis points on Nov. 5, and Intel's reached a record 149.7 basis points. UBS estimated that roughly 30% of expected US data center capacity additions in 2028 could be at risk from construction pauses, and more than 50 projects have already been cancelled in 2026. The pledges do not create permission. They buy it — and the 2028 numbers now depend on permission nobody has granted.

OpenAI's compact is easy to misread as optics. The Nov. 5 document sets out what changes for every compute campus the company owns and operates, and what it is asking every state legislature to copy. The scope is deliberately narrow. The obligations are specific, dated and auditable, and they carry a cost that is easy to miss.

OpenAI had already shown it can move a grid. Its Grid Protectors Program, launched Nov. 2, sets out how future campuses will cut electricity demand when the grid is strained, working with utilities and grid operators on on-site generation, storage and power controls. The Georgia Power arrangement allows OpenAI to reduce power delivered by up to 1,000 megawatts during high-demand periods, and Georgia Power estimates roughly $950 million in annual customer savings beginning in 2029 — at least $15 a month for a typical residential customer using 1,000 kWh. The campus trades flexibility for ratepayer protection, and the utility gets a customer it can switch off instead of a rate it has to build.

Google's Nov. 5 announcement is the same idea in hard assets. It committed $17 billion to its first Wyoming data center, on a 716-acre parcel eight miles south of Cheyenne in the Switchgrass Industrial Park, with four data halls and an office hub, logistics and network buildings, and completion targeted for 2031. Earlier reporting put the campus at up to 2.7 gigawatts; the company did not disclose capacity. What it did disclose was the financing model. Partners named are Tallgrass Energy, Black Hills Energy, American Electric Power and Related Digital.

The price of that power is now explicit. American Electric Power's roughly $4 billion for 702 megawatts of fuel cells works out to about $5,700 per kilowatt of capacity. Black Hills's $1.8 billion for 564 megawatts of gas works out to about $3,200 per kilowatt. The fuel cells cost roughly 80% more per kilowatt than the gas turbines standing in the same campus, and the premium buys what no ratepayer can be asked to absorb: on-site generation, carbon capture and a private grid. The power bill did not get cheaper. It moved from a regulator's balance sheet to a corporation's, at a markup.

There was a second, quieter move. On Nov. 6 Google acquired Crusoe's southeast Wyoming site outright — the permitted, power-planned parcel Crusoe had developed — and the campus is now being built by a Google subsidiary, Jupiter Star Holdings. Pre-development work, from site selection through permits and power planning, has become a priced asset, and the developer that could not lock financing was bought out. That is where the buildout is consolidating: away from developers and toward the balance sheets that can pay for interconnection.

In the third quarter of 2026 Alphabet spent $51.4 billion on property and equipment against $42.3 billion of operating cash flow, and reported negative $9.1 billion of free cash flow in a quarter when revenue grew 25% to $127.6 billion and Google Cloud revenue rose 103% to $30.7 billion. Trailing twelve-month free cash flow is $19.7 billion against trailing capital expenditure of $159.8 billion, and the gap was bridged with $96.6 billion of new debt and $51.0 billion of equity issuance in the same period. A single $17 billion Wyoming campus absorbs 86% of a year's worth of trailing free cash flow.

OpenAI is funding its own version of the same bet. On Feb. 27, 2026 it announced $110 billion of new investment at a $730 billion pre-money valuation — $30 billion from SoftBank, $30 billion from Nvidia and $50 billion from Amazon — alongside Nvidia commitments of 3 gigawatts of dedicated inference capacity and 2 gigawatts of training on Vera Rubin systems. Its largest project is a 10 gigawatt campus in southern Ohio built and managed by SB Energy, SoftBank's energy subsidiary. Nvidia committed up to $105 billion in credit and compute for that campus on Aug. 17, after an earlier plan for a roughly $250 billion guarantee was scaled back on Aug. 14. The pledge is real, and it is underwritten — by a vendor that has already trimmed once.

The clearest failure is CoreWeave. Its trailing interest expense of $1.87 billion is 24.7% of trailing revenue of $7.59 billion, and it still posted a trailing net loss of $1.93 billion on $3.94 billion of trailing EBITDA. That interest bill alone is roughly a tenth of what Google spent on the entire Wyoming campus. A company that cannot cover its own interest bill cannot also pre-pay a 20-year power purchase agreement, and it cannot build behind the meter.

Bloom Energy sits at the other end of the scale and shows what the market has already paid for the theme. Trailing EV to EBITDA of 83.6x and a trailing free cash flow yield of 0.9% leave no margin for a slipped order, even though third-quarter revenue rose 138.1% to $1,235.8 million at a 33.8% gross margin with full-year guidance raised. A policy that creates one very large buyer carries a matching risk: concentration reaches the order book, not just the balance sheet.

The pledge does not shrink the buildout. It relabels where the money is borrowed and who carries the risk. The money now goes to the companies that sell generation and equipment directly to the campus owner, and the contracts run for decades rather than quarters.

Two risks sit underneath all of it. The first is political: the pledges work only if voters believe them, and one state commission that decides a self-funded campus should still be charged into the local rate base would reset the precedent everywhere. The second is financial: Alphabet is funding a $195-205 billion annual capital expenditure program with debt and equity rather than cash flow, and the moment money markets reprice that funding, the cost of every campus rises for every hyperscaler at once. That is the real transmission channel from this story into the utility and equipment supply chain, and it runs through the cost of capital rather than through electricity prices.