Oracle's reported $300 billion, five-year computing arrangement with OpenAI helped add more than $240 billion to its stock-market value in a single day in September 2025.[1] Oracle also has negative $23.69 billion in free cash flow and has used bond financing as it expands data-center capacity.[2, 3] A more expensive path to constructing those facilities could leave the OpenAI agreement in place while increasing the capital Oracle must raise to fulfill it.
That outcome has become plausible since the US midterm elections. Cleanview identifies 159 gigawatts of planned data-center capacity, 41% of the national pipeline, in states where incoming governors favor halting new construction. Another 69.2GW, or 18%, is in states whose election winners favor changing incentives or operating rules without supporting a pause.[4] These figures measure exposure to political decisions, not lost capacity. They identify two ways in which the election could affect an infrastructure expansion backed by substantial financial commitments.
Governors could suspend projects outright. They could instead impose additional taxes, reduce preferential treatment or require developers to secure their own electricity supply. In the latter case, sites might still be built, but their construction and power costs would rise.[4] The customer agreements underpinning the projects would not automatically disappear. That would leave cloud suppliers facing higher delivery costs, AI developers facing existing payment commitments, and financiers exposed to the difference between spending today and collecting revenue later.
The threat is politically significant. In a voter survey, 56% said data centers played a major or crucial part in their election choice, and 76% opposed building one in their community.[5] The consequences for companies still depend on whether incoming governors prevent construction or make it more costly.
Multiyear AI contracts preserve demand but leave suppliers exposed to rising construction costs
OpenAI has agreed to buy $250 billion in computing services from Microsoft and $38 billion from Amazon, alongside the reported $300 billion Oracle agreement.[1, 6] These arrangements give suppliers prospective demand on a scale that does not vanish when technology stocks fall. They do not themselves guarantee that all scheduled capacity can be built, that clients can pay or that the nominal commitments will become recognized revenue.
Oracle's 2025 stock-market jump assigned substantial value to computing orders scheduled over several years.[1] Its negative free cash flow reflects the timing of infrastructure expenditure rather than a reported decline in existing operations.[3] Higher facility budgets could therefore increase Oracle's external funding needs while its reported OpenAI backlog remains largely intact.
The financial capacity of the customer is a separate constraint. HSBC estimates that OpenAI will need $207 billion more funding by 2030 than its projected revenue can support.[6] OpenAI had $24 billion of unused debt capacity in the financing picture described in 2025, while Oracle and other data-center providers had been turning to bond markets.[2] HSBC identified contract flexibility and growth in paying users as critical to OpenAI's ability to manage its obligations.[6] Additional construction costs would enter an arrangement already dependent on future funding and revenue growth.
Anthropic illustrates how binding the spending commitments can become before the underlying economics are settled. Its amended pre-IPO disclosures put purchase and lease obligations associated with computing infrastructure at approximately $895.4 billion, with nearly $70 billion due in the year following September 30, 2026.[7] A breakdown of compute commitments describes around $417 billion as noncancelable, although some individual agreements allow termination on 90 days' notice.[8] The amount that survives any construction interruption will depend on the enforceable terms, not the headline total.
Enforceable commitments could sustain infrastructure spending even as the economics of AI demand weaken
This creates two distinct pressures for Oracle. A contract that remains enforceable can sustain demand for its facilities while forcing the company to finance construction against a customer that HSBC already expects to need substantial additional capital. A delay severe enough to prevent delivery would create a different problem: contracted capacity might be unusable on schedule, and agreements with cancellation rights could be abandoned. The 159GW identified by Cleanview remains a measure of sites potentially affected by such decisions, not evidence that this second outcome has begun.[4, 6, 8]
Large cloud providers have substantial contracted demand of their own. Microsoft's and Amazon's agreements with OpenAI, and commitments to Google Cloud, Amazon Web Services and Microsoft appearing in Anthropic's compute arrangements, span multiple suppliers.[6, 8] These dollar amounts cannot determine which provider must absorb a new electricity bill or tax. Contract amendments and cost-sharing provisions would decide whether state action reduces suppliers' margins, raises customers' payments or increases the capital developers need to fund upfront.
The customers' economics are already changing independently of elections. Silicon Data's index of prices for one million model tokens fell to $0.97, more than 50% below its summer high.[9] Lower-cost Chinese models accounted for 48% of tokens traded on OpenRouter, while a survey found the proportion of enterprises relying primarily on public cloud for production inference had fallen from 56% to 41%.[10, 11] Lower prices can encourage usage, but the revenue earned from that usage must still cover long-term computing obligations. A shift toward private infrastructure could also reduce the demand available to public-cloud suppliers.
The IMF has also questioned whether AI deployment requires anything like the industry's current infrastructure spending, or whether the economic payoff will match expectations.[12] That challenge concerns the volume of capacity ultimately needed, not just the price of building it. Even a state policy that permits every affected facility to proceed cannot ensure that future AI usage will justify the full contracted expenditure.
Market losses reveal mounting financial concerns, but state decisions will determine how much capacity is actually constrained
The stock market has repriced those risks, but not in a way that isolates the election's effects. Roughly $2.5 trillion was erased from Nasdaq market value on November 5 as concerns about Democratic victories and data-center development intensified.[13] Technology valuations were already under pressure from negative free cash flow, deteriorating credit conditions and scrutiny of returns on investment.[14] A share-price decline driven by all those concerns cannot establish that projects will be stopped or that new regulatory expenses are already reflected in Oracle's future earnings.
State decisions during 2027 should identify how much construction is actually constrained. Any moratorium can be measured against the 159GW exposed to potential pauses; changes to tax incentives and electricity requirements can be checked against the additional 69.2GW facing possible rule revisions.[4] By December 31, 2027, projects still proceeding under costlier rules would support continued orders with greater funding needs. Verified cancellations would instead support the risk of lost or deferred delivery.
Corporate spending and contracts will show whether that difference persists beyond policy announcements. BofA forecasts aggregate hyperscaler investment above $860 billion in 2026 and close to $1.2 trillion in 2027.[15] Actual 2027 spending approaching the latter figure would be consistent with continued expansion; a total below the projected 2026 figure would point to a sharp pullback. Updated obligations from OpenAI and Anthropic, together with Microsoft, Amazon and Alphabet contracted-revenue disclosures, would establish whether customers are maintaining or reducing their purchases.
Oracle's 2027 financial statements will provide the most direct test. Changes in its borrowing costs, new debt, free cash flow, construction budgets and conversion of OpenAI-related backlog into revenue would reveal whether the company is earning from committed demand while paying more to deliver it. Oracle already has the reported $300 billion customer agreement and negative $23.69 billion free cash flow.[1, 3] If states permit construction but make electricity and facilities more expensive, the immediate cost will be carried by the party the contracts leave responsible for financing the buildout—and Oracle's balance sheet is already exposed.