Two filings landed on Wednesday, September 24, and together they contain more information about the health of the AI infrastructure cycle than any earnings call has offered this year. Read them separately and you see a tenant protecting itself and a content-delivery company landing a blockbuster deal. Read them together and you see the oldest pattern in capital markets: late-cycle financing dressed up as strategic investment.
Start with Oracle. Oracle sent a force majeure notice to Stack Infrastructure, the Blue Owl Capital-backed developer of Project Jupiter, a 2.45-gigawatt campus in southern New Mexico that is one of the flagship sites of the $500 billion Stargate program with OpenAI and SoftBank. The notice would allow Oracle to delay payments should the facility miss its 2028 target to come online. Oracle insists everything is fine. “Project Jupiter remains on our planned schedule,” the company said in a statement to CNBC. Companies that are genuinely confident in their schedules do not typically file contractual escape hatches. The notice follows a string of setbacks at the site tied to its energy supply: the campus is designed to run on gas-powered fuel cells from Bloom Energy, and an Energy Transfer pipeline intended to deliver gas has been delayed nearly six months, to February 1, 2027, after regulators repeatedly denied permits. Oracle shares fell about 3% to 4% on the day, and Bloom Energy fell about 6% to 7%.
The Akamai filing is the more structurally interesting of the two. Akamai disclosed an approximately $11.6 billion contractual commitment over seven years with Anthropic. Tucked inside the 8-K is a detail that deserves more attention than the headline number: Akamai issued a warrant to Anthropic for the purchase of non-voting convertible Series B Preferred Stock representing 7.7 million shares of Akamai’s common stock on an as-converted basis, at an exercise price of $111.33 per share. A portion representing approximately 2% of Akamai’s common stock outstanding is expected to vest with the initial $11.6 billion commitment, with the remaining approximately 3% vesting on successful expansion up to an additional $9 billion. In plain language: the customer received equity in its own supplier as part of the deal. That is not a standard commercial arrangement. It is a financing instrument layered into what looks like a procurement contract.
The Funding Structure Is the Signal
The technology underlying these projects may be genuinely transformational. That question is separate from whether the financing supporting this cycle is sound. History has a clear answer to what happens when the answer to both is not the same.
In the last 18 months, hyperscalers Alphabet, Amazon, Meta, Microsoft, and Oracle have moved from almost fully self-funded capex to raising external capital at scale. FactSet reported that incremental annual debt rose from 9% of capex in FY24 to 32% on a trailing-twelve-month basis by mid-2026, and equity has returned to the funding mix: Alphabet priced an $84.75 billion equity raise in June 2026, and Oracle has said it intends to raise roughly $40 billion through a mix of debt and equity in fiscal 2027.
The revenue side is growing more slowly than the capex side. Alphabet reported Google Cloud revenue growth of 82% year over year in Q2 2026. Amazon reported AWS revenue up 37% year over year to $42.2 billion, which the company described as its fastest growth in 18 quarters. But capex is outrunning both.
CoreWeave is the sharpest illustration of where this leads. CoreWeave’s guidance calls for $31 to $35 billion of capital spending in 2026, and it reported net interest expense of $536 million in the first quarter. The financing structure has also shifted: its most recent $2.6 billion loan carries a roughly five-year maturity, while the customer contracts underpinning it average about three years.
Cash Flow Is the Dividing Line
The question a disciplined investor must answer is not whether AI demand is real. It is whether a given company’s share of that demand is being captured with its own earnings or with borrowed money and structured warrants. Those are different businesses with different survival profiles.
Nvidia sits clearly in the first category. The company sells the picks to every miner and collects cash regardless of which mine produces gold. The hyperscalers with diversified revenue, strong free cash flow histories, and balance-sheet leverage still around 1x debt-to-EBITDA are strained but not broken. The companies in the second category, those whose growth depends on continuously rolling structured credit against customer contracts with shorter durations than the debt, need the cycle to keep accelerating to survive. Oracle filing a force majeure notice on a site it says is on schedule, and Anthropic receiving equity warrants in the infrastructure company it just committed $11.6 billion to, are both signs that parties inside this cycle are hedging against each other. That is what late-cycle financing looks like. The companies worth holding in a decade are the ones generating enough cash flow that they do not need to.