Big Tech's $300B AI Guarantees Raise Hidden Investor Risk

Oracle Corporation reported on September 10 that its revenue rose 30% in the fiscal first quarter, driven by a 121% jump in cloud infrastructure revenue, the business that rents computing power to AI companies. The company said its remaining performance obligations, the value of contracts signed but not yet delivered, reached $664 billion, more than seven times the revenue it expects this year, and on the back of that backlog Oracle raised its full-year revenue outlook to at least $90 billion. Roughly half of the backlog should turn into revenue within three years, more than three times what the company expects to collect in the whole of this year, and Oracle signed more than $30 billion of new AI cloud deals in the quarter, most structured as prepayments or customers bringing their own hardware. Delivering that backlog is costly: Oracle spent $28.5 billion on data centers and equipment in a single quarter, more than the cash its business generated, free cash flow stayed negative, gross margin fell more than five points to 61%, and a day after the results the company expanded its job-cut plan. Concentration risk looms as well, since a $300 billion contract with OpenAI signed last year begins in 2027 and is close to half of the current backlog, while OpenAI is still losing money. Management has pointed to an investor day in October for a fuller plan on margins and cash flow ahead of the next quarterly report in December.

NVIDIA, Google, and Emerald AI have created the AI Energy Management Alliance to coordinate electricity use in AI data centers, and the group has started building what it calls the first purpose-built power-flexible AI data center, designed to adjust consumption with grid needs. AEMA plans to publish shared frameworks for grid-responsive data centers with input from utilities, regulators, and large AI infrastructure users. The alliance's first power-flexible facility is in Manassas, and the inflection point will be whether it secures faster grid interconnection terms or preferential tariffs by late 2026. Concrete regulatory wins or standardized grid-responsive design templates would show AEMA is more than marketing and could influence how future DSX style AI factories are specified and approved. NVIDIA operates as a data center scale AI infrastructure provider across the US, Asia, Europe, and other regions, so any shift toward power-flexible facilities directly affects how its hardware and software fit into future grid-aware computing projects.