Amazon Web Services and Nvidia announced Wednesday a major expansion of their partnership, committing to deploy 2 million additional Nvidia GPUs across AWS's global infrastructure in 2027 and 2028, on top of the more than 1 million GPUs AWS had already pledged to add starting this year. [1] The companies said demand "has exceeded those expectations" since the earlier plan was announced at Nvidia's GTC conference — the justification offered for an expansion that lands one day after this paper reported Nvidia's own quarterly filing disclosing, for the first time, indebtedness as a standalone risk factor.

The announcement bundles far more than raw chip volume. AWS will bring Nvidia's new Vera CPU-based infrastructure onto its platform, extend NVIDIA NVLink Fusion interconnect technology to work with custom high-bandwidth memory for AWS's own Trainium chips, and build AI factories for the U.S. government that include 100,000 GPUs on secure infrastructure cleared for workloads classified at Impact Level 6 and above. [1] The companies are also expanding GPU-accelerated data processing on Amazon EMR — delivering up to 3.7 times faster processing at 30 percent better price performance than CPU-based configurations, according to the release — and vector indexing on Amazon OpenSearch up to 9 times faster. [1] AWS CEO Matt Garman said the expansion reflects a strategy of making "AWS the best place to run NVIDIA AI technologies," while Nvidia founder and CEO Jensen Huang called the partnership "one of the great growth engines of the AI era," adding that "demand is running ahead of every forecast." [1]

That demand framing sits uneasily against the financial picture this paper documented Wednesday. Nvidia's own quarterly filing disclosed debt due within one to five years jumping to $15 billion from $2.75 billion in a single quarter — more than a fivefold increase — alongside $33.5 billion in senior notes outstanding and a new $25 billion commercial paper program. Nvidia chief financial officer Colette Kress told analysts that combined capital expenditure among the "top five hyperscalers" is projected to rise to $1.3 trillion next year from $800 billion in 2026, and that Huang separately told CNBC's Jim Cramer the AWS deal includes "millions of CPUs" built on the new Vera chip. Kress's own figures are the demand pool Nvidia is now borrowing hundreds of billions of dollars against — and the AWS expansion announced Wednesday is the largest single order in that pool.

The customer on the other end of that order is not, by its own most recent disclosure, generating the cash to self-fund purchases at this scale. Amazon reported negative free cash flow for the quarter, as did Alphabet, Tesla and SpaceX, while Meta's cash generation shrank by roughly 90 percent — a detail CNBC's own live coverage of Nvidia's earnings raised without resolving. [2] Those companies are among Nvidia's largest buyers, and Goldman Sachs projects hyperscaler AI infrastructure spending will reach $1.2 trillion in 2027 — spending increasingly financed through debt and vendor arrangements rather than organic operating cash. Yahoo Finance's own aggregated market commentary on the earnings put the tension plainly, describing it as "hard" to argue that Nvidia's financing arrangements with its largest customers aren't, in some meaningful sense, circular: Nvidia extends capital and credit into the same companies placing record hardware orders, and those orders in turn justify Nvidia's own borrowing against future revenue.

Nvidia's press release makes no mention of financing terms, delivery-timeline covenants, or how AWS intends to fund a commitment measured in the millions of GPU units. [1] The companies describe the expansion purely in terms of capability — broader model choice, faster data pipelines, new physical-AI use cases through a robotics collaboration with Amazon Robotics — without addressing the balance-sheet mechanics underneath it. [1] That omission matters because AWS's parent company is, by its own most recent numbers, not generating enough cash to cover this order out of operations; whatever combination of debt, vendor credit, or compute-for-equity arrangement fills that gap is exactly the kind of detail Nvidia's own 10-Q has just begun treating as material enough to warn investors about, even as its press releases keep describing record deals in terms of demand alone. Nvidia's next earnings report, covering the period through late October, will be the first opportunity to see whether the AWS order shows up as revenue Nvidia can point to or as another line item in the debt disclosure it only started making this week.

-- THEO KAPLAN, San Francisco