The Program Addressed a Real Problem
Smaller cloud providers who want to build capacity need to show lenders revenue certainty, which is exactly what they lack. The AI Compute Partnership was designed to fill that gap: Nvidia provides a take-or-pay commitment on a portion of a facility's capacity — a minimum revenue guarantee that gives lenders confidence to underwrite the project — and shares in revenue above that floor. That is how CFO Colette Kress described the mechanics on the earnings call. For a lender, it means part of the cash flow is effectively backed by Nvidia rather than resting entirely on the operator's ability to find customers. It is the same shape as Nvidia's backstop on OpenAI's Ohio campus lease, which this site covered earlier — using its own credit to lever someone else's project, at the cost of its own exposure. Among the first adopters, Firmus deployed 170,000 GPUs in Batam, Indonesia, and Sharon AI deployed 40,000 GB300s.
What Stalled It Was Control, Not Money
Per the reporting, the friction was in the attached conditions. Nvidia had reportedly told some customers they could only rent chips to Nvidia-approved customers, and signaled a preference for capacity spread across several smaller AI firms rather than concentrated with one large buyer. Some prospective partners were uncomfortable with that degree of control from the start, and certain six-year proposals would have let Nvidia take up to 50% of partner revenue once specified thresholds were crossed. The internal concern followed: staff warned that being this deeply involved in how customers used the processors could attract antitrust scrutiny. That detail is worth noting — what reportedly drove the pause was not a regulator's action but the company's own anticipation of one.
The Two Accounts Do Not Match
Keep the framings separate. The reporting says some deals were paused, and explicitly that this is a pause rather than a scrapping — Nvidia could still modify the financing model or fold it into another program. Nvidia's spokesperson denies it, saying the July business model opening compute access to the AI ecosystem is still in place and continues to evolve due to high demand. The two are not in direct conflict — "paused some deals" and "the model is still in place" can both be literally true — but from outside, those two sentences are all there is, with no visibility into how many deals actually got signed. The filings offer one reference point: for fiscal Q2 2027, ended July 26, 2026, Nvidia disclosed $36 billion in commitments related to AI cloud deals, while the same document notes the figure rests on memorandums of understanding that may not become definitive agreements. News of the pause briefly weighed on Nvidia shares and some neocloud stocks. The practical implication: if you are evaluating renting compute from a smaller cloud provider, the signal here is not that they are about to fail. It is that this class of provider may depend on a chip vendor's credit backing more than you would assume — worth asking, before signing a long-term contract, who stands behind their capacity commitment.
via: Reuters via DCD, Tom's Hardware (with Nvidia's denial), CloudNews