How much new capacity-purchase backstop support will NVIDIA contractually commit to independent GPU-cloud providers from November 13, 2026 through June 30, 2027?

Latest forecast

Generated Nov 13, 2026, 11:22 AM · High effort

P10 0.75
Median 2.25
P90 35.25
0.0%13.1%26.2%39.3%52.5%0387511315080% intervalMedian

Write-up

Executive Summary

The forecast’s 50th percentile is $2.25 billion, with a 10th–90th percentile range of $0.75–$35.25 billion. The middle case is limited new publicly verifiable support, but the distribution has a large upside tail: a relaunch or financing rescue could create much larger multiyear obligations.

This measures gross incremental contractual maximums under qualifying agreements signed November 13, 2026–June 30, 2027, using public evidence available through December 31, 2027. It is not a forecast of cash payments, expected losses, or all NVIDIA financing support. The existing $36 billion of AI-cloud commitments is outside the new-support subtotal.

Key drivers

  • The purchase-backstop mechanism is established, not merely proposed. NVIDIA’s July 26, 2026 quarterly disclosure reported $36 billion of AI-cloud commitments, typically lasting six years. NVIDIA may have to purchase capacity that providers cannot sell to third parties. This establishes willingness and scale, but the outstanding balance does not reveal a reliable rate of new signings. It also falls as capacity is consumed, so changes in that balance cannot identify gross additions. NVIDIA quarterly filing

  • Commercial and approval friction weigh against rapid expansion. Reuters reported a pause in some revenue-sharing arrangements, while NVIDIA said the model remained in place and continued to evolve. October reporting described further redesign after provider resistance. The evidence supports slower, more selective execution—not a verified permanent shutdown. Reuters; The Information

  • The leadership transition does not imply categorical opposition. Colette Kress became interim CEO effective November 11 while retaining the CFO role. She had coauthored the program’s launch and publicly explained its take-or-pay mechanism: a binding capacity-purchase floor that supports provider revenue even if third-party demand is insufficient. Her record supports tighter selection and sizing more clearly than abandonment. No capacity-specific moratorium was verified. Leadership filing; program launch

  • Customer financing can substitute for NVIDIA support. CoreWeave projected substantial customer prepayments and deposits over the following three months; those were expected receipts, not cash already received. Nebius reported strong prepayment coverage for its third-quarter agreement cohort, excluding short-duration agreements from that calculation. These disclosures weaken any assumption that provider expansion must generate proportional new NVIDIA backstops. CoreWeave results; Nebius shareholder letter

  • Unresolved financing needs preserve the upside tail. Bloomberg reported that NVIDIA’s separate asset-manager financing initiative had produced no transactions after months of work, with lenders seeking stronger support. Nscale’s November 12 prospectus also identified financing without binding commitments for its Anthropic deployment. Neither establishes a qualifying purchase obligation, but both preserve a reason to negotiate stronger demand floors. Bloomberg reporting; Nscale prospectus

Scenarios

Most weight rests on restraint and selective continuation. A successful relaunch receives less weight, while a large financing rescue is a smaller but consequential tail case.

ScenarioWhy it carries weightHow it affects the outcome
Restraint, excluded financing, or insufficient disclosureThe partial pause, redesign, leadership transition, and strict documentary requirements can all limit countable additions. Support may instead take forms excluded by the question.Little or no publicly quantified qualifying support, even if NVIDIA continues helping providers in other ways.
Selective qualifying continuationThe instrument has institutional backing, and Kress previously sponsored it. Financing needs persist, but approval and commercial obstacles remain.A limited number of new purchase floors produce meaningful full-duration commitments without a broad rollout.
Successful relaunchRevised economics could address provider objections while retaining a lender-supporting purchase floor. This is plausible, but not yet demonstrated.More signings and larger contractual maximums move the subtotal higher.
Large financing rescue or expansionFinancing difficulties could prompt stronger NVIDIA commitments. This receives the least weight because financing pressure has not yet demonstrated executable qualifying contracts.Several large, long-lived obligations create the distribution’s substantial upper tail.

Implications

  • Do not use the median as a cash-budget estimate. The target assumes eligible capacity is delivered and remains unsold, and counts the full contractual maximum. Actual purchases can be much lower without changing the measured commitment.
  • A median-only risk view would miss the upside exposure. Long contract durations mean a small number of agreements can materially increase gross obligations.
  • Broad financing headlines are poor accounting proxies. Equity, loans, hardware financing, and property or lease guarantees can support the ecosystem without increasing this subtotal. Conversely, a flat outstanding commitment balance can conceal new signings offset by consumption.

Signs to watch

Timing or developmentWhat would matterDirection for the forecast
November 17, 2026: scheduled earnings callSpecific comments on new AI-cloud approvals, revised terms, or restrictions would be more informative than general AI-demand commentary. The reporting date appears in the August earnings-call transcript.Confirmed continuation would move it higher; explicit restrictions lower. Clear policy guidance could narrow uncertainty and warrants rerunning the forecast.
November 18, 2026, 09:00 Pacific: scheduled employee meetingPublicly reported capital-allocation or approval changes could clarify the transition’s effect. The meeting was announced in employee emails.New constraints would move it lower; restored authority to approve qualifying agreements could move it higher. Generic continuity statements would resolve little.
Any time during the signing window: an executed agreement or incremental amendmentPublic terms must establish the signing date, qualifying purchase floor, maximum amount, and absence of overlap with earlier obligations.A verified new obligation would move it higher and narrow documentary uncertainty. Rerun when a material agreement is disclosed.
Any time: redesign, formal suspension, or permanent-CEO policy announcementRevised terms retaining the floor would distinguish continuation from substitution into excluded financing. A formal moratorium would settle a major policy uncertainty.A workable relaunch would move it higher; suspension or substitution lower. Either merits rerunning.
Any time: provider financing succeeds or deterioratesReceipt of customer prepayments or binding outside financing would reduce the need for NVIDIA support. Failed financing or stronger lender demands could increase pressure for purchase floors.Successful substitutes point lower; financing stress points higher or wider, although it does not guarantee qualifying contracts.
June 30, 2027 signing cutoff; December 31, 2027 evidence deadlineLater filings may retrospectively establish an eligible signing date and contractual maximum. Announcements and deployment dates alone do not suffice.Verified additions move the subtotal higher; evidence of earlier signing or overlap removes candidates. Contract-level documentation should narrow uncertainty.

Research and methodology

The research combined a targeted regulatory-filing review, historical contract comparisons, official program and leadership disclosures, and reporting on financing negotiations.

The most important agreement-level comparisons were:

Historical agreementWhat it establishes—and does not establish
CoreWeave, signed September 9, 2025The filing explicitly describes residual-unsold-capacity purchases and an initial value of $6.3 billion. It is a strong structural precedent, but “initial value” does not independently establish the exact incremental maximum. The publicly filed master agreement is redacted. Original filing; master agreement
SharonAI, agreement dated June 8, 2026A six-year agreement with reported contract value up to $4.88 billion supports the possibility of multibillion-dollar deals. The complete executed commercial terms were not located, so that headline could not be fully reconciled to a NVIDIA-only incremental backstop maximum. Original filing

Both agreements predate the measurement window and serve only as comparisons. NVIDIA’s quarterly filing supplied the broader commitment mechanics; its official launch and earnings remarks established policy sponsorship. Counterparty filings supplied financing alternatives and unmet needs. News reporting informed the pause, redesign, and execution-risk assessment rather than establishing countable contractual amounts.

The quantitative approach combined scenarios for the number of future obligations that would become publicly verifiable with a widely varying distribution of contractual sizes. More active scenarios also allowed larger agreements. These were judgmental assumptions, not statistically reliable estimates from a large historical sample. The model assessed whether obligations would be created and qualify; it did not discount a qualifying maximum for expected utilization.

Limitations

  • Executed monetary schedules were unavailable for the principal historical comparisons. This prevented independent verification of their precise incremental maximums.
  • There was no comprehensive public signing ledger or reconciliation of gross additions. The selected historical examples cannot establish a reliable signing rate or disclosure frequency.
  • No named, publicly quantified qualifying in-window pipeline or capacity-specific post-transition approval policy was verified. Some negotiation reporting was accessible only through excerpts or secondary accounts.

Unknown contractual amounts remain unknown—not zero-valued agreements. A zero resolution would mean no qualifying publicly quantified commitments were identified by the evidence deadline, not that NVIDIA signed no backstops.

Forecast history

Hover a data point to trace its series, or click to view the forecast generated at that time.

050.0100150Nov 13
P10
Median
P90

Signed forecast receipt

Externally anchored

Signed Nov 13, 2026, 11:22 AM with ed25519 key preseen-prod-ed25519-20260523 and externally timestamped Nov 13, 2026, 11:22 AM.

sha256:d4769e6164cbaf...5dbde66d79
Receipt JSON

Want forecasts like this?

Preseen researches your questions and answers them with calibrated, source-backed forecasts.