Tech · Explainer

Nvidia’s Next Test: Rival Chips and Its Financing Web

Michael Cembalest’s new commentary examines the business beyond Huang. The underlying filings show why the definitions matter.

Fictional analyst examines the connections between a large chip generator and cloud workshops as an engineer brings a rival chip.
AI-generated illustration: Kunkafa

Kunkafa News · Published · 2 min read

Kareem FaridBy Discuss with the author on X

Key points

  • Prospective computing capacity and existing chip-revenue market share measure different things.
  • Nvidia’s supply commitments, customer arrangements and credit backstops carry different cash-payment conditions.

Michael Cembalest sees Nvidia’s cash generation as resilient after Jensen Huang’s departure, while warning about competition and customer-financing ties. His November 14 commentary, The Icarus Files, presents that tension as a central test for the AI industry [1].

Cembalest chairs Market and Investment Strategy at J.P. Morgan Asset & Wealth Management [2]. The document identifies its views as his judgment and distinguishes them from JPMorgan Research [1].

A different measure of competition

2026 Cembalest/JPMorgan estimate: Nvidia has 39% of capacity in future frontier-lab deals — Kunkafa
0%10%20%30%40%Nvidia GPUsBroadcom customGoogle TPUsAMD GPUsAWS Trainium

Cembalest/JPMAM, 14 Nov 2026, p. 4. Capacity agreements; not installed capacity or revenue share.

The report assigns Nvidia 39% of prospective frontier-lab computing agreements, measured in gigawatts [1]. This covers leading AI developers’ future capacity arrangements. It cannot be treated as a current global GPU sales share or proof that announced projects are already operational.

Corporate announcements independently show the diversification. AMD and OpenAI announced a multiyear 6GW agreement in October 2025, with the first 1GW scheduled to begin deployment in the second half of 2026 [3]. Anthropic announced multiple gigawatts of Google/Broadcom TPU capacity starting in 2027 and said Claude uses AWS Trainium, Google TPUs and Nvidia GPUs [4].

For Nvidia, the commercial question is how much demand it retains as customers qualify alternative hardware. A multi-supplier strategy can reduce dependence on one vendor while total AI spending grows; a smaller slice of future agreements need not mean falling absolute revenue.

What the financing figures actually measure

Nvidia’s August filing lists $422bn of commitments across two tables and $108.5bn of maximum credit backstop exposure. Together they reach $530.5bn, combining multiyear obligations and contingent exposure. It separately reports $32.4bn of long-term debt [5].

The largest credit backstop, capped at $105bn, supports an SB Energy campus serving OpenAI. Payments depend on specified tenant defaults; the credit backstop takes effect in stages as leases begin. The full cap is not an immediate payment obligation [5].

The economic connection is important: a supplier that also supports customers’ financing can become exposed to both hardware demand and customers’ ability to fund it. Changes in that support could affect smaller cloud operators’ expansion, but the figures alone do not establish customer defaults or losses.

The practical questions extend beyond the next CEO: which capacity agreements become operating sites, whether alternatives deliver competitive economics, and when financing promises turn into cash payments. Those outcomes require operating results and contractual evidence, rather than a single headline total.

Sources: [1] Cembalest, The Icarus Files, 14 Nov 2026, pp. 3–4, 7–10 · [2] JPMorgan biography, accessed 15 Nov 2026 · [3] AMD/OpenAI, 6 Oct 2025 · [4] Anthropic, 6 Apr 2026 · [5] Nvidia 10-Q, 27 Aug 2026, notes 9–10

Kunkafa forecasts

NVIDIA

How far it could move, which way, by when, and how confident Kunkafa is.

Check the latest forecasts

More in Tech

From the Blog

All articles