NVIDIA OpenAI $100 billion deal: What CFO Colette Kress said and why it matters for AI infrastructure

Key Points

NVIDIA OpenAI $100 billion deal remains a letter of intent, not a signed contract, according to NVIDIA (Níngwēidá 英伟达) CFO Colette Kress (Kēlǎitè Kèléisī 科莱特·克雷斯).

At UBS’s Global Technology & AI Conference, Kress answered direct questions on AI hype, GPU demand, competition, and big customer commitments.

She confirmed the late-September plan to deploy “at least 10 gigawatts” of NVIDIA systems for OpenAI is still an investment intent and has not been finalized.

Conversion note: key figures below use an indicative exchange rate of ¥7.20 per $1 USD for clarity; actual FX rates vary.

Quick context: deal status, scale, and revenue implications

The OpenAI deployment was described in NVIDIA’s announcement as roughly ¥720 billion RMB ($100 billion USD).

NVIDIA estimated that deployment could translate into about $400–500 billion USD of revenue over time — about ¥2.88 trillion–¥3.60 trillion RMB.

Kress was explicit: NVIDIA is “working with them” but there is no executed final agreement and related revenues are not included in current revenue guidance.

Other commercial commitments worth noting

• Anthropic (up to $10 billion USD) — about ¥72 billion RMB ($10 billion USD) — disclosed as a potential commitment that could add to chip and system prebooking.
• Blackwell and Vera Rubin systems — expected by NVIDIA to create roughly $500 billion USD of demand across 2025–2026 — about ¥3.60 trillion RMB ($500 billion USD).
• Inventory and purchase commitments — surged about $25 billion USD last quarter — about ¥180 billion RMB ($25 billion USD) — versus prior quarters that typically moved by only a few hundred million dollars.

Selected highlights from the interview — what investors and builders should care about

1. Is there an AI bubble? (AI infrastructure and data center capex)

Kress said the industry is still in an early migration from CPU workloads to GPU-based accelerated computing.

She forecast data center infrastructure investment through the rest of the decade could total roughly $3–4 trillion USD, with about half driven by the shift from CPUs to accelerators.

Her point: this is additive compute capacity, not just one-for-one replacement.

2. Is NVIDIA’s lead shrinking? (competitive moat, platform play)

Her answer: “Absolutely not. Our lead has not shrunk.”

Kress framed NVIDIA’s advantage as system-level design — multiple chips working together to deliver the performance next-gen models need.

She emphasized the difference between NVIDIA’s programmable platform and fixed-function ASICs, and the value of a hardware-plus-software stack centered on CUDA and NVIDIA libraries.

Her claim: “everyone is using our platform” — across cloud and on-prem deployments — and most models run in NVIDIA’s ecosystem.

3. Can NVIDIA’s customers make money? (the AI economics flywheel)

Kress described a clear “flywheel” effect where model scale drives token generation, which increases willingness to pay for inference, which funds more model expansion and compute investment.

This is the core commercial loop that turns raw infrastructure demand into recurring, monetizable workloads.

4. Are huge, prebooked compute reservations risky?

Kress acknowledged developers book capacity ahead of revenue and that compute demand is often a long-term capital question tied to developer funding.

She said NVIDIA focuses on near-term orders (this year, next year, and the year after), evaluating demand by purchase orders and customer capital availability.

5. Vera Rubin timing and performance (product roadmap)

Vera Rubin has completed tape-out and NVIDIA expects market roll-out in the second half of next year.

Kress expects Vera Rubin to deliver multiple-times performance improvements relative to prior generations.

6. Gross margin outlook (HBM costs and profitability)

Despite concerns about rising HBM memory costs, NVIDIA expects to maintain gross margins in the mid‑70% range next year, roughly at the 70% midpoint.

7. Inventory and procurement surge (preparing for growth)

The company’s inventory and procurement commitments rose nearly $25 billion USD last quarter — about ¥180 billion RMB ($25 billion USD).

Kress described this as production ramp and shipments likely to be delivered to customers in the near term.

8. Capital allocation priorities (growth, M&A, buybacks)

NVIDIA is guiding revenue for the coming year in the range of $350–400 billion USD — about ¥2.52–2.88 trillion RMB.

Kress outlined priorities: fund internal capacity and supply expansion (including Vera Rubin), return cash via buybacks and dividends, and invest free cash flow into the ecosystem.

She added that very large, transformational M&A is difficult right now; NVIDIA favors strategic deals that bring engineering teams and technology that strengthen the platform and R&D.

What this all means for investors, founders, and builders

If you’re an investor: the headline ~¥720 billion RMB ($100 billion USD) OpenAI number is a signal of potential scale — but it’s nonbinding right now, so treat it as optional upside rather than booked revenue.

If you’re a founder or model developer: expect NVIDIA’s platform and ecosystem (hardware + CUDA + libraries) to remain a dominant force for large-scale model training and inference.

If you’re a cloud or enterprise buyer: plan for multi-year capacity growth and budget for accelerated computing as additive infrastructure, not just replacement.

Bottom line

Kress positioned NVIDIA as squarely in an early, demand-driven phase of a long transition to accelerated computing.

Headline intentions like the reported ~¥720 billion RMB ($100 billion USD) OpenAI deployment remain nonbinding, but NVIDIA’s near-term orders, product roadmap (including Vera Rubin), and platform strength are framed as drivers of sustained multi-year demand.

Keywords: AI infrastructure, GPU demand, accelerated computing, NVIDIA OpenAI $100 billion deal

NVIDIA OpenAI $100 billion deal

References

• NVIDIA CFO interview: OpenAI $100B deal not finalized – Cailian Press
• NVIDIA CFO interview: OpenAI $100B deal not finalized – Eastmoney (Dōngfāng Cáifù 东方财富)
• NVIDIA comments on Vera Rubin and demand outlook – Reuters