0
This article says if AI demand stalls Nvidia is on the hook, but I would say if American businesses smarten up and start paying DeepSeeks seeks $0.18 per million output versus OpenAI's $25, Nvidia could be on the hook.
Price competition from DeepSeek and other cheap models is a real risk to the OpenAI-Nvidia bet, but the evidence so far cuts the other way — cheaper AI has driven more total compute demand, not less, which is exactly why Nvidia has kept doubling down.
Why it matters: The Nvidia guarantee only pays off if OpenAI's revenue keeps growing fast enough to cover its Ohio lease and power bills, so anything that erodes OpenAI's pricing power is a genuine threat, not just a hypothetical one.
- Current pricing gap is real if smaller than your numbers: DeepSeek's output tokens run about $0.66–$3.96 per million under its new peak/off-peak scheme (up from roughly $0.28–$0.87 before an Aug. 16, 2026 repricing), while OpenAI's flagship GPT-5.5 runs $30 per million output — though its cheaper tiers (GPT-5.4 Nano at $0.20) already undercut most rivals.
- The 'DeepSeek moment' already happened once: its January 2025 launch wiped roughly $600 billion off Nvidia's market cap in a day on fears cheap AI would gut chip demand — but GPU demand kept climbing afterward, the textbook Jevons paradox where efficiency gains expand total usage rather than shrink it.
- OpenAI's own numbers argue against your thesis, for now: CFO Sarah Friar told investors on Aug. 14, 2026 that enterprise revenue hit $40 billion annualized, up 20% month-over-month in July, with business customers up 32%.
- But the pressure is showing up in market share, not just price: OpenAI's share of enterprise AI spending has slid from roughly 50% in 2023 to 27-29% in early 2026, with Anthropic now edging ahead in new business adoption — that's the slower-moving version of the risk you're describing.
Room for disagreement
- Bulls, including Jensen Huang and most infrastructure analysts, argue cheap models are additive, not substitutive — workloads that don't exist at $30 per million tokens get built at $1, so falling prices grow the total compute market Nvidia sells into.
- Skeptics point to OpenAI's eroding enterprise market share as evidence commoditization compresses margins across the stack, and if OpenAI loses pricing power to cheaper rivals faster than volume grows, its revenue — the thing backing Nvidia's $105 billion guarantee — could fall short even as industry-wide AI demand keeps rising.
AI-generated with web search — may be wrong. Check the linked sources.
