Nvidia CEO Dismisses OpenAI Chip Challenge
Nvidia CEO Jensen Huang dismissed concerns about OpenAI's custom AI-chip effort on August 27, citing Nvidia's 33 years of experience and scale. Bloomberg reported on August 25 that OpenAI said its Jalapeno chip outperformed Nvidia's GB300 in tests measuring work per unit of power and response speed. The exchange puts proprietary benchmark results and hardware deployment scale at the center of AI infrastructure competition.
Nvidia CEO Jensen Huang dismissed concerns about OpenAI's custom AI-chip development, arguing that Nvidia's 33 years of experience, technology, and ability to operate at scale give it an advantage. India Today reported the comments on August 27, following a CNBC interview in which Huang addressed the prospect of major AI customers building potentially competing processors.
"You know. I am okay with it. There are so many XPUs that are being announced and as we know, it is not easy what we do. We have been doing this for 33 years," Huang said, according to India Today. He also said Nvidia was increasing its AI-market share, accelerating growth, and extending its technology leadership.
The comments follow reporting by Bloomberg that OpenAI's custom Jalapeno processor outperformed Nvidia's GB300 in OpenAI's tests on two measures: AI work per unit of power and the speed of returned responses. Bloomberg reported that the comparison used Nvidia's GB300, described as the leading option in the test.
Competing claims, different layers of evidence
The available reporting does not provide the test configuration, workloads, batch sizes, precision formats, software stack, or pricing assumptions behind OpenAI's Jalapeno results. Those details are consequential when comparing inference hardware, since throughput, latency, energy efficiency, memory capacity, interconnects, compilation tooling, and total cluster utilization can produce different winners for different workloads.
OpenAI's reported benchmark claim therefore establishes a performance assertion from the chip developer, rather than an independently replicated comparison. Huang's remarks, meanwhile, emphasize Nvidia's established ability to develop and supply AI infrastructure at scale. They do not identify a specific benchmark response or disclose a technical comparison with Jalapeno.
Nvidia's financial position
India Today linked Huang's confidence to Nvidia's fiscal 2027 second-quarter results, reporting $96.2 billion in revenue, more than double the prior-year level. Separately, CNBC reported that Huang characterized the capital demands of AI development and deployment as unusually large, saying frontier AI startups need tens of billions of dollars in funding.
CNBC also reported that Nvidia has invested across the AI ecosystem, including model developers such as OpenAI and Anthropic and providers of Nvidia-powered cloud capacity. The outlet noted that Nvidia has supported data-center projects financially and announced an arrangement with Wall Street firms for up to $500 billion in data-center financing.
For infrastructure teams, the dispute illustrates that accelerator selection is no longer only a question of peak chip performance. Companies evaluating purpose-built silicon typically need reproducible workload-specific benchmarks alongside evidence on compiler maturity, framework support, fleet availability, networking, reliability, and procurement terms. Comparable transitions in the sector have often placed greater value on those system-level factors as deployments grow beyond a single model or workload.
OpenAI's stated Jalapeno test results and Huang's scale argument address different parts of that evaluation. Public reporting has not established when, where, or at what scale Jalapeno will be deployed, nor has it supplied an independent technical assessment of the chip against Nvidia hardware.
Key Points
- 1Huang cited Nvidia's 33-year operating history and scale, while OpenAI said Jalapeno performed favorably against GB300 in testing.
- 2OpenAI's reported comparison lacks configuration and cost details, limiting conclusions about performance across real-world inference workloads and deployment environments.
- 3Industry hardware evaluations increasingly require workload-level benchmarks plus software, networking, availability, and total-cost evidence rather than peak-chip metrics alone.
Scoring Rationale
The reported Jalapeno benchmark claims and Nvidia CEO response concern competition in the AI accelerator market, a core constraint for large-scale model training and inference. The technical evidence remains incomplete and has not been independently replicated, but the story is notable because it involves OpenAI and the leading AI infrastructure supplier.
Sources
Public references used for this report.
Practice interview problems based on real data
1,625 SQL & Python problems across 15 industry datasets — the exact type of data you work with.
Try 250 free problems
