Jensen Huang plays down OpenAI’s “Jalapeño” chip, saying many AI projects get started and canceled

NVIDIA CEO Jensen Huang responded to the latest push for in-house AI chips by OpenAI with a blunt take, saying he is not personally tracking every new prototype and that customers’ plans can change quickly. In comments reported Wednesday, Huang said he remains confident in NVIDIA’s technology and in the company’s ability to supply the AI industry, even as some clients explore designing their own accelerators.

The remark centered on OpenAI’s so-called “Jalapeño” chip. While details of the device and its deployment path were not addressed in the cited report, Huang’s message was essentially that early-stage chips and internal initiatives do not automatically displace established suppliers, because programs often evolve or get dropped.

Huang’s characterization of the market emphasized volatility and uncertainty in AI product development. He said, in effect, that “lots of projects get started” and also that “lots of projects get canceled,” framing customer-designed hardware as one part of a larger pattern rather than a decisive turning point for NVIDIA’s business.

That stance comes at a time when the broader AI hardware ecosystem has been intensely competitive, with major companies seeking leverage across compute, networking, and software stacks. NVIDIA, which has built its data center strategy around providing not only chips but also the surrounding platform, has frequently argued that integrating AI workloads is not just a matter of raw accelerator performance.

In practical terms, Huang’s comments suggest NVIDIA views OpenAI’s hardware exploration as something that can coexist with continued demand for NVIDIA systems. Even if customers aim to reduce dependency over time, the transition from prototype to large-scale production is long, and it often intersects with software compatibility, system integration, and supply-chain execution.

The report also reflects a familiar dynamic for NVIDIA. As hyperscalers and researchers intensify their efforts to tailor compute, NVIDIA continues to position its ecosystem as the default path for deploying and scaling AI models. Huang’s messaging, at least as conveyed in the report, leaned toward reassurance rather than direct confrontation.

For readers, the key uncertainty is what “Jalapeño” ultimately represents for production use. The cited reporting conveyed Huang’s dismissal but did not provide additional specifics on timelines, customer adoption, or performance comparisons that would clarify whether Jalapeño is a near-term substitute for NVIDIA accelerators or a longer-term experiment.

Going forward, the market will likely watch for indicates that translate from words to procurement decisions. The most important datapoints would be any disclosures about whether customers adopting internal chips expand them into broader deployments, and whether NVIDIA’s AI infrastructure revenue trajectory remains resilient amid the growing appetite for custom silicon.