Nvidia CEO shrugs off OpenAI’s Jalapeño chip: ‘Lots of projects get canceled’

OpenAI has published the first performance results for Jalapeño, its custom AI inference chip co-developed with Broadcom ↘ 0.74%, showing it outperforms Nvidia's ↘ 4.58% Blackwell-generation systems on efficiency and speed. Nvidia CEO Jensen Huang responded Wednesday by downplaying the threat, telling CNBC's Mad Money that "lots of projects get started. Lots of projects get canceled."benzinga

The benchmarks, presented Tuesday at the Hot Chips conference at Stanford University, showed Jalapeño delivering 1.5 to 1.9 times higher peak throughput per kilowatt and 1.7 to 3.6 times lower end-to-end latency than Nvidia's GB200 and GB300 systems across three models ranging from 120 billion to 1 trillion parameters. The chip draws 700 watts, roughly half the power of the Nvidia systems it was tested against.cnbc+1

A Cost and Efficiency Play

OpenAI chip chief Richard Ho said Jalapeño should "drop [costs] by a lot" as it rolls out, according to Bloomberg. The chip was tested on the public SemiAnalysis InferenceX benchmark against Nvidia's GB300, which held the previous best results.latimes+1

"Jalapeño can serve more AI work per unit of power, while also returning responses more quickly," Ho said in a press call. The chip is rated at 700 watts but measured sustained power stayed at or below 550 watts during testing.egamers+1

Analysts noted caveats. SemiAnalysis said the comparison was "somewhat incomplete and unfair" because Jalapeño uses newer HBM4 memory, making Nvidia's Rubin platform — which also uses HBM4 and has already begun shipping — a more appropriate benchmark. Jalapeño is also inference-only and was not tested for model training, where Nvidia remains dominant.cnbc+2

Nvidia Unfazed

Huang expressed confidence in Nvidia's position during his Wednesday appearance. "We have the supply chain and the technology scale to be their largest supplier," he said, adding that Nvidia is "increasing our market share of the AI market" and that "our growth is accelerating."benzinga

OpenAI itself acknowledged the continued relationship. "Nvidia is a really good partner, and we continue to need a lot of Nvidia," Ho said.latimes

Deployment Timeline and Industry Implications

Jalapeño will begin deployment in small volumes within OpenAI's infrastructure by the end of 2026, with larger-scale rollout in 2027. Omdia analyst Alexander Harrowell told CNBC that custom ASIC chips like Jalapeño are expected to exceed GPUs in volume by 2028, calling this "the biggest competitive threat to Nvidia."egamers+2

OpenAI is already working on second and third generations of the chip, with Gen 2 tape-out expected in "the coming months." The company joins Google Alphabet Inc. ↗ 1.53%, Amazon Amazon.com, Inc. ↗ 3.97%, and Meta ↗ 1.21% in the growing push toward custom AI silicon — a trend that Yole Group analyst Adrien Sanchez described as a "threat to Nvidia's inference margins, which is the field growing the most at the moment."cnbc+2