OpenAI had announced impressive results for its Jalapeno chip yesterday, showing it to be beating NVIDIA’s older Blackwell chips, but NVIDIA CEO Jensen Huang doesn’t appear to be too concerned.
In a new interview with Jim Cramer, Cramer put it to Huang directly, describing a hypothetical conversation where he warns Huang about a chip that could “really hurt” NVIDIA, before naming it: Jalapeno. Cramer then asked whether Huang would really be as unbothered by that as he claims to be, adding that he personally would be hurt if a customer did to him what OpenAI has just done to NVIDIA.
Huang’s response was calm bordering on dismissive. He told Cramer he was okay with it, pointing out how many custom chips, or XPUs, are being announced across the industry right now, and reminding him that building this kind of hardware is genuinely difficult. “It’s not easy doing what we do,” Huang said. “We’ve been doing this for thirty-three years, and so lots of projects get started. Lots of projects get cancelled.” It’s the kind of line that works as both a factual observation about the graveyard of failed chip programs and a subtle jab at how early OpenAI still is in this process.
Huang went on to lay out why he isn’t losing sleep over Jalapeno specifically. He said NVIDIA is there to support its partners, and that the company intends to keep building what he called the world’s best technology, adding that he has “every confidence” in that. He made the same claim about NVIDIA continuing to be the most productive infrastructure option available to its customers, and again about NVIDIA having the supply chain and manufacturing scale to remain their largest supplier. Three separate claims, each closed out with the same phrase, delivered with the practiced confidence of someone who has answered a version of this question many times before.
The larger point Huang was making is about market share, not any single chip. He noted that despite a wave of XPU announcements and a growing list of startups entering the space, NVIDIA’s own share of the AI chip market keeps climbing, its growth is accelerating, and its technology lead keeps extending. That’s what he says lets him watch a rival’s benchmark results without treating them as a personal or business threat. Whether that holds up as more of these chips reach real production volume is a separate question, but for now, the numbers he’s citing are on his side.
Huang has been fielding versions of this question in nearly every recent public appearance, and OpenAI’s Jalapeno isn’t the only project he’s had to respond to. His answer here fits a pattern of downplaying the competitive threat from custom silicon while still acknowledging it exists, something he’s had to balance while also publicly praising the same AI labs and hyperscalers building these chips as customers who need NVIDIA hardware to keep growing.
The competitive backdrop makes his confidence easier to understand, even if it doesn’t settle the argument. Nvidia still controls roughly 92 percent of the GPU market by most estimates, and every major hyperscaler building its own silicon, including Google, Amazon, and Meta, continues to buy NVIDIA chips in large volume alongside its in-house alternatives. Google remains the furthest along of any custom silicon program, having spent more than a decade on its TPUs and still running that fleet near full utilization, yet Google itself remains one of NVIDIA’s largest customers. Amazon’s Trainium chips have crossed a meaningful annual revenue run rate of their own, and OpenAI’s Jalapeno now joins a growing list that also includes AMD’s newly acquired Taalas and rumored in-house projects at other labs.
None of this changes the fact that NVIDIA’s upcoming Rubin platform, not the older Blackwell generation, is the product actually built to compete with chips like Jalapeno on their own terms. Huang has previously said perf-per-watt, reliability, and long lifetime will define the next generation of GPUs, and Rubin’s HBM4 memory and redesigned architecture are meant to be NVIDIA’s answer to exactly the kind of efficiency argument custom chipmakers are now making. Whether Jalapeno’s early lead over Blackwell survives that comparison, and whether OpenAI can scale production fast enough to matter, will likely shape how seriously the rest of the industry takes today’s answer from Huang.