Custom AI Chips Are Coming for Nvidia. OpenAI’s Jalapeño Could Be the Biggest Warning Yet.

Nvidia's dominance of the artificial intelligence chip market is facing a growing challenge as OpenAI and some of the world's largest technology companies push deeper into custom silicon, potentially threatening one of the chipmaker's most lucrative businesses.

OpenAI's first in-house AI chip, called Jalapeño, is emerging as the latest test of Nvidia's grip on the infrastructure powering the AI boom. OpenAI said that initial benchmarks showed the processor delivering "industry-leading speed and efficiency." The report comes a day after OpenAI said Jalapeño outperformed Nvidia's GB300 in the amount of AI work completed per unit of power and the speed at which the system returned responses.

Jalapeño is designed specifically for inference, the stage after an AI model has been trained when it responds to users, generates content, or carries out tasks. OpenAI says the chip can deliver both high throughput, allowing it to serve large numbers of users efficiently, and low latency, which translates into faster responses.

Adrien Sanchez, a technology analyst at Yole Group, told CNBC that Jalapeño demonstrates that a "hyperscaler-designed chip can now match or beat Nvidia's Blackwell-class GPUs on inference efficiency.

"Nvidia still controls the "vast majority" of AI computing and benefits from the enormous ecosystem built around CUDA, its software platform, Sanchez said. But he described OpenAI's processor as a "threat to Nvidia's inference margins, which is the field growing the most at the moment."

But there are limits to the Nvidia comparison. Jalapeño was not tested against Nvidia's newer Vera Rubin generation, which has begun shipping. It also is not intended to train AI models, an area where Nvidia remains crucial to OpenAI. Instead, Jalapeño is optimized for serving models after training is complete.

That means OpenAI is not preparing to abandon Nvidia. OpenAI hardware chief Richard Ho told Bloomberg that Nvidia remains a key partner and that OpenAI expects to continue buying large quantities of its processors as demand for computing power grows.

OpenAI is developing Jalapeño with Broadcom and plans to deploy the processor within its own computing infrastructure by the end of 2026. The company is already developing second- and third-generation versions of the semiconductor.

Alexander Harrowell, senior principal analyst at Omdia, called Jalapeño an "impressive achievement, most of all in terms of efficiency." At the scale at which OpenAI operates, efficiency improvements can translate into lower spending on electricity, cooling and power-distribution infrastructure, he said.

Research firm SemiAnalysis visited OpenAI's labs to benchmark Jalapeño and found that it outperformed Nvidia's Blackwell architecture on performance per watt in nearly all of the scenarios tested. But the firm cautioned that the comparison was "somewhat incomplete and unfair" because OpenAI's processor uses newer HBM4 high-bandwidth memory.

Nvidia's newer Rubin platform also uses HBM4, making it a more appropriate competitor. "Jalapeño is really competing against chips like Rubin that also use HBM4," SemiAnalysis analysts wrote. Nvidia's Vera Rubin systems are already beginning to ship to customers, they noted, while OpenAI remains some distance from deploying Jalapeño beyond engineering samples.

"This is the biggest competitive threat to NVIDIA," Harrowell told CNBC, noting that roughly half of AI infrastructure capital spending comes from hyperscale cloud providers that either already have custom chip programs or are capable of developing them.

The stakes may be particularly high because OpenAI has historically been one of Nvidia's biggest customers. Sanchez said the company has been "one of the largest single consumers of Nvidia GPUs."