OpenAI unveiled benchmark results for its first custom AI chip, Jalapeño, on Tuesday, and analysts told CNBC the semiconductor poses a fresh threat to Nvidia's inference business. The chip beat Nvidia's Blackwell systems on performance per watt in nearly all tested scenarios, though analysts say Nvidia's newer Rubin platform is a fairer comparison.
Nvidia's near-monopoly over the most advanced AI chips is under pressure as OpenAI and other tech giants push forward with custom-built semiconductors, analysts told CNBC. OpenAI unveiled its Jalapeño chip on Tuesday, describing it as industry-leading on speed and efficiency. Google, AWS and Meta are also developing their own AI chips.
The company's share price has rocketed amid the data center buildout, which has created huge demand for its chips in both model training and inference, the day-to-day running of AI systems.
A threat to Nvidia's fastest-growing segment
Jalapeño, designed for inference, shows that a hyperscaler-designed chip can now match or beat Nvidia's Blackwell-class GPUs on inference efficiency, Adrien Sanchez, technology analyst at Yole Group, told CNBC. Nvidia still owns the vast majority of AI compute and benefits from ecosystem lock-in through its CUDA software platform, Sanchez said. According to Sanchez, the new chip is a "threat to Nvidia's inference margins, which is the field growing the most at the moment."
The chip is being developed with Broadcom and will be deployed within OpenAI's compute infrastructure by the end of the year, with generations two and three already in progress.
How Jalapeño stacks up against Blackwell
Research firm SemiAnalysis visited OpenAI's labs to benchmark Jalapeño and found it beat Blackwell on performance per watt in nearly all tested scenarios. But the firm said the comparison was somewhat incomplete because Jalapeño uses newer HBM4 memory, making Nvidia's Rubin platform, which also uses HBM4, a more like-for-like rival. Vera Rubin systems are already shipping to customers, while OpenAI has so far produced only engineering samples of Jalapeño.
Fion Chiu, analyst at TrendForce, said OpenAI's custom chip could reduce its reliance on Nvidia over time for inference workloads. But for large-scale model training and frontier AI workloads, Nvidia GPUs will remain important given their broad programmability, performance and software ecosystem, Chiu said.
Other AI companies build custom silicon
OpenAI is one of several AI companies looking to build custom silicon. In April, Google unveiled new training and inference chips it calls tensor processing units. Meta agreed to deploy one gigawatt of custom AI chips using Broadcom technology. Anthropic committed to spending more than $100 billion on AWS technology over the next 10 years, including current and future generations of Amazon's Trainium chips. Startups Cerebras, SambaNova, D-Matrix, Etched and Fractile are also building AI chips.
Alexander Harrowell, senior principal analyst at Omdia, said custom ASIC chips like Jalapeño could exceed GPUs in volume by 2028, though revenue will take longer to catch up since GPUs are considerably more expensive. He called the trend the biggest competitive threat to Nvidia, since roughly half of AI infrastructure capital spending comes from hyperscale cloud providers that either have a custom chip program or could reasonably build one.
OpenAI has been one of Nvidia's largest single customers for GPUs, and Sanchez said Jalapeño raises the stakes for that relationship specifically.
Source: CNBC
Trading involves risk.
Risk Disclaimer
