NvidiaAnalysts told CNBC that the company’s near-monopoly on cutting-edge AI chips is under “threat” as OpenAI and other tech giants introduce custom-made semiconductors.
OpenAI announced its first AI chip, Jalapeno, with “industry-leading speed and efficiency” when it unveiled its semiconductor on Tuesday. Google, AWS, and Meta are also developing their own AI chips.
Nvidia has seen its stock soar amid a data center ramp-up, creating huge demand for the company’s chips, both for model training and inference, the way AI systems perform everyday tasks.
But hyperscalers and AI companies are increasingly developing their own silicon to power AI systems.
The Jalapeño chip, designed for inference, shows that “chips with hyperscalar designs can match or exceed Nvidia’s Blackwell-class GPUs in inference efficiency,” Adrien Sanchez, a technology analyst at Yole Group, told CNBC.
He added that while Nvidia still owns the “vast majority” of AI computing and has the ecosystem locked in to its software platform CUDA, OpenAI’s new chips are “a threat to Nvidia’s inference margins, which is the fastest growing area right now.”
Nvidia has been contacted for comment.
OpenAI announced the first benchmark results with Jalapeño, saying users will get “faster responses, more responsive agents, and more reliable access” as demand increases.
The new chip is being developed as follows broadcom. It is expected to be deployed within OpenAI’s computing infrastructure by the end of the year, and OpenAI said it is already working on the second and third generations of this semiconductor.
lead the industry
OpenAI announced Jalapeño in June, saying it would be “built from the ground up for current and future LLMs across the industry.”
Alexander Hallowell, senior principal analyst at Omdia, told CNBC that the jalapeño is “first and foremost a remarkable achievement in terms of efficiency.”
“In large-scale deployments, this saves on power, cooling and distribution infrastructure, contributing significantly to unit economics,” he added.
TrendForce analyst Fionn Chiu told CNBC that OpenAI’s custom chips could reduce reliance on Nvidia for inference workloads over time.
But for more compute-intensive workloads, such as large-scale model training and frontier AI workloads, “we believe Nvidia GPUs remain important given their broad programmability, performance, software ecosystem, and ability to handle a wide range of workloads,” Chiu said.
Comparing OpenAI’s Jalapeño chip to Nvidia
Research firm Semianalysis said it visited OpenAI’s research lab to benchmark the jalapeño. We found this chip outperformed Blackwell in performance per watt in nearly every scenario tested. However, he added that this comparison is “somewhat incomplete and unfair” because Jalapeno uses newer HBM4 memory. Nvidia’s Rubin platform is a better comparison since it also uses HBM4.
“Jalapeño is really competing with chips like Rubin, which also uses HBM4,” SemiAnalysis analysts said in a Tuesday blog post titled “OpenAI Jalapeño: Outperforming Nvidia Blackwell.”
“The Vera Rubin system is now beginning to ship to customers, but it will be some time before OpenAI delivers more than engineering samples of jalapeños,” the blog post added.
Nvidia’s competitors have an advantage
OpenAI is one of several AI companies looking to build custom silicon.
April saw a number of deals announced for custom AI chips, also known as application-specific integrated circuits (ASICs).
Google has announced a new chip for AI training and inference called a tensor processing unit (TPU). Meta said it has agreed to deploy 1 gigawatt of custom AI chips using Broadcom technology as part of the multi-GW agreement. Anthropic said it has committed to spending more than $100 billion on AWS technology over the next 10 years, including current and future generations of Amazon’s custom AI chip Trainium.
“Omdia expects the volume of custom ASIC chips like Jalapeño to surpass GPUs by 2028, but since GPUs are significantly more expensive, returns will take longer,” said Omidia’s Harrowell.
“This is the biggest competitive threat to NVIDIA because about half of our capital spending on AI infrastructure comes from hyperscale cloud providers that have, or can reasonably be able to have, custom chip programs,” he added.
Startups like Cerebras, SambaNova, D-Matrix, Etched, and Fractile are also developing chips for AI.
OpenAI has been a big purchaser of Nvidia GPUs to train and run huge AI models. Having a proprietary chip could affect that relationship.
Sanchez said AI Labs was “one of the largest single consumers of Nvidia GPUs,” and Jalapeno “specifically raised the stakes for Nvidia’s largest customer relationship.”
