OpenAI presented benchmarks at Hot Chips on Tuesday, claiming its new Jalapeño chip beats Nvidia's GB300 GPU on speed and power efficiency. Anyone tracking AI hardware should care: OpenAI is now positioning itself as a chipmaker, not just a chip buyer, which could reshape supply chains for AI and future devices.
What actually happened
According to Tom's Hardware, OpenAI unveiled its first in-house chip, code-named Jalapeño, at the Hot Chips conference on Tuesday. The company claims the chip delivers up to 1.9 times the throughput per kilowatt compared with Nvidia's GB300 GPU. OpenAI also claims 3.6 times lower latency. The Jalapeño chip runs at 700W, versus 1,400W for Nvidia's flagship GB300, according to the report. The chip was co-developed with Broadcom, an established semiconductor manufacturer. These are OpenAI's first published benchmarks for an in-house chip design, per the outlet. OpenAI has not disclosed a release timeline or pricing for the Jalapeño chip, per available reporting.
How we got here
Hot Chips is an annual conference where semiconductor companies present technical benchmarks to engineers and researchers, according to the report. OpenAI has not previously published performance data for an in-house chip design. The Jalapeño chip project involved Broadcom, a semiconductor manufacturer, as a development partner. Nvidia's GB300 is described as Nvidia's current flagship GPU for AI workloads. The comparison marks the first time OpenAI has released benchmark data pitting its own silicon against Nvidia's leading hardware, a signal of movement beyond software into chip design.
Why this matters for you
For builders integrating AI into devices, lower power draw per unit of throughput could mean cheaper, cooler-running hardware for on-device AI, relevant to future AR and smart glasses that need efficient chips. For Nvidia investors, a credible rival raises competitive risk for GB300 sales. For OpenAI, controlling chip design could reduce dependence on Nvidia and lower long-term compute costs, which could affect pricing of OpenAI's API and consumer products. Users of AI-powered apps might eventually see faster response times if OpenAI deploys Jalapeño at scale.
The bigger question
OpenAI's benchmarks come from the company itself, not from independent third-party testing labs. If outside researchers confirm the claimed 1.9 times efficiency gain and 3.6 times latency drop, will other AI companies accelerate their own custom chip programs? And if Nvidia's dominance in AI chips weakens over time, who benefits most: chipmakers, cloud providers, or the developers building the next generation of AI hardware and devices?
What to watch
Watch for independent verification of these benchmark claims from outside researchers or rival chipmakers. Watch for any response from Nvidia. No public release date or pricing for the Jalapeño chip has been announced yet. Bonuz will track how power-efficient AI chips could affect future on-device compute, including AR and smart glasses hardware.
