OpenAI has unveiled Jalapeño, its first in-house AI chip, designed with Broadcom to run the models behind ChatGPT and Codex. Built for inference rather than training, the chip is pitched on efficiency, with OpenAI claiming far better performance per watt than off-the-shelf options. The move cuts its reliance on Nvidia and puts OpenAI alongside Google, Amazon, Microsoft and Meta in building custom silicon. First deployments are expected by year’s end, scaling through 2027.
Related Posts
White House deletes thousands of energy-saving web pages as US heat wave intensifies
As a record heat wave grips the US, the Trump administration’s Department of Energy has deleted roughly 6,000 web pages…
‘Boat overturned within seconds’: Survivor recounts Vietnam tragedy that killed 15 Indians
Fifteen Indian tourists died when their speedboat capsized near Vietnam’s Phu Quoc Island. Survivors recounted the tragic moments as the…
Google’s AI safety team tells job seekers: Skip our AI, talk to a human
Google DeepMind’s safety team uses a special form for job applications. This ensures human review, as AI tools may incorrectly…