OpenAI's Jalapeño Chip Challenges Nvidia as Meta's AI Plan Backfires

OpenAI has unveiled Jalapeño, its first custom AI inference chip, and the early results could challenge one of Nvidia's biggest advantages.

OpenAI says Jalapeño delivers between 1.5 and 1.9 times more AI work per watt than the systems it tested against. SemiAnalysis watched benchmark runs inside OpenAI's lab and reported strong performance-per-watt results against Nvidia Blackwell. But the bigger story may be software: OpenAI is using Codex to write optimized kernels, potentially weakening the nearly 20-year CUDA software moat that has helped protect Nvidia from competitors.

Also on this episode of Hashtag Trending:

A Reuters investigation reveals how Meta's ambitious AI-driven restructuring plan fell apart. AI-assisted code increased dramatically, but gains in actual product features were far smaller, while technical incidents increased and employee morale deteriorated.

Canada is investing more than $500 million in 64 new research chairs as it tries to turn decades of brain drain into brain gain. The real test will be whether Canada can keep the intellectual property, companies and economic benefits created by those researchers.

And Twitter is back. Sort of. Operation Bluebird has launched a new social network using the Twitter name and the familiar language of tweets and retweets. Elon Musk's X says it still owns the trademarks and is suing.

Hashtag Trending is hosted by Jim Love.

00:00 Headlines
00:27 OpenAI unveils the Jalapeño AI chip
01:39 Can OpenAI challenge Nvidia's CUDA moat
03:02 Meta's AI restructuring plan backfires
04:50 Canada makes a major brain-gain push
06:33 Can Canada turn research into Canadian industry
07:38 X battles a new company called Twitter
09:18 Wrap-up