August 25, 2026 · OpenAI · The Verge · TechCrunch
OpenAI shows its Jalapeño chip: why it could change AI speed
OpenAI has published the first results of its own inference chip. The promise is clear: faster responses, more users served per watt and cheaper AI to serve if production scales.
- Own Chip
- Inference
- Speed
- Energy cost
Contrasted sources
OpenAI published the results on August 25, 2026. The Verge y TechCrunch Covered the ad on the same day.
OpenAI said what
According to OpenAI, Jalapeño delivers between 1.5 and 1.9 times more AI work per watt in tested models, and between 1.7 and 3.6 times less latency from end-to-end compared to comparable systems on the InferenceX benchmark by SemiAnalysis.
Why matters
Inference is the moment when a pre-trained model responds to the user or executes steps of an agent. If this part gets simplified and accelerated, it can affect prices, usage limits, response speed, and product availability with many users.
What you shouldn't buy yet
TechCrunch notes an important caveat: the initial rollout would be small toward the end of 2026, with more volume in 2027. It is significant news, but it does not mean every ChatGPT user will automatically notice a change tomorrow.