OpenAI's new Jalapeño chip announced at Hot Chips beats Vera Rubin's July results on Output Throughput per MW. (1/7)🧵
On performance/TCO, Jalapeño is roughly competitive with Rubin's July results, with significant upside once speculative decoding is added. (2/7)
Codex helps design chip blocks and automatically writes/tunes thousands-line kernels, dramatically accelerating hardware and software bring-up. (3/7)
If production ramps successfully through 2027, Jalapeño could materially weaken Nvidia's CUDA moat and reshape OpenAI’s dependence on Nvidia, AMD, and Cerebras. (4/7)
speculative decoding — A speed trick where a small model drafts several tokens ahead and the big model verifies them in one pass, often doubling generation speed.
inference — Running a trained model to get answers — the phase where AI is actually used, as opposed to trained.
Why it matters
If Jalapeño ramps through 2027 it erodes Nvidia's CUDA lock-in for inferenceRunning a trained model to get answers — the phase where AI is actually used, as opposed to trained.Full definition → — and Codex authoring production kernels is a concrete data point on agents in hardware bring-up.
Read the full thread on X ↗
