Broadcom IncBroadcom's AI semiconductor revenue jumped 143% to $10.8 billion, and it handles physical implementation and networking for OpenAI's new chip.
OpenAI has unveiled its first homegrown inference chip, Jalapeño, which it says can perform up to 1.9 times more AI work per watt than comparison systems, and it went from initial design to tape-out in just nine months. Broadcom handled the chip's physical implementation and supplied Tomahawk networking, while Celestica managed board, rack, and system integration, and TSMC manufactured the chip. In published InferenceX results, Jalapeño delivered 1.5 to 1.9 times more work per watt and 1.7 to 3.6 times lower end-to-end latency across three large open models, compared with Nvidia GB200 and GB300 systems. However, Jalapeño is an inference chip, not a training accelerator, and was not compared with Nvidia's Vera Rubin; initial deployment is expected by year-end, so production economics remain unproven. Broadcom's AI semiconductor revenue jumped 143% to $10.8 billion last quarter, and the companies plan to deploy 10 gigawatts of OpenAI-designed accelerators through 2029, though bears cite customer concentration and margin pressure. Hedge funds remain heavily invested in both Broadcom and Nvidia, with short interest at just 1.20% and 1.23% of float, respectively.
Broadcom IncBroadcom's AI semiconductor revenue jumped 143% to $10.8 billion, and it handles physical implementation and networking for OpenAI's new chip.
NVIDIA CorporationOpenAI's Jalapeño chip outperforms Nvidia's GB200 and GB300 in inference per watt and latency, posing a competitive threat.
Taiwan Semiconductor Manufacturing Co. Ltd.TSMC manufactures the Jalapeño chip, benefiting from OpenAI's production orders.
Celestica Inc.OpenAI unveiled its first homegrown inference chip, Jalapeño, with superior performance and rapid development.