OpenAI and Broadcom have unveiled Jalapeño, OpenAI's first custom AI chip designed to run large language model inference workloads. Developed in a rapid nine-month window with assistance from OpenAI's own models, the chip aims to reduce inference costs by roughly 50% compared to standard GPUs. The move comes as OpenAI attempts to curb massive financial losses, which reached an operating deficit of $20.92 billion in 2025 due to soaring compute costs, while seeking to reduce its reliance on Nvidia.
Jalapeño chip technical specifications
- ▪The Jalapeño chip is being tested in OpenAI's labs with the GPT-5.3-Codex-Spark AI model and is operating at target power and performance
- ▪Celestica will build the server systems, boards, and racks for the Jalapeño chips, which will be used exclusively by OpenAI
- ▪Early testing of the Jalapeño chip shows cost savings of roughly 50% compared with typical AI graphics processing units
- ▪OpenAI and Broadcom unveiled Jalapeño, a custom artificial intelligence chip designed specifically for large language model inference to run AI workloads faster and cheaper
ASIC development timeline
- ▪Broadcom CEO Hock Tan stated that the Jalapeño chip will begin to ramp up in 2027 and go full tilt in the first half of 2028
- ▪OpenAI and Broadcom completed the design of the Jalapeño chip in nine months, utilizing OpenAI's own AI models to accelerate the process
- ▪OpenAI plans to begin initial deployment of the Jalapeño chip in data centers by the end of 2026, with a full rollout projected by late 2029
OpenAI financial losses
- ▪OpenAI's research and development costs, driven largely by model training and serving infrastructure, accounted for $19.18 billion in 2025
- ▪OpenAI paid Microsoft over $10.59 billion for research and development and compute infrastructure in 2025
- ▪OpenAI generated $13.07 billion in revenue in 2025 but incurred $34 billion in total operational expenses, resulting in an operating loss of nearly $20.92 billion
Existing chip partnerships
- ▪Nvidia finalized a $30 billion direct investment into OpenAI in February 2026, securing an agreement to deploy 10 gigawatts of computing systems
- ▪Amazon invested $50 billion into OpenAI in February 2026, which included a commitment for OpenAI to use two gigawatts of AWS Trainium computing capacity over eight years
- ▪OpenAI has signed agreements to use hardware from Advanced Micro Devices and Cerebras, the latter of which held its initial public offering in May 2026
Global AI chip competition
- ▪ByteDance entered negotiations with Qualcomm in June 2026 to design custom application-specific integrated circuits for its data centers
- ▪Google, Amazon, Meta, and Microsoft have developed their own custom AI accelerators, such as Google's TPUs, Amazon's Trainium, Meta's MTIA, and Microsoft's Azure Maia
- ▪Chinese technology firms are developing custom AI hardware, including Alibaba's Zhenwu M890 chip and Huawei's upcoming Ascend 950DT chip
Story comments
Loading comments…