Broadcom and OpenAI have jointly unveiled the Jalapeño custom AI inference chip designed to accelerate OpenAI’s next-generation models.
Broadcom Inc. and OpenAI have unveiled Jalapeño, OpenAI’s first custom Intelligence Processor designed specifically for large language model (LLM) inference. Co-developed from initial design to manufacturing tape-out in nine months, the processor utilizes OpenAI models to optimize its design cycle. Engineering samples are currently running machine learning workloads, including GPT-5.3-Codex-Spark. Celestica Inc. assisted with board and rack system integration.
The technical design of Jalapeño emphasizes minimizing data movement, a critical factor in improving overall system efficiency. By better integrating compute, memory, and networking resources, the processor seeks to maximize utilization rates. This approach is intended to bridge the gap between theoretical performance capabilities and actual realized performance in production environments, focusing on balancing resources to achieve realized utilization closer to theoretical peak performance.
Jalapeño marks the first step in a multi-generation compute platform planned for initial deployment by the end of 2026. "By co-developing our industry-leading silicon directly with OpenAI, we are enabling the deployment of gigawatt-scale data centers with Microsoft and other partners beginning in 2026," stated Broadcom President and CEO Hock Tan. Early testing indicates the chip delivers performance per watt substantially better than current state-of-the-art accelerators. "Based on early testing, Jalapeño will efficiently execute our most important workloads close to the hardware’s theoretical limits," said Richard Ho, leader of OpenAI’s hardware program.
Broadcom has quietly emerged as a key partner for hyperscalers and AI companies seeking specialized hardware tailored to specific workloads. Unlike NVIDIA, which sells its own accelerators, Broadcom’s opportunity lies in helping customers build theirs. The company’s role in the project includes providing silicon implementation and networking technologies, such as Tomahawk networking silicon, to facilitate the platform's large-scale production. The shift toward custom silicon suggests the industry’s largest AI companies increasingly want hardware optimized for their own models and workloads rather than relying entirely on off-the-shelf chips. Alphabet Inc. has spent years developing its Tensor Processing Units, or TPUs. Amazon.com, Inc. built Trainium and Inferentia. Microsoft Corp. introduced Maia. Meta Platforms, Inc. has been developing its own MTIA accelerators. Now OpenAI has joined the list.
JPMorgan has reiterated its 'Overweight' rating on Broadcom Inc., urging investors to buy the dip following a 2.22% decline over the last month. The bank dismissed rumors of delays in the TPU v9 2nm program with Alphabet Inc.'s Google, citing a five-year agreement that locks in increasing TPU revenues through 2031. Analysts project significant AI revenue growth for 2027 and 2028, maintaining a December 2026 price target of $580.00.
In a research note titled 'Ignore The Noise', JPMorgan analysts Harlan Sur and Mayur Ramdhani pushed back against reports suggesting the joint Broadcom-Google next-generation artificial intelligence program was in jeopardy. They emphasized that contrary to recent noise regarding delays or cancellations of the TPU v9 2nm program, the team remains on track. They stated the program is proceeding with "NO delays; NO cancellations," reinforcing Broadcom's execution track record. According to JPMorgan, Broadcom maintains an "18mos+ lead" over Google’s internal custom chip-design team.
The firm highlighted a five-year agreement signed in March that locks in Broadcom’s TPU design win roadmap for the next four generations of TPU (v8, v9, v10, v11). Crucially, the bank emphasized that this pact "includes commitments for increasing TPU revenues on an annual basis through 2031," ensuring a predictable and expanding pipeline of AI revenue. JPMorgan expects Broadcom's AI revenue to rise two to 2.5 times in 2027 and double again in 2028.
From a trend perspective, Broadcom is in a constructive long-term setup, trading 10.7% above its 100-day simple moving average (SMA) and 13.2% above its 200-day SMA. However, the stock is working through a shorter-term digestion phase, sitting 1.7% below its 20-day SMA and 0.8% below its 50-day SMA. The Relative Strength Index (RSI) is at 46.68. Shares of AVGO have advanced by 18.85% year-to-date, closing 4.70% higher at $411.35 on Thursday. JPMorgan reiterated its "Overweight" rating with a December 2026 price target of $580.00.