What's happening
OpenAI and Broadcom announced Jalapeño on June 24, 2026 — the first custom AI inference accelerator chip produced through their partnership and the inaugural entry in what both companies describe as a multi-generation silicon platform. The chip was developed from initial design through tape-out in nine months, with chip samples already running GPT-5.3-Codex-Spark at target frequency and power. Manufacturing is handled by TSMC, while Celestica is serving as the partner for board and rack systems integration. Initial deployments are planned before the end of 2026, targeting 10 gigawatts of power capacity at gigawatt scale.
The announcement was made with participation from OpenAI CEO Sam Altman, President and Co-Founder Greg Brockman, and hardware chief Richard Ho on the OpenAI side, alongside Broadcom President and CEO Hock Tan and President Charlie Kawwas. Broadcom's Hock Tan stated that Jalapeño is designed to match Nvidia's Blackwell architecture and Google's tensor processing units in performance. Early testing results, according to the companies, show performance per watt that is substantially better than current state-of-the-art offerings. Greg Brockman noted: "The degree to which our models have been able to accelerate it was very surprising to us."
Why it matters for markets
Jalapeño represents OpenAI's first concrete step toward reducing its dependence on third-party GPU suppliers — most notably Nvidia — by developing proprietary silicon optimized for its own inference workloads. Nvidia currently carries a market capitalization of $4.66 trillion and reported revenue of $253.49 billion, with its data center GPU lineup, including the Blackwell architecture, serving as the dominant compute platform for AI training and inference across the industry. A sustained shift by a customer of OpenAI's scale toward custom silicon could alter demand dynamics for high-end GPU procurement, though the degree and timeline of any such displacement remains contingent on Jalapeño's performance at production scale.
For Broadcom, the announcement reinforces its position as the leading merchant silicon partner for hyperscale custom AI accelerator programs. The company, which carries a market capitalization of $1.74 trillion and reported revenue of $75.46 billion, has seen its shares rise nearly seven times since the end of 2022 and approximately 10% in 2026 alone. Broadcom shares climbed further following the June 24 announcement. Hock Tan framed the partnership in structural terms: "Our collaboration with OpenAI represents a fundamental commitment to scaling the physical infrastructure required for the next decade of AI." Greg Brockman added broader context: "The world is moving to a compute-powered economy."
The planned deployment scale — targeting 10 gigawatts of power capacity — underscores the capital intensity of the buildout and the potential revenue magnitude for Broadcom as the chip designer and TSMC as the manufacturer. The multi-generation platform structure also signals that this is not a one-cycle engagement but an ongoing, deepening supply relationship between OpenAI and Broadcom, with successive chip generations implied.
Sectors and assets to watch
The primary tickers directly implicated are Broadcom (NASDAQ: AVGO), as the chip design partner and a central beneficiary of the program's expansion, and Nvidia (NASDAQ: NVDA), whose Blackwell GPU architecture was explicitly cited by Broadcom's CEO as the performance benchmark Jalapeño is designed to meet. Nvidia's data center GPU business has been the dominant revenue driver behind its $253.49 billion in annual revenue and $4.66 trillion market capitalization, making any credible custom silicon alternative from a major AI lab a development worth monitoring for its longer-term demand implications. Nvidia's P/E ratio of 29.4 reflects a valuation profile that incorporates continued data center GPU dominance, a premise that custom inference silicon programs like Jalapeño could test over successive product generations.
Beyond the two primary tickers, the supply chain participants named in the announcement are also relevant to watch. TSMC, as the manufacturing partner for Jalapeño, stands to capture foundry revenue from the program's gigawatt-scale deployment ambitions. Celestica, named as the partner for board and rack system integration, occupies a direct role in the physical infrastructure buildout. More broadly, the custom ASIC segment — in which Broadcom competes alongside players such as Marvell Technology — may see increased strategic interest from AI labs and cloud providers seeking inference-optimized alternatives to general-purpose GPU platforms.
What to watch next
Key developments to monitor include whether Jalapeño deployments proceed on the stated end-of-2026 timeline and whether the 10-gigawatt power target is met at scale, as these milestones will determine the program's near-term commercial significance. Subsequent generations of the multi-generation platform — and any disclosures about their design specifications or timelines — will indicate the depth and durability of the OpenAI-Broadcom silicon partnership. Nvidia's response, whether through accelerated Blackwell successor announcements, pricing adjustments, or new inference-specific product lines, will be a critical variable. Additionally, any disclosures from OpenAI regarding the proportion of its inference workloads migrating to Jalapeño versus continued GPU procurement will provide a clearer picture of the competitive displacement dynamic between custom silicon and merchant GPU architectures.