The first chip OpenAI designed is already running in production, but the smart money is betting on the companies that didn't build it.
The Summary
- OpenAI and Broadcom launched Jalapeño, OpenAI's first custom AI chip designed specifically for LLM inference, and it's already handling customer queries in production
- Despite the headlines, institutional money flow shows large investors quietly favoring rivals like Micron and AMD over Broadcom
- The chip marks OpenAI's first move toward controlling its own hardware stack, reducing dependency on Nvidia while Broadcom gets manufacturing revenue but not investor confidence
The Signal
OpenAI and Broadcom built Jalapeño in record time, an LLM-focused accelerator now processing real ChatGPT traffic. This is not a lab experiment. The chip is already deployed for customer queries, making it one of the fastest custom silicon projects to reach production in AI history. For context, Google spent years developing TPUs before they handled meaningful workloads. OpenAI compressed that timeline.
The speed matters because inference costs are the hidden tax on AI scale. Every ChatGPT response burns silicon. Multiply that by millions of users, and you see why OpenAI needed hardware designed specifically for LLM inference rather than continuing to rent Nvidia's training-optimized GPUs at premium rates.
"The chip marks OpenAI's first step toward controlling the hardware stack behind ChatGPT and beyond."
But here's where the story turns. Broadcom's positioning data and money flow tell a different story than the partnership announcement. Institutional investors are not piling into AVGO. They're rotating into Micron and AMD. The relative strength indicators show capital moving away from Broadcom even as it ships OpenAI's first custom chip.
Why? Three possibilities:
- Broadcom is a fab partner, not an innovation play. They build what others design.
- The margins on custom AI chips are thinner than memory or compute where Micron and AMD operate.
- Large investors see Jalapeño as OpenAI reducing dependency on the entire chip supply chain, not enlarging it.
The Broadcom-OpenAI partnership positions both companies in the AI infrastructure race, but the market is pricing in a future where the real money flows to commodity components and specialized compute, not custom design-for-hire. Broadcom gets the headline. AMD and Micron get the inflows.
The Implication
Watch the next wave of AI companies. If OpenAI can go from zero to production silicon this fast, every major AI lab will try. That means more custom chips, more fragmentation, and a growing market for the companies that make the building blocks everyone needs: memory, packaging, interconnects. The headline partnerships matter less than the infrastructure bets happening in the background.
If you're tracking where AI infrastructure money actually goes, follow the capital flows, not the press releases. Broadcom built the chip. The market bought the picks and shovels.