AVGO
New chip, same AI money chase
OpenAI just showed off Jalapeño, its first custom inference chip, and the pitch is very on-brand for the AI boom: more work per watt, lower latency, and lower costs for serving models. The company says it’ll start rolling the chip into its own infrastructure by the end of 2026, with production ramping further in 2027.
Where Broadcom fits in
This is the part investors care about: Broadcom is the partner helping bring the chip to life. That puts AVGO squarely in the custom silicon club, the place where hyperscalers and AI labs go when they get tired of paying premium prices for off-the-shelf horsepower.
OpenAI claims Jalapeño outperformed Nvidia GB200 and GB300-based comparison systems across several benchmarks, including better performance per watt and lower end-to-end latency. That’s not exactly a “Nvidia is done” moment — more like a reminder that the AI stack is becoming a lot less monogamous.
So is Nvidia sweating?
Jim Cramer’s answer: not really. He basically shrugged at the idea that every new chip challenger suddenly means Nvidia is losing its crown, and honestly, that tracks with how this market has worked so far. Nvidia still gets the giant share of the attention, the dollars, and the “tell me more” investor obsession.
Big picture: this is another signal that custom AI silicon is becoming a real business, not a science fair project. Broadcom gets more credibility in the AI hardware race, while Nvidia keeps its heavyweight belt — at least for now.