OpenAI is designing its own AI accelerator chip in partnership with Broadcom, marking a major vertical-integration play by a leading AI lab into semiconductor design.
On June 24, OpenAI and Broadcom unveiled Jalapeño, OpenAI's first custom AI chip. Designed by the maker of ChatGPT and built with Broadcom, it represents the first piece of a sweeping plan the two companies unveiled last fall to deploy 10 gigawatts of OpenAI-designed accelerators between the second half of 2026 and the end of 2029. Some interpreted the announcement as a direct strike at Nvidia, prompting the natural question: should Nvidia shareholders be concerned?
The answer is a measured yes—but probably less than headlines suggest.
Broadcom's custom-silicon business is experiencing explosive growth. In its fiscal second quarter of 2026 (ended May 3, 2026), the company's AI semiconductor revenue jumped 143% year over year to $10.8 billion. Management reported booking more than $30 billion of AI orders during the quarter alone, signaling substantial future demand. Broadcom reaffirmed that its AI chip revenue should exceed $100 billion in fiscal 2027, roughly double current-year expectations. "Demand for XPUs and networking is simply insatiable," said Broadcom CEO Hock Tan, referring to the custom accelerators Broadcom co-designs with major customers.
OpenAI is one of several deep-pocketed customers building custom silicon. Broadcom also designs chips for Alphabet's Google and Meta Platforms. Across the industry, major AI buyers face the same calculation: designing silicon tailored to their workloads while reducing reliance on Nvidia's expensive graphics processing units offers compelling economics. Over time, this could pressure Nvidia's pricing power and the roughly 75% gross margins it maintains.
Yet Nvidia operates on a different scale entirely. In its fiscal first quarter of 2027 (ended April 26, 2026), Nvidia's revenue rose 85% year over year to $81.6 billion, with data center revenue climbing 92% to $75.2 billion. Nvidia generates more AI hardware revenue in a single quarter than Broadcom projects to make from its entire AI chip business over a full year.
Jalapeño is an application-specific chip built for inference—the step where a trained model answers queries—and was taped out in approximately nine months. Application-specific chips can be cheaper and more power-efficient for narrow workloads, but they lack the flexibility of Nvidia's general-purpose processors and don't support CUDA, the software layer that anchors developers to Nvidia. A custom inference chip is largely additive to Nvidia's business: it absorbs a slice of OpenAI's inference workload, while training and much of the remaining workload continues on Nvidia hardware.
Market growth dynamics also favor both suppliers. Nvidia guided for fiscal second-quarter revenue of $91 billion, up from $81.6 billion, indicating that AI demand continues expanding faster than any single competitor can capture. When markets grow this rapidly, custom silicon and Nvidia's chips can both gain share; the question is how spoils divide as the industry matures, not whether Nvidia stops growing.
Trading around $194 per share, Nvidia trades at approximately 22 times forward earnings—a significant decline from a year earlier, suggesting the market has already incorporated some erosion of dominance. Broadcom trades at a forward price-to-earnings ratio in the low 30s, showing that enthusiasm for custom silicon is no secret to investors.
Custom chips from Broadcom and its customers merit attention: they're performant, scaling quickly, and will chip away at Nvidia's pricing power as the AI buildout matures. But "chip away" is the operative phrase. Nvidia's advantages in performance, software, and sheer scale remain enormous, and overall spending is accelerating. The nearer-term risk to Nvidia arguably stems more from shifts in AI sentiment than from a homegrown chip built by a customer that continues buying Nvidia hardware for most of its workload.
For now, Nvidia's competitive moat is narrowing at the edges—not closing.