OpenAI unveils its first custom inference chip, built by Broadcom
TechCrunch reports that OpenAI has introduced its first custom chip, called Jalapeño, designed for the company’s inference workloads and built by Broadcom. The announcement suggests OpenAI is moving beyond reliance on off-the-shelf AI hardware by developing silicon tailored to its own serving infrastructure.
Why it matters: A custom inference chip from OpenAI would be a notable strategic shift in AI infrastructure. Purpose-built silicon can affect cost, performance, supply-chain leverage, and dependence on dominant GPU vendors, making it important for both OpenAI’s economics and the competitive structure of the AI compute stack.
Sources
- OpenAI unveils its first custom chip, built by Broadcom TechCrunch · June 24, 2026
Related stories
Independent events that offer a meaningful comparison, without implying that one caused the other.
-
Google reportedly weighs Marvell against Broadcom for AI chip supply as Marvell offers a $12.2 billion stake deal
In August, The Register reported Google was weighing Marvell against Broadcom for AI chip supply while Marvell proposed a $12.2 billion stake deal, offering a comparison on how hyperscalers may use supplier competition to reshape bargaining power in custom AI silicon.
-
AMD acquires Taalas to develop model-specific AI inference chips
AMD’s acquisition of Taalas adds a separate comparison point focused on model-specific inference chips, showing another route toward specialized serving hardware through acquisition rather than an in-house chip launch.
Connections
Context and precedents—not claims of causation or corroboration.
-
Cerebras unveils CS-4 rack systems with higher chip performance and denser configurations
As context for OpenAI's custom inference chip launch, Cerebras' later CS-4 rack systems provide a comparison from another part of the hardware stack, focused on higher per-chip performance and denser rack-level AI compute deployments.