Take note, Nvidia: OpenAI is tightening its grip on the AI economy. With a new partnership, the company is targeting the single biggest bottleneck in its supply chain: the chip.
According to a press release published Monday, OpenAI has struck a deal with Broadcom to co-develop its first in-house AI processor. The move is part of an effort to reduce reliance on third parties, such as Nvidia, and exert greater control over the infrastructure that drives its models.
“Partnering with Broadcom is a critical step in building the infrastructure needed to unlock AI’s potential,” OpenAI CEO Sam Altman said in a statement. “Developing our own accelerators adds to the broader ecosystem of partners all building the capacity required to push the frontier of AI to provide benefits to all humanity.”
This news comes just a week after OpenAI signed a multi-year partnership with chipmaker AMD to supply the processors powering its next-generation AI systems.
10 gigawatts of compute power
Under the agreement, OpenAI will focus on chip design and architecture, while Broadcom brings the hardware manufacturing and deployment expertise. Production is expected to ramp up in 2026, with early chips potentially handling inference workloads — the process of running trained AI models at scale.
The goal is staggering: an infrastructure footprint of up to 10 gigawatts of compute, roughly equivalent to the electricity demand of several million US homes. Analysts say that scale could allow OpenAI to run its largest models faster and more efficiently — a crucial advantage as training costs balloon across the industry.
The partnership also reflects a broader trend among tech giants like Google, Amazon, and Meta, each of which has pursued in-house chip design to avoid supply shortages and optimize for their own workloads.
OpenAI has not yet revealed the financial details of the agreement.
The gamble ahead
For all its promise, the path forward is steep. Chip design and manufacturing are notoriously complex, expensive, and prone to delays. Even with Broadcom’s expertise, integrating new hardware at OpenAI’s scale will be a high-wire act.
Analysts warn that execution risk is significant: costs could spiral, timelines could slip, and new architectures could underperform against Nvidia’s well-honed GPUs. Yet, if OpenAI and Broadcom succeed, they could spark a shift across the industry — away from off-the-shelf graphics chips and toward custom-built silicon optimized for intelligence itself.
In that scenario, OpenAI won’t just be training the world’s most advanced AI models. It’ll be powering them with a brain it built from the ground up, which could be a fitting evolution for a company whose ambitions have always stretched from code to creation.
For more on how OpenAI is pushing its technology forward, read how the tech giant trained to beat the world’s best coders.