OpenAI has signed a major deal with Broadcom to produce its first in-house AI processors to secure the computing power it needs as demand for its services accelerates.
The companies announced on Monday that OpenAI will design the chips, while Broadcom will handle development and deployment. Production is set to begin in the second half of 2026.
They plan to roll out 10 gigawatts of custom AI chips. This amount of power equals the electricity needed by more than 8 million U.S. households.
It is also about five times the power output of the Hoover Dam. All the new systems will run on Broadcom’s Ethernet and other networking gear.
This will create a direct challenge to Nvidia’s InfiniBand technology, which currently dominates AI data centers. Broadcom’s stock rose by more than 10% after the announcement.
Infrastructure
OpenAI’s partnership with Broadcom is part of a strategy to build a stronger infrastructure base.
For years, OpenAI has relied heavily on Nvidia’s powerful chips to train and operate its models.
But demand for these chips has surged, creating supply bottlenecks and raising costs.
OpenAI’s intended self-designed chips will give it control over performance, availability, and expenses.
It will also place the company alongside major cloud providers such as Google and Amazon, which have developed their own chips to reduce dependence on Nvidia.
Industry Impact
Analysts say this deal will not immediately threaten Nvidia’s dominant position in the AI accelerator market.
Designing, scaling, and manufacturing chips from the ground up is complex and expensive.
Other tech firms, including Microsoft and Meta, have tried to build in-house chips but have not matched Nvidia’s performance, according to media reports.
Even so, the demand for custom processors has made Broadcom one of the biggest beneficiaries of the AI boom.
Its stock has risen almost six-fold since the end of 2022. And in September, Broadcom announced a $10 billion order for custom AI chips from a new customer.
Analysts speculated that the buyer might be OpenAI. On Monday, a company executive said OpenAI was not that client.
The Build-Out Timeline
OpenAI’s build-out plan is ambitious. The company expects to complete deployment of the new custom chips by the end of 2029.
According to Nvidia CEO Jensen Huang, constructing a single one-gigawatt data center can cost between $50 billion and $60 billion.
Nvidia’s products account for more than half of these costs. Financing such a large project will likely involve multiple strategies.
These may include funding rounds, pre-orders, strategic investments, credit facilities, and support from Microsoft.
Analyst Gadjo Sevilla of eMarketer noted that OpenAI is well-positioned to raise the necessary capital because of investor confidence.
Major Chip Deals
The Broadcom partnership comes after a series of large chip investments by OpenAI. Last week, the company announced a 6-gigawatt AI chip supply deal with AMD.
The agreement also gives OpenAI the option to buy a stake in the chipmaker.
Just days before, Nvidia unveiled plans to invest up to $100 billion in OpenAI and supply it with data-center systems capable of at least 10 gigawatts.
These deals reflect the intense race among technology companies to secure the computing power required for increasingly advanced AI models.
OpenAI’s Strategic Position
“Partnering with Broadcom is a critical step in building the infrastructure needed to unlock AI’s potential,” OpenAI CEO Sam Altman said.
This collaboration places OpenAI among a growing group of technology leaders working to build custom chips.
If successful, the company will reduce its reliance on Nvidia, gain more control over its operations, and strengthen its long-term strategic position in the global AI market.
