OpenAI is preparing to produce its own artificial intelligence chip for the first time next year, according to people familiar with the partnership. The project, co-designed with Broadcom, reflects a wider industry push toward custom AI chips as demand for computing power keeps rising.
The chip is not expected to become a product for outside buyers. One person close to the project said OpenAI plans to use it internally, making the effort a direct response to the computing needs behind products such as ChatGPT and the training and operation of AI models.
A move toward more control over compute
OpenAI has become one of the most visible examples of the AI industry’s appetite for hardware. The company was one of the earliest customers for Nvidia’s AI chips and has continued to consume large amounts of Nvidia hardware.
That reliance now sits alongside a second strategy: building a specialized chip with Broadcom. The goal, based on the source reporting, is not to replace the entire market for AI hardware, but to give OpenAI another path to the computing power it needs.
OpenAI chief executive Sam Altman has been vocal about the pressure created by demand from both businesses and consumers. Last month, Altman said the company was prioritising compute “in light of the increased demand from [OpenAI’s latest model] GPT-5” and planned to double its compute fleet “over the next 5 months.”
That statement helps explain why an internal AI chip matters. If a company’s services and model development depend on a steady expansion of compute, then hardware strategy becomes central to product strategy. For OpenAI, the Broadcom project gives it a way to pursue more capacity while reducing some dependence on a single dominant supplier.
Broadcom’s custom chip business gets a major signal
Broadcom’s role came into focus after chief executive Hock Tan discussed a new customer on a call with analysts. Tan said Broadcom had secured a fourth major customer for its custom AI chip business while reporting earnings that topped Wall Street estimates.
Broadcom did not name the customer. However, people familiar with the matter confirmed that OpenAI was the new client. Broadcom and OpenAI declined to comment.
Tan also referred to a mystery new customer committing to $10 billion in orders. He said the deal created “immediate and fairly substantial demand,” with chips for that customer shipping “pretty strongly” from next year.
Those details place the OpenAI partnership within Broadcom’s broader custom AI chip business rather than treating it as a one-off experiment. Last year, OpenAI began an initial collaboration with Broadcom, according to earlier reports, but the schedule for mass production of a successful design had been unclear until this new reporting.
Why custom AI chips are gaining ground
OpenAI is not alone in pursuing specialized silicon. Google, Amazon and Meta have also designed their own chips for AI workloads. The common thread is straightforward: AI models require large amounts of computing power to train and run, and major technology companies are looking for hardware shaped around their own needs.
Custom chips can be part of that answer because they are designed for a specific set of workloads. In OpenAI’s case, the source reporting says the chip is meant for internal use, which suggests the design is tied to the company’s own compute demands rather than a plan to sell hardware to the wider market.
The shift also says something about the shape of the AI infrastructure market. Nvidia continues to dominate AI hardware, and the Big Tech “hyperscalers” remain a significant part of its customer base. But growth in Nvidia’s business has slowed compared with the extraordinary figures seen at the start of the AI investment boom.
That does not mean Nvidia’s role has disappeared. The source makes clear that Nvidia remains dominant. The important change is that large AI customers are also looking at custom silicon as another route to scale.
Market reaction and the Nvidia comparison
Broadcom has benefited from investor interest in this custom chip trend. The prospect that custom AI chips will capture a growing share of the AI infrastructure market has helped push Broadcom’s shares more than 30 percent higher this year. The shares also rose almost 9 percent in pre-market trading in New York on Friday.
Analysts have also been watching the comparison with Nvidia. HSBC analysts have recently said they expect Broadcom’s custom chip business to show a much higher growth rate than Nvidia’s chip business in 2026.
That comparison is important, but it should be read carefully. The source does not say Broadcom is overtaking Nvidia in AI hardware. It says custom AI chips are becoming a stronger part of the market, and Broadcom’s business could grow faster from its own base.
For OpenAI, the immediate issue is compute. For Broadcom, the OpenAI connection highlights demand for custom AI chips. For the wider industry, the partnership shows how the AI hardware market is moving beyond a simple dependence on off-the-shelf supply, even while Nvidia remains the central player.