OpenAI is reportedly changing how it thinks about the hardware behind its AI systems. Instead of pursuing a network of factories for chip manufacturing, the company is said to be concentrating on in-house chip design while widening the set of chips it uses for training.
According to Reuters, OpenAI has been working with TSMC and Broadcom on an AI chip of its own. The reported chip is aimed at running models and could arrive as soon as 2026.
A shift from factories to chip design
The most important change in the report is not only that OpenAI is exploring its own AI chip. It is that the company has, at least for now, abandoned plans to establish a network of factories for chip manufacturing.
That distinction matters. Building factories and designing chips are different paths, with different levels of complexity. The Reuters report says OpenAI is focusing on in-house chip design rather than trying to create a manufacturing network.
In practical terms, that means OpenAI would be trying to shape the hardware it needs without taking on every part of the chip production process. The source does not describe the full design, the expected performance, or the business model. What it does say is that OpenAI has been working with Broadcom for months to create an AI chip for running models.
What the reported OpenAI AI chip is for
The chip described in the report is meant for running models. That is a narrower and clearer purpose than simply saying OpenAI wants a chip because chips are important.
Running models is one of the core activities of an AI company. A custom AI chip could give OpenAI more control over the hardware used for that task, if the project reaches production. Reuters reports that the chip could arrive as soon as 2026, which frames the effort as a future plan rather than something already in use.
The companies named in the report also show how OpenAI is approaching the effort. TSMC and Broadcom are part of the reported work around the in-house AI chip. The source does not add further detail about each company's exact role, so the clearest takeaway is that OpenAI is not described as acting alone.
Why AMD enters the training picture
The report also points to a second hardware move: OpenAI plans to use AMD chips through Microsoft's Azure cloud platform for model training.
That matters because Reuters says OpenAI previously relied almost entirely on Nvidia GPUs for training. The shift does not mean Nvidia disappears from the picture. The source says AMD chips would be used alongside Nvidia's, which makes this a diversification story rather than a full replacement story.
The reasons given are also specific. Reuters connects the move to chip shortages and delays, as well as the high cost of training. Those pressures have pushed OpenAI to explore alternatives.
Put simply, OpenAI appears to be looking for more flexibility in the hardware stack. The reported plan includes both a future in-house AI chip for running models and the use of AMD chips through Azure for training.
The broader hardware strategy
Taken together, the reported moves suggest a hardware strategy with two tracks.
- Design control: OpenAI is reportedly working on an in-house AI chip with TSMC and Broadcom.
- Training options: OpenAI plans to use AMD chips through Microsoft's Azure cloud platform alongside Nvidia GPUs.
- Manufacturing restraint: OpenAI has reportedly stepped back from plans to establish a network of factories for chip manufacturing.
The logic is straightforward. AI companies need large amounts of computing power to train and run models. If the available supply is constrained, delayed, or expensive, relying almost entirely on one kind of hardware becomes a strategic risk.
The Reuters report does not say whether OpenAI's chip project will succeed, how much it will cost, or how much it will change the company's dependence on outside chips. It does show that OpenAI is actively looking beyond its previous near-total reliance on Nvidia GPUs for training.
What to watch next
The timeline in the report gives the clearest marker: the in-house AI chip could arrive as soon as 2026. Until then, the more immediate change may be OpenAI's plan to use AMD chips through Microsoft's Azure cloud platform.
For OpenAI, the reported effort is about reducing constraints around AI computing. For the wider AI market, it signals that chip access, chip cost, and chip availability remain central to how advanced models are built and deployed.
The key point is not that OpenAI is leaving existing suppliers behind. The report says the company is adding options: working on an AI chip of its own, using AMD chips through Azure, and continuing to train with Nvidia GPUs. That combination is the real story.