OpenAI is moving closer to owning more of the hardware stack behind its AI systems. According to a Reuters report released Monday, the ChatGPT creator is in the final stages of designing a long-rumored AI processor intended to reduce its reliance on Nvidia hardware.
The chip has not been formally announced. Its full capabilities, technical specifications, and exact schedule remain unknown. But the reported plan points to a clear strategic goal: OpenAI wants more control over the infrastructure that powers its models.
A custom chip built around leverage
OpenAI reportedly plans to send its chip designs to Taiwan Semiconductor Manufacturing Co. (TSMC) for fabrication within the next few months. TSMC is also the company that produces Nvidia’s chips, which makes it a central player in the AI hardware supply chain.
The near-term goal is not necessarily to replace Nvidia overnight. The more immediate value may be strategic. A chip design OpenAI controls could give the company more leverage when negotiating with suppliers, while also creating a path toward greater independence over time.
That matters because Nvidia has become the dominant supplier of high-powered GPUs for data center use, including hardware such as the Blackwell series. For companies building large AI systems, access to these processors is a major operational concern.
OpenAI is not alone in exploring custom silicon. Microsoft, Amazon, Google, and Meta have also built AI acceleration chips, with motivations that include lowering costs and easing shortages in Nvidia-supplied hardware.
What the first OpenAI AI processor is expected to do
The first version of OpenAI’s chip will reportedly be aimed mainly at running AI models, a workload often called “inference.” That is different from training, the process of building or improving models, which typically requires very large amounts of compute.
Focusing first on inference would make sense for limited deployment across the company. Running models is a recurring demand for a chatbot maker, and even a partial shift in that workload could help OpenAI test its own hardware in real operating conditions.
The chip is expected to include high-bandwidth memory and networking features similar to those found in Nvidia’s processors. Those details suggest the project is not just about making a generic processor. It is being designed for the data movement and connectivity needs that matter in AI systems.
The reported timeline points to possible mass production at TSMC in 2026. Before that, the project must pass through tape-out and an initial manufacturing run. Reuters notes that those steps carry technical risk, and fixes could push the effort back by months.
The project is expensive and technically demanding
Designing an AI processor is a large investment before a finished chip ever reaches production. Industry experts told Reuters that a single version of this kind of processor could cost as much as $500 million to design.
The total burden may be higher once the surrounding ecosystem is included. Supporting software and hardware could potentially double that amount, according to the same report.
The current OpenAI effort is reportedly led by Richard Ho, a former Google chip designer. Reuters says the project involves a team of 40 engineers working with Broadcom on the processor design.
That structure highlights the scale of the undertaking. OpenAI is not only working on a chip; it is also coordinating across design expertise, manufacturing, memory, networking, and future software support. Each layer affects whether the hardware becomes useful beyond a first experiment.
Why the timing matters
OpenAI’s chip work comes as major technology companies increase spending on AI infrastructure. Reuters notes that Microsoft plans to invest $80 billion in 2025, while Meta set aside $60 billion for the next year.
OpenAI has also been linked to a larger infrastructure push. Last month, OpenAI, working with SoftBank, Oracle, and MGX, announced the “Stargate” infrastructure project aimed at building new AI data centers in the US.
The chip project fits into that broader pattern. AI companies need more than models and applications; they need access to compute, data centers, manufacturing partners, and hardware roadmaps that can keep pace with demand.
Sam Altman’s chip ambitions have been visible for some time. In October 2023, Ars Technica covered a report about OpenAI’s intention to create its own AI accelerator chips. In early 2024, Altman also spent considerable time traveling around the world trying to raise up to a reported $7 trillion to increase world chip fabrication capacity.
A long game, not a quick exit from Nvidia
The first OpenAI chip, if it reaches production, is expected to have limited deployment. That means Nvidia hardware would likely remain important to OpenAI’s infrastructure in the near term.
Still, the direction is significant. A custom AI chip could become a negotiating tool, an internal platform, and the start of a hardware roadmap that OpenAI can refine through future iterations.
The biggest unknowns remain the same: what the chip can do, how well it performs, how quickly it can be manufactured, and whether the first tape-out works as planned. Until OpenAI formally announces the processor, the project remains defined by reporting rather than company confirmation.
But the shape of the strategy is now easier to see. OpenAI appears to be treating AI hardware as a core part of its future, not just a supply chain problem to manage.