OpenAI Wants to End Reliance on NVIDIA Chips

Rumor has it OpenAI will finalize the proprietary chip it has been working on “in the next few months,” according to Reuters. OpenAI plans to send the chip to Taiwan Semiconductor Manufacturing Co (TSMC) for manufacturing once the design has been finalized. The first-time manufacturing process can take about six months.

OpenAI plans to put its proprietary chip into production by 2026.

Custom chips are strategic and technological moves

Running generative AI on custom chips could free OpenAI from its close ties with NVIDIA, which has grown so rapidly over the past few years partially on the success of OpenAI’s ChatGPT. Reuters said stakeholders inside OpenAI see the endeavor as “a strategic tool to strengthen OpenAI’s negotiating leverage with other chip suppliers.” OpenAI will continue to develop processors after the initial project to keep up with the advanced chip market.

NVIDIA holds roughly 80% of the market share for processors capable of running advanced generative artificial intelligence models. 

TSMC will use its advanced 3-nanometer process technology for OpenAI’s chip. It will use the same systolic array architecture with high-bandwidth memory as NVIDIA’s chips and have “extensive networking capabilities,” Reuters’ sources said.

A small team and a big investment

OpenAI is working with Broadcomm on the project; team lead Richard Ho doubled the size of the company’s in-house team in the last few months, Reuters said. But the 40 people still make up a relatively small team for the type of work.

According to sources contacted by Reuters, brand-new chip design could cost $500 million for just one iteration; software and peripherals could cost the same amount again. Some of OpenAI’s funding over the next few years is expected to come from the Stargate program, a U.S. government-lauded initiative started under the Biden administration in 2024 to build data centers for AI systems. Stargate could pour up to $500 billion into AI.

Other tech companies wanting to make in-house AI chips

Meta and Microsoft are both trying to make in-house AI chips, too. Meta has pledged $60 billion on all aspects of AI infrastructure in 2025, while Microsoft plans to devote $80 billion this year. Apple runs some of its mobile AI tools on name-brand M-series chips and outsources some to OpenAI.
The post OpenAI Wants to End Reliance on NVIDIA Chips appeared first on eWEEK.