OpenAI has reportedly partnered with semiconductor giant Broadcom to develop its own custom chip, expected to launch in 2026.
According to the Financial Times, the chip is intended for OpenAI’s internal use rather than for external customers. It will power OpenAI’s own systems, including training large models such as GPT and managing day-to-day operations. This move would allow OpenAI to reduce its heavy reliance on Nvidia, as the company currently makes extensive use of Nvidia’s GPU chips to train and run its models. The custom chip aims to diversify the supply chain and reduce costs by decreasing dependence on the chip giant, the report said.
In February, further reports from Reuters confirmed that OpenAI was boosting its efforts to reduce its dependence on Nvidia by developing its own AI silicon. According to Reuters, OpenAI was finalising the design of its first in-house chip and preparing to send it to Taiwan Semiconductor Manufacturing Company (TSMC) for production, as the company sought to diversify its hardware supply chain and gain greater control over performance and costs.
The decision to develop custom AI chips puts OpenAI on par with tech giants such as Google, Apple, Amazon and Meta, which already design their own chips to optimise performance and control costs.
In June, Apple said it was exploring the use of genAI to speed up the design of custom chips that power its devices, according to the company’s top executive in the hardware technology sector. Johny Srouji, Apple’s senior vice president of hardware technologies, outlined the company’s interest in AI-assisted chip design during a speech in Belgium earlier this year, while he was receiving an award from Imec, an independent semiconductor research and development group that collaborates with major global chipmakers.
In February, Amazon Web Services (AWS) joined the list of tech firms that expanded their horizons in the field of quantum computing by unveiling a new quantum computing chip. The tech giant presented ‘Ocelot’, a prototype quantum computing chip designed to test the effectiveness of the company’s quantum error correction architecture. According to Oskar Painter, AWS director of quantum hardware, the new chip can accelerate the time it takes to build a practical quantum computer by up to five years.
These developments highlight the intensifying competition among tech giants to control the semiconductor technologies that underpin AI and next-generation computing. As demand for high-performance chips continues to surge, companies are investing heavily in custom solutions to secure their supply chains and maintain a competitive edge.
