OpenAI taps Broadcom for rumored 2026 expansion into AI chip production

ChatGPT maker OpenAI is scheduled to start producing its first AI chip in 2026 in a partnership with semiconductor firm Broadcom, although it will not be sold externally.

The initiative is reportedly to keep up with the rising demand for computing power and market control, as opposed to mere expansion. Both OpenAI and Broadcom have not officially commented on the matter, although an article by Financial Times citing people familiar with the issue indicates the chip will be ready by next year.

The internal chip will reduce overreliance on Nvidia

For OpenAI, producing hardware is a leap, as training and running large language models eats through computing power and cash at an astonishing rate. Chipmaking giant Nvidia has until now largely held the majority of the market, with GPUs that power billions of queries.

However, relying on a single supplier is not wise as prices are high and supply can tighten without warning. Ultimately, the leverage lies with the chipmaker, not the user.

Sources familiar with the matter told the Financial Times that the chip will be used internally by OpenAI and will not be offered to external customers.

Last year, it was reported that OpenAI had begun scouting alternatives.

Broadcom CEO Hock Tan said on Thursday that the company expects artificial intelligence revenue growth for fiscal 2026 to “improve significantly”, after securing more than $10 billion in AI infrastructure orders from a newn unnamed customer. Its share price jumped 4% in response to the news.

A new prospect placed a firm order last quarter, making it into a qualified customer, Tan said on an earnings call.

Tan further suggested four other companies are in advanced discussions to design their own chips alongside Broadcom.

The aim is familiar across the sector, to reduce reliance on Nvidia, cut costs, and optimise for in-house workloads. But while the goal is simple, the path is not.

OpenAI joins growing industry trend

OpenAI joined the bandwagon a bit late and has lagged behind its industry peers like Google, which already has its Tensor Processing Units while Amazon has its Graviton and Trainium processors. Social media giant Meta has also pushed ahead with proprietary AI chips.

Designing, testing, and manufacturing silicon is expensive, technically demanding, and risky. Even the biggest tech firms have stumbled. For OpenAI, a software company at heart, the challenge is steep.

Sources say OpenAI will finalize its chip design soon and hand it off to TSMC for fabrication. If all goes according to plan, the chip could shift the company’s economics: lower running costs, faster experimentation cycles, and tighter control over infrastructure. Yet questions remain.

Will the chips remain internal forever, or could OpenAI one day join the ranks of Google and Amazon in selling AI-specific hardware? ChatGPT, DALL·E, and other internal systems are likely to be the first users.

Such a partnership exposes a broader truth that AI is not just about clever algorithms or datasets as the hardware, the engines behind the models, are equally crucial. Whoever controls it wields influence over the pace of innovation.

Analysts compare it to the early oil booms; chips are the fuel, and control over supply chains determines who wins. In that sense, OpenAI’s decision is as much about power and strategy as it is about technology.

The industry is evolving quickly. The ones who can design, build, and run their own silicon will have an edge, not just in cost but in speed, flexibility, and innovation. OpenAI’s Broadcom partnership may be messy, risky, and ambitious, but it could mark the beginning of a new phase: software companies taking the reins of the hardware that makes their AI dreams real.

If you're reading this, you’re already ahead. Stay there with our newsletter.