According to the Financial Times, the ChatGPT creator has teamed up with Broadcom to co-design a custom AI processor that insiders say will start shipping in 2026. The project has been underway since last year, but the timeline for mass production was previously murky.

Broadcom chief executive Hock Tan teased investors by talking up a mystery customer coughing up $10 billion (€9.2 billion) in orders. He told analysts the deal brought “immediate and fairly substantial demand”, with chips heading out “pretty strongly” from next year.

Sources later confirmed that OpenAI was the secret client, although both companies are keeping their lips sealed. Broadcom has a track record of doing bespoke jobs, having previously cooked up Google’s tensor processing units.

The cocaine nose jobs of Wall Street lapped it up, with Broadcom shares climbing 4.5 per cent in after-hours trading after the earnings call. Its stock has already jumped more than 30 per cent this year thanks to the AI hype machine.

The new parts have been dubbed “XPUs” to make them sound more exciting than the GPUs that Nvidia and AMD still flog. HSBC analysts claim Broadcom’s custom silicon arm could grow faster than Nvidia’s graphics racket by 2026.

OpenAI boss Sam Altman has been shouting for months about the desperate need for more compute power. He admitted last month that the outfit was doubling its server fleet within five months to keep up with demand from GPT-5.

Job’s Mob, Google, Amazon and Meta have already jumped into designing their own AI chips, so Altman’s gang is only catching up with the big boys. But given the speed at which OpenAI chews through GPUs, Nvidia will still be minting money from them for a while.