Broadcom slides 4% as OpenAI custom chip financing stalls at $18bn

Broadcom shares slid about 4 per cent on Thursday afternoon after The Information reported that the chipmaker’s custom-silicon partnership with OpenAI had stalled over the financing of its first phase, a roughly $18bn outlay covering 1.3 gigawatts of data-centre capacity.

The chipmaker, run by chief executive Hock Tan, has told OpenAI it will only finance the opening tranche if Microsoft commits to buy about 40 per cent of the resulting chips, or OpenAI lines up other buyers, Yahoo Finance reported on the back of an internal memo and two people involved in the talks. Under the proposed structure, Microsoft would install the accelerators in its own data centres and lease them back to OpenAI, giving Broadcom a backstop for its upfront capital.

The dispute is the first publicly reported friction in a partnership unveiled in October 2025, when the two companies sketched a roughly $500bn multi-year hardware buildout spanning chips, networking, power and data-centre capacity. The full programme covers 10 gigawatts of OpenAI-designed accelerators, with volume deployment targeted for 2027. Broadcom has long required that OpenAI put up a dollar of its own for every dollar of vendor financing, but the internal memo cited by The Information indicates the company has recently signalled willingness to invest more capital upfront in exchange for the Microsoft commitment.

For Microsoft, the proposed lease structure is at odds with its preference for standardised, multi-tenant infrastructure. The company has built its hyperscale fleet around general-purpose Nvidia and AMD silicon and has historically resisted hard-wiring data halls to a single tenant’s chip design. OpenAI, by contrast, wants the facilities tuned around its own accelerators to extract better performance per watt as it scales inference for ChatGPT and its application programming interface, with operating expenses projected at about $200bn through 2029.

The custom-chip programme is central to OpenAI’s push to reduce its dependence on Nvidia and Sam Altman’s effort to bring more of the cost stack in-house. Broadcom has emerged as one of the largest beneficiaries of that diversification: Tan named OpenAI as the firm’s sixth major AI customer on the 4 March earnings call, and the company has since guided second-quarter AI semiconductor revenue to $10.7bn, up roughly 140 per cent year on year, with a 2027 target above $100bn from chips alone across six named accelerator customers.

Thursday’s pullback put AVGO at about $412.56, roughly 6 per cent below the 52-week high of $437.68 set earlier this year. The reaction tracks a broader unease in AI-exposed semiconductors over the financing assumptions underpinning the sector’s capital plans, after Micron’s 38 per cent weekly surge on AI memory demand and Qualcomm’s 8 per cent rally on its own OpenAI chip tie-up earlier in the week. The Broadcom programme is the single largest piece of OpenAI’s vendor diversification away from Nvidia, and a structural break would force the lab back onto the open market for general-purpose accelerators, according to The Information, which described the financing question as a stress test for the wider AI capital expenditure narrative.

Broadcom and OpenAI did not respond to requests for comment. Microsoft declined to comment to The Information. Investors will get the next official read on the partnership at Broadcom’s second-quarter earnings call on 3 June.