Report: OpenAI faces $18B financing shortfall for custom‑AI chips

OpenAI and Broadcom are trying to finance a custom-chip buildout that outside reports say has run into an $18 billion funding gap. The reported shortfall has drawn attention because the two companies disclosed only seven months ago that they planned to deploy 10 gigawatts of OpenAI-designed AI accelerators, with rack deployments targeted to begin in the second half of 2026. OpenAI has not publicly detailed the project’s financing structure. Broadcom, for its part, has told investors that demand for custom AI accelerators and networking products is driving a sharp rise in revenue. ### Where does the $18 billion figure come from? The Information reported on May 8 that OpenAI’s custom-chip deal with Broadcom had hit an $18 billion financing snag, tied to the first phase of the project, according to summaries carried by other outlets. OpenAI and Broadcom announced the partnership on October 13, 2025, saying they would collaborate on 10 gigawatts of custom AI accelerators. The companies said OpenAI would design the accelerators and systems, while Broadcom would develop and deploy them in partnership with OpenAI. They said rack deployments were targeted to start in the second half of 2026 and be completed by the end of 2029. (theinformation.com) ### What have OpenAI and Broadcom actually said in public? OpenAI’s October 13 statement said the companies had “long-standing agreements on the co-development and supply” of the accelerators and had signed a term sheet to deploy racks incorporating the chips and Broadcom networking gear. The statement did not disclose financial terms, committed purchase volumes, or outside financing arrangements. (openai.com) Sam Altman, OpenAI’s chief executive, said in that announcement that partnering with Broadcom was “a critical step” in building AI infrastructure. Hock Tan, Broadcom’s chief executive, said the collaboration would co-develop and deploy “10 gigawatts of next generation accelerators and network systems.” ### Why does this matter for Broadcom now? (openai.com) Broadcom said on March 4 that first-quarter fiscal 2026 AI revenue rose 106% year over year to $8.4 billion. Hock Tan said the gain was driven by “robust demand for custom AI accelerators and AI networking,” and the company forecast AI semiconductor revenue of $10.7 billion for the second quarter. (openai.com) Palo Alto-based Broadcom also reported total first-quarter revenue of $19.31 billion, up 29% from a year earlier, and guided for about $22.0 billion in second-quarter revenue. Those figures have made the company a central supplier to large AI infrastructure programs, especially for customers pursuing custom silicon alongside standard GPU deployments. (investors.broadcom.com) ### Has Broadcom identified OpenAI as a customer? Broadcom has publicly tied its AI growth to custom accelerator programs, and several market reports have said OpenAI is among its named custom-silicon customers. Those reports cite Broadcom’s growing roster of hyperscale customers as support for its long-term AI revenue targets. OpenAI’s own October announcement confirms the companies are working together on OpenAI-designed accelerators and Broadcom networking systems. (investors.broadcom.com) That public statement establishes the commercial relationship even though Broadcom’s March earnings release did not list customer names in the text of the release. ### What is the financing problem likely to affect first? (tech-insider.org) The October 2025 announcement set the first visible milestone: deployments starting in the second half of 2026. If financing is delayed, that timetable is the most concrete public benchmark against which investors and suppliers can measure slippage, though neither company has announced a revised schedule. (openai.com) The project’s scale also matters. OpenAI and Broadcom said the program would reach 10 gigawatts by the end of 2029, a size that implies not only chip production but also associated networking, power and data-center capacity. The companies did not publish a full project budget when they announced the deal. ### What should readers watch next? (openai.com) Broadcom’s next scheduled checkpoint is its fiscal second-quarter earnings report in June 2026, when investors will look for updated AI semiconductor revenue guidance and any comments from Hock Tan on custom-accelerator demand. The second marker sits in the calendar already disclosed by OpenAI and Broadcom: rack deployments were targeted to begin in the second half of 2026. (openai.com) Any confirmation, delay or financing update from OpenAI, Broadcom or named infrastructure partners would show whether that milestone still stands. (investors.broadcom.com)