Broadcom arranges over $50 billion in financing for OpenAI to fund AI chip purchases and custom silicon development.
The breakneck pace of spending in the AI compute race is forcing chip suppliers to transform from mere sellers into financial engineers who structure capital on behalf of their clients. Broadcom is reportedly assembling a financing package exceeding $50 billion for OpenAI to support the latter's purchase of co-developed custom AI chips. Though the deal's final contours remain unfinialized, it already reveals an industry accelerating its tilt toward debt leverage.
According to multiple media reports, Broadcom has initiated preliminary discussions with major private credit firms including Apollo Global Management and Blackstone, with the financing size under discussion ranging between $30 billion and $50 billion. Sources indicated that no formal process has been launched, negotiations remain in early stages, and no deal is expected before the end of 2026. Representatives from Broadcom, OpenAI, and Blackstone all declined to comment.
At the core of this potential arrangement is the partnership Broadcom and OpenAI publicly announced in October 2025, in which they revealed plans to jointly develop custom AI accelerators—processors specifically designed for large-model training and inference. These chips stem from an internal OpenAI project codenamed Nexus, with the first two generations named Jalapeño and Serrano. Both parties aim to deploy AI accelerators totaling 10 gigawatts of power capacity by the end of 2029. Broadcom expects to deliver chips equivalent to 5 gigawatts of compute to OpenAI by 2028, with CEO Hock Tan stating he expects this business to become the company's second-largest custom chip customer. In June, Tan and OpenAI jointly unveiled the Jalapeño chip, with accelerators set to enter service later this year and broader deployment planned for 2027.
Broadcom's fundraising effort for OpenAI is not its first venture into this "chip delivery plus bundled financing" model. The company has launched a debt financing arrangement of up to $60 billion for Anthropic, another top-tier AI lab, currently being marketed. Banks participating have begun sending syndicated loan commitment letters for a $42 billion Class A senior secured tranche, while Blackstone is leading an $18 billion Class B junior debt portion and has already committed $9 billion from its various funds. Broadcom is providing a partial guarantee on the senior financing to attract lender participation.
Anthropic has simultaneously committed to a five-year TPU leasing contract totaling $125.2 billion, with Broadcom potentially providing up to $42 billion in convertible financing covering roughly one-third of that commitment. The arrangement would give Anthropic access to multiple gigawatts of next-generation TPU compute capacity starting in 2027. In September, Tan said Anthropic will become Broadcom's largest custom chip customer in 2027, with deliveries expected to reach 5 gigawatts next year and 10 gigawatts the year after.
In June, Broadcom formally announced a financing platform established jointly with Apollo and Blackstone to provide financing support for more than 20 gigawatts of compute capacity to AI developers including Anthropic and OpenAI by 2028. The platform launched with an initial transaction of $35 billion, supporting more than 1 gigawatt of Anthropic compute infrastructure. Fulfilling the platform's entire commitment could ultimately require hundreds of billions of dollars in capital.
Broadcom's negotiations with OpenAI are hardly isolated. The enormous capital expenditure required for AI infrastructure is pushing the entire technology industry to seek funding sources beyond traditional equity financing. Oracle is reportedly in separate talks with Apollo Global Management and Goldman Sachs to raise funds for its chip purchases, exploring a structure in which investors would fund a separate entity to buy chips, which would then be leased back to Oracle, allowing the cloud services provider to avoid additional borrowing to bridge the gap between equipment payments and cloud revenue collection.
Elon Musk's SpaceX has begun discussions with banks and investors to raise $40 billion for purchasing Nvidia AI chips, with roughly $10 billion in bank loans and $30 billion in investment-grade bonds. People familiar with the matter indicated the deal is not expected to close until 2027.
Earlier in August, Nvidia partnered with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs, and KKR to establish an independent financing platform aimed at progressively mobilizing more than $500 billion in third-party capital for AI infrastructure. Nvidia said the initiative is designed to transform AI computing and infrastructure into an investable asset class.
Broadcom's financing strategy may accelerate chip sales and lock in long-term orders, but it also exposes the company to the risk of its customers' ability to perform. Bank of America Securities analyst Tom Curcuruto estimates that if Broadcom's AI financing platform reaches its 20-gigawatt capacity target, its financing vehicles could carry approximately $370 billion in senior debt by mid-2029, including roughly $150 billion in new issuance during 2027. This $370 billion represents potential debt within the financing vehicles, not debt directly on Broadcom's balance sheet. However, Broadcom provides guarantees on some customer lease payments, with its maximum guarantee exposure on the first transaction at approximately $29 billion. As the financing platform continues to expand, these guarantee obligations could become increasingly burdensome.
Market concerns about such arrangements are already reflected in Broadcom's stock price. Following the news, Broadcom shares fell nearly 2% in premarket trading, reflecting investors weighing the financial risks associated with its AI financing commitments and the broader pressure on technology stocks from rising Treasury yields.
From a macroeconomic perspective, AI industry borrowing is expanding rapidly. Goldman Sachs estimated in early August that AI-related debt issuance had reached approximately $500 billion, while JPMorgan projects total issuance could hit $4.1 trillion by 2030. Morgan Stanley's estimates show AI infrastructure will require $1.5 trillion in external financing by 2028. The Bank for International Settlements has issued warnings on this front, noting that a downturn in AI investment could ripple through broader financial markets, with heavily indebted smaller companies bearing the initial brunt. The central question remains: can AI companies generate sufficient revenue to cover debt service while continuing to amortize the cost of expensive computing infrastructure?
While Broadcom aggressively uses financial instruments to bind large customers, rival Marvell Technology is sprinting full-speed on the same track. At its investor day, Marvell projected fiscal 2031 revenue of $70 billion to $90 billion, driven by demand for custom AI chips and high-speed data center connectivity, and forecast the custom silicon market to reach approximately $235 billion by 2030.
Custom chips offer enterprises an alternative to Nvidia's general-purpose AI processors, enabling optimization of compute performance and cost for specific workloads. OpenAI's decision to deeply bind itself to Broadcom for the Nexus chip series stems from a core desire to reduce dependence on a single supplier while optimizing hardware architecture for the specific characteristics of its large-model training and inference workloads. Yet this path is not without challenges. The cycle from chip design, tape-out, to validation and iteration is lengthy; the software stack development workload is enormous; and post-production requires sustained investment.