Anthropic’s proposed $42 billion financing arrangement with Broadcom shows how the next phase of artificial-intelligence expansion is being built through tightly linked relationships between chip suppliers, equipment lessors, financiers and cloud infrastructure providers. The arrangement, disclosed in Anthropic’s IPO prospectus, is not simply a procurement contract: Broadcom may supply the hardware, help finance it and become the AI company’s largest compute customer as early as 2027.
That structure offers Anthropic a way to secure the computing capacity required to train and operate increasingly large models. It also creates dependencies that the company itself has warned could affect its ability to obtain computing infrastructure on suitable terms. The filing identifies potential conflicts of interest arising from Broadcom’s simultaneous roles as hardware supplier and financing partner.
The arrangement is part of a much larger infrastructure commitment. Anthropic has committed $125.2 billion over five years to lease tensor processing unit, or TPU, computing capacity. The convertible note that Anthropic could issue under the Broadcom financing may fund about one-third of that obligation. Anthropic said it does not expect any notes to be sold before completing its initial public offering.
The prospectus also states that Broadcom could designate a financing partner and that the debt instruments could be converted into Anthropic shares. In April 2026, Anthropic deposited cash into a restricted account for Broadcom’s benefit and may have to contribute more in certain circumstances. The filing warns that payment or performance defaults could make a substantial portion of the lease obligations immediately due while restricting Anthropic’s ability to use the $42 billion facility to meet those payments.
These terms illustrate how AI infrastructure is moving beyond the conventional model of a company buying computing capacity from an independent provider. Anthropic’s relationship with Broadcom spans compute supply, equipment leasing and financing. That gives the semiconductor company a central position in the expansion of Anthropic’s physical and digital infrastructure, while also tying Broadcom’s future chip revenue to the AI lab’s ability to raise capital and generate income.
The arrangement differs from the role played by some of Anthropic’s other major partners. Amazon, for example, primarily provides cloud infrastructure and distribution for Anthropic’s Claude model, according to the prospectus described in the report. Google and Broadcom have collaborated on several generations of TPUs, the specialised processors that Anthropic plans to use at scale.
Anthropic announced in April an expanded partnership with Broadcom and Google that would provide access to multiple gigawatts of next-generation TPU compute capacity beginning in 2027. The reference to gigawatts is significant because it describes infrastructure at a scale closer to a utility or industrial system than to a conventional corporate technology purchase. Large AI deployments require specialised processors, high-capacity power systems, cooling, buildings, network equipment and long-term capital commitments. The filing provides a view of how those requirements are being assembled through commercial contracts.
The financial arrangement also reflects the growing importance of semiconductor companies in shaping the supply of AI infrastructure. Broadcom is expected to become Anthropic’s largest compute customer in 2027. At the same time, Broadcom projects AI semiconductor revenue of about $115 billion in fiscal 2027 and $230 billion in fiscal 2028. Anthropic’s infrastructure plans therefore form part of a wider strategy in which chip companies are not only selling components but helping customers finance the capacity needed to buy them.
That strategy resembles the approach Nvidia has taken in recent years, according to comments cited in the report. Seaport Research analyst Jay Goldberg said Nvidia was committing a large amount of its balance sheet and that Broadcom was having to follow suit. The use of a semiconductor company’s financial strength to support chip demand creates a feedback loop: financing helps an AI company secure more compute, while the resulting infrastructure commitments support projected chip revenue.
The model can accelerate construction and deployment because it aligns supply, financing and demand. An AI company does not have to arrange every part of the infrastructure independently, and a chip supplier gains greater visibility into future orders. For cities and infrastructure providers, however, such arrangements can concentrate responsibility and risk. A single commercial relationship may influence equipment specifications, financing conditions, leasing obligations and access to computing capacity.
Anthropic’s own disclosure points to the central governance problem. The company said Broadcom’s role in supplying hardware and acting as a financing partner creates potential conflicts of interest that could affect Anthropic’s access to the computing power needed for its work. It also warned that Broadcom’s decisions on pricing and hardware could affect its ability to procure enough computing infrastructure.
This is not only a corporate finance issue. The physical expansion of AI depends on data-centre campuses, power connections, advanced cooling systems and networks capable of moving large volumes of data. When infrastructure is secured through long-term leases and convertible debt, the financial structure can influence how quickly facilities are built, which technologies are installed and how much capacity remains available to other users.
The filing does not establish that the arrangement will fail or that Broadcom will misuse its position. It does establish that Anthropic sees the relationship as carrying risks requiring disclosure. Those risks include the possibility that lease obligations become immediately due after certain defaults and that the financing facility cannot be used freely to cover them. The conditions could become particularly important if the AI company’s revenue growth does not keep pace with its infrastructure commitments.
Robert Leitao, managing partner of Rothschild & Co, described the situation as a concentrated bet on two companies generating enough revenue to support the financing already committed. His observation captures the wider financial question embedded in the infrastructure buildout: whether expected demand for AI services will be sufficient to support the cost of the computing systems being ordered today.
Anthropic is preparing for a public offering that could value it at $2 trillion, according to the report. That prospective valuation and the scale of its infrastructure commitments are connected. Access to large amounts of compute is necessary for the company’s expansion plans, but securing that access creates substantial fixed obligations. The more infrastructure is committed in advance, the greater the pressure to keep that infrastructure utilised and commercially productive.
For Broadcom, the arrangement provides a route to support projected AI semiconductor growth. For Anthropic, it provides access to capacity and financing. For the wider AI infrastructure market, it demonstrates how supply chains are becoming vertically connected through contracts that combine chips, leases, debt and equity-like instruments.
The immediate evidence does not show how the arrangement will perform after Anthropic’s IPO, whether the notes will be issued, or how the parties will respond if performance or payment conditions change. It does show that the physical expansion of AI is being financed through relationships that can blur the lines between supplier, landlord, lender and strategic partner. As next-generation TPU capacity comes online from 2027, the key issue will be whether this integrated model delivers reliable infrastructure without making access, pricing and financial stability dependent on a small number of companies.