Broadcom Seeks Up to $100B Debt Package for AI Infrastructure
Broadcom is reportedly in discussions with lenders to raise more than $60 billion in debt financing for a massive AI infrastructure program that could ultimately reach $100 billion.
The proposed financing would help AI developers—including Anthropic—secure custom AI accelerators and the supporting data center infrastructure required to deploy them at enormous scale.
Rather than requiring AI companies to fund the hardware entirely themselves, the structure would use private debt and special-purpose vehicles to finance the chips and lease them to customers.
If completed, the transaction would represent another major step in the transformation of AI infrastructure into a debt-financed asset class.
💰 A Potential $100 Billion Financing Structure #
The proposed package reportedly consists of multiple layers of debt.
The senior-secured portion could reach $60 billion to $70 billion, while another approximately $30 billion could come from junior debt.
Broadcom would reportedly backstop part of the senior financing, helping provide additional protection for lenders and potentially allowing the senior tranche to achieve an investment-grade credit profile.
The exact structure remains under negotiation and could change before closing. Financing may also be deployed in multiple stages rather than as a single $100 billion transaction.
The sheer scale demonstrates how much capital is now required to build competitive AI infrastructure.
🏦 Blackstone and Apollo Enter the Financing Picture #
Private-credit heavyweights Blackstone and Apollo Global Management are reportedly negotiating participation in the deal.
Their involvement builds on the AI XPV partnership established with Broadcom in June 2026.
The central idea is relatively straightforward: rather than having an AI company purchase enormous quantities of specialized hardware directly, outside investors finance the equipment through an SPV.
The SPV then leases the hardware to the AI customer.
This transforms expensive AI accelerators from an upfront capital expenditure into a long-term infrastructure obligation.
For investors, the arrangement provides exposure to AI infrastructure demand. For AI companies, it can provide access to compute without requiring the same level of immediate capital investment.
🔄 How the SPV Model Works #
The proposed arrangement can be simplified into four steps:
- Investors provide capital to a special-purpose vehicle.
- The SPV purchases custom AI chips and infrastructure supplied by Broadcom and its partners.
- The hardware is leased to an AI company, such as Anthropic.
- The AI company makes long-term payments under the lease agreement.
Broadcom’s partial guarantee provides additional protection to senior lenders.
This is similar to the financing structure reportedly used in the companies’ earlier $35 billion agreement announced in June.
The model effectively turns AI accelerators into infrastructure assets that can be financed in a manner similar to other large-scale physical infrastructure projects.
⚡ The Bigger Goal: More Than 20 GW of Compute #
The broader AI XPV initiative reportedly targets more than 20 gigawatts of computing capacity for leading AI laboratories.
That is an extraordinary amount of infrastructure.
At a conceptual level, 20 GW represents power consumption on a scale comparable to the output of roughly 20 large nuclear power plants.
And the electricity itself is only one part of the equation.
A deployment of this magnitude requires:
- AI accelerators
- Custom CPUs and networking
- Data center buildings
- Power generation and transmission
- Cooling infrastructure
- High-speed networking
- Storage systems
- Long-term operations and maintenance
The hardware financing therefore represents only one component of a much larger capital cycle surrounding AI.
🧠 Anthropic Gets Access to Custom Silicon #
For Anthropic, the proposed arrangement could provide a major strategic advantage.
Training and serving increasingly capable AI models requires enormous amounts of compute. Purchasing all of that hardware directly would require massive upfront capital expenditures.
Leasing custom accelerators through an SPV could allow Anthropic to secure long-term computing capacity while shifting some of the financing burden to infrastructure investors.
The approach could also provide greater hardware customization than simply purchasing general-purpose GPUs.
Broadcom specializes in custom silicon and has established relationships with some of the world’s largest technology companies.
For AI laboratories that want specialized accelerators optimized for their own workloads, that capability could become increasingly valuable.
🥊 Broadcom’s Bigger Challenge to NVIDIA #
The financing strategy also has significant implications for the competitive landscape.
NVIDIA currently dominates the AI accelerator market through a combination of GPUs, networking, software, and a massive developer ecosystem.
Broadcom is pursuing a different path.
Rather than attempting to replicate NVIDIA’s entire platform, Broadcom is positioning itself as a major supplier of custom AI silicon for hyperscalers and leading AI companies.
Long-term financing agreements could make that strategy considerably more powerful.
If Broadcom can convert future AI infrastructure demand into committed multi-billion-dollar chip orders, it gains greater visibility into revenue while customers gain access to specialized hardware without bearing all the upfront costs.
Broadcom CEO Hock Tan has previously indicated that the company’s annual AI semiconductor revenue could exceed $100 billion next year.
The company already has major custom-silicon relationships with technology giants including Apple and Meta.
📈 AI Infrastructure Is Becoming a Financial Market #
The most important implication may extend beyond Broadcom and Anthropic.
AI infrastructure is increasingly becoming a financing opportunity for banks, private-credit firms, asset managers, and institutional investors.
The logic is similar to infrastructure finance in other industries: enormous physical assets generate predictable long-term cash flows, allowing investors to finance them with debt rather than relying entirely on corporate equity.
AI adds a new variable—the extraordinary pace of technological change.
A data center can remain useful for decades, but an AI accelerator may become economically obsolete much faster.
That makes the financing structure particularly important.
Lenders need confidence that the underlying hardware will continue generating sufficient economic value throughout the repayment period.
🏛️ NVIDIA Is Pursuing a Similar Financial Strategy #
Broadcom’s financing push is occurring alongside a broader expansion of institutional capital into AI infrastructure.
NVIDIA has also been working with major financial institutions on a large-scale computing infrastructure platform targeting more than $500 billion in AI buildouts.
Participants include major names from the banking, private-equity, and asset-management industries.
The convergence is significant.
AI infrastructure is no longer being financed solely by technology companies. Increasingly, Wall Street and private credit are becoming part of the AI hardware supply chain itself.
That could dramatically accelerate infrastructure deployment—but it also creates new financial dependencies.
⚠️ The Risk Behind AI’s Debt-Fueled Expansion #
Guarantees from semiconductor companies can reduce initial credit risk for lenders, but they do not eliminate the fundamental risks of the AI business.
The ultimate question is whether AI companies can generate enough revenue from their models and services to justify the enormous computing commitments being made today.
If demand continues growing rapidly, the financing model could become extremely powerful.
But if AI monetization fails to keep pace with infrastructure spending, highly leveraged hardware deployments could create pressure across the entire ecosystem.
This makes the relationship between chip manufacturers, AI laboratories, infrastructure providers, and financial institutions increasingly interconnected.
🔭 Conclusion: AI Compute Is Becoming Infrastructure Finance #
Broadcom’s reported effort to raise as much as $100 billion illustrates how dramatically the economics of AI infrastructure are changing.
For Anthropic and other AI developers, the model offers a way to secure enormous amounts of computing capacity without funding every accelerator directly.
For Broadcom, it turns future custom-chip demand into potentially long-term, contract-backed business.
For private-credit investors, it creates exposure to one of the fastest-growing infrastructure markets in the world.
But the model also creates a new dependency: the financial system is increasingly betting that AI compute will remain economically valuable for years to come.
The next phase of the AI race may therefore be determined by more than who designs the fastest chip.
It may also depend on who can finance the infrastructure needed to deploy those chips at planetary scale.