Key Takeaways
Google and Broadcom are systematically copying the financial engineering playbook that built Nvidia's $3 trillion AI chip empire, deploying billions in loan guarantees and circular financing to win customers for their own processors.
The clearest example sits on the southern shore of Lake Ontario. Google provided a $3.2 billion financial guarantee for the Lake Mariner AI data center cluster in western New York, according to the Wall Street Journal. The site, developed by TeraWulf and Google-backed cloud provider FluidStack, will lease computing power from thousands of Google's tensor processing units to Anthropic. The guarantee helps the data center raise cheaper debt — the same tactic Nvidia has used for years to lock in chip orders.
"They want to be in the game, they don't want to be left behind," TeraWulf co-founder Nazar Khan said.
Google's financial commitments extend well beyond Lake Mariner. The company is also backstopping a $7 billion Anthropic project called River Bend near Baton Rouge, Louisiana, and has provided $1.4 billion in guarantees for AI computing infrastructure in Colorado City, Texas. Combined with a $5 billion partnership with Blackstone to launch a new cloud services company — one that will compete directly with Nvidia-backed providers CoreWeave and Nebius — Google is signaling it will use its balance sheet as a competitive weapon. The company said this month it plans to raise $85 billion in equity financing, primarily for AI infrastructure.
The Circular Financing Model Takes Hold
Broadcom is pursuing an identical strategy through a different structure. The chipmaker partnered with Apollo and Blackstone last week to launch the AI XPV platform, a $35 billion special purpose vehicle that buys Google TPUs and Broadcom networking chips and leases them to Anthropic. The debt is structured in three tranches: $6 billion in A1 notes sold to banks at Treasuries plus 100 basis points, $24 billion in A2 notes at a 5.75% yield sold to institutional investors, and $4.5 billion in junior notes yielding 8.5% that carry no Broadcom backstop. Apollo's Atlas SP Partners contributed $800 million in equity.
The key innovation is Broadcom's "deficiency guarantee" — if Anthropic defaults and chip sale proceeds can't cover the principal and interest, Broadcom makes investors whole on the senior tranches. This allows the platform to raise cheap debt, which in turn funds more chip purchases from Broadcom and Google. Broadcom CEO Hock Tan, who was cautious about using the company's balance sheet for such guarantees as recently as March, changed course after seeing Nvidia use similar supplier financing to accelerate sales. He called the deal "the first of many" and said the platform could finance over 20 gigawatts of AI computing capacity by 2028, with potential chip procurement reaching $700 billion.
Nvidia's Moat Faces Its First Real Test
Nvidia still controls more than 90% of the AI chip market, defended by its CUDA software ecosystem and plug-and-play hardware. Some smaller cloud providers say they fear losing their Nvidia allocation — a dynamic the industry calls "Jensen jail" — if they buy from competitors. Nvidia CEO Jensen Huang has publicly dismissed the threat, arguing in April that Anthropic is Google's only significant external TPU customer and challenging Google to prove its chips offer a cost advantage.
"Our market reach is far greater than any TPU or ASIC can possibly have," Huang said.
But Google is making progress. The company announced in May it would sell TPUs directly to enterprise customers for the first time and launched its inaugural inference-optimized chip, positioning it against Nvidia's Groq 3 LPU. Citadel Securities, an early adopter, reports running some workloads at 30% lower cost and up to four times faster on TPUs. Google Cloud AI infrastructure vice president Mark Lohmeyer said the inference chip has attracted customers who previously had not considered TPU technology.
Bernstein analyst Stacy Rasgon said Google is "more opportunistic and more aggressive" in commercializing its chip technology compared with previous years. The internal push comes from Amin Vahdat, promoted in December to oversee Google's AI infrastructure expansion, who reports directly to Google Cloud leader Thomas Kurian and CEO Sundar Pichai. Colleagues describe him as relentless in pursuing performance gains, pushing engineers for 10% incremental improvements in chip capabilities.
The broader context is a massive wave of AI infrastructure financing. Morgan Stanley estimates US AI capital markets financing will reach $400 billion, potentially surpassing $1 trillion by 2028, to match roughly $1.8 trillion in capital expenditure needs over the next two years. Meta completed a $27.3 billion SPV transaction around its Louisiana Hyperion data center, and Amazon raised about $10 billion in Canadian dollar bonds, the largest single issuance in that market's history.
For investors, the question is whether financial engineering can erode Nvidia's technological moat. Nvidia shares trade at roughly 35 times forward earnings, reflecting expectations of continued dominance. If Google and Broadcom succeed in commoditizing AI compute through cheaper financing and competitive silicon, that premium could face pressure. Vahdat, for his part, insists the market is big enough for everyone.
"There's so much demand out there," he said.
This article is for informational purposes only and does not constitute investment advice.
