If AI demand is so effortless, why is Broadcom reportedly exploring a financing package approaching $100 billion to turn chip demand into deployed capacity?
Chip orders and capacity commitments point to genuine demand. But equity returns depend on who supplies the capital, guarantees the debt, bears utilization and refinancing risk, and absorbs dilution.
Broadcom’s Bookings Prove Demand, Not the Quality of the Financing
Broadcom reportedly has more than $30 billion of Q2 AI bookings, compared with $10.8 billion shipped. Its FY26 AI revenue guidance exceeds $56 billion, with an FY27 target above $100 billion.
The opportunity appears real. Yet Broadcom is also reportedly exploring an AI financing package approaching $100 billion, potentially involving senior and junior debt, Apollo, Blackstone, and a special purpose vehicle.
Some reported structures would have financing vehicles buy Broadcom custom AI chips and lease them to Anthropic. Broadcom could guarantee part of the senior tranche.
Financing can turn customer commitments into installed capacity, allowing Broadcom to ship more silicon and customers to access more compute. But it changes the analysis. Investors must ask whether the capacity will be sufficiently utilized, whether the customer can support its lease commitments, whether the debt can be refinanced on acceptable terms, and whether guarantees move risk back toward Broadcom.
Strong demand and significant financing risk can coexist.
AI Suppliers Are Becoming Part of the Customer’s Capital Strategy
Google’s agreement with Marvell illustrates another shift. Google receives warrants to purchase almost 59 million Marvell shares at $206.58 per share. Another 240,000 warrants vest for every $500 million in qualifying revenue.
The agreement permits up to $120 billion in cumulative qualifying revenue through January 2033. That figure is a ceiling within the warrant structure, not a revenue forecast.
Google is doing more than placing orders. It is using equity-linked economics to support a broader supplier ecosystem.
Broadcom reportedly remains Google’s primary partner for core TPU silicon under an April 2026 long-term agreement. Marvell is being added around the TPU ecosystem, including AI inference accelerators, storage controllers, network interface controllers, memory interface controllers, and near-memory compute.
That suggests diversification, not necessarily replacement. Google gains another important supplier, while Marvell’s economics improve if qualifying revenue grows. Broadcom’s reported financing discussions go further: selling AI silicon may now include helping arrange the capital required to deploy it.
The Cost of Capital Is Becoming Part of Product Economics
UBS reportedly expects combined capital expenditures by Google, Amazon, Microsoft, and Meta to reach approximately $1.62 trillion by 2028.
Capital at that scale has a price. Big Tech borrowing costs were cited at 6% to 8%, while US 10-year and 30-year yields were cited at 4.7% and 5.2%.
A project can have strong demand and still produce disappointing equity returns if interest expense, refinancing risk, or an expensive capital structure consumes too much of its economics.
Nebius Shows Where Infrastructure Ambition Reaches Shareholders
Nebius raised approximately $4.94 billion through an upsized convertible notes offering, potentially increasing to roughly $5.68 billion if the 13-day greenshoe is exercised.
Separately, exchanges of $400 million of 2029 notes and $400 million of 2031 notes created immediate dilution through approximately 15.8 million Class A shares.
This does not prove that Nebius raised capital on unattractive terms. It does show that growth capital is not free. Shareholders must track both the assets being built and the claims created against them.
Broadcom’s bookings and revenue targets suggest substantial underlying business. Google’s Marvell agreement suggests customers want deeper, more diversified supplier capacity. Nebius shows that infrastructure providers can raise billions to pursue the opportunity.
But every layer of deployment introduces another potential bearer of risk: the chip company, financing vehicle, cloud provider, customer, creditor, or shareholder.
The next phase of AI investing is not merely about identifying demand. It is about tracing the liability attached to that demand and determining whether shareholders are being paid for it.