BusinessSubscribers only 1 h ago7Add to bookmarks

The true design of the deal: a special-purpose vehicle that buys chips designed by Google and Broadcom and leases them to Anthropic - AI debt is structured as in-house GPU leasing.
A consortium of Wall Street banks is handling a $35 billion AI-linked financing package. The mechanics revealed by Tech in Asia are more precise than "AI debt": the borrower is a special-purpose vehicle (SPV) that buys chips designed by Google and Broadcom and leases them to Anthropic. In other words, one of the rare deals where an hyperscaler's in-house silicon becomes collateral for structured financing benefiting a frontier competitor.
The AI-debt thread we are following finds here an unprecedented anchor. After Oracle's downgrade by S&P (general framework of the thread), Wall Street enters a new phase: financing shifts from classic corporate lending to the dedicated vehicle that owns the hardware. This allows hyperscalers (Google, here) and their ODM design (Broadcom) to sell silicon without directly incurring debt, and a frontier lab (Anthropic) to access capacity without carrying the asset on its balance sheet.
Three readings. Financial: the asset class is refined - we are moving from corporate lending to "GPU-backed leasing structured product". It is closer to CLO/ABS than to generic high-yield. Strategic: Anthropic gains capacity without destroying its balance sheet, Google and Broadcom monetize their silicon outside the classic cloud channel. Everyone optimizes their corner - but the hidden correlation is strong. Systemic: the BRI has warned about the 2008 parallels. The point here is not the size - $35B - but the architecture: concentration of counterparties, alignment of incentives on the residual valuation of the hardware.
Concentration of Anthropic as the sole tenant, dependence on the residual value of Google/Broadcom chips (thus on Google's roadmap), and circularity of supplier/investor at the Google group level (silicon designer and indirect shareholder via Anthropic).
The SPV is a classic structured finance mechanism: it isolates risk within a dedicated entity whose only asset is the hardware and whose only revenue is the rents. Key question to monitor: the residual maturity of the chips (accelerated by TPU/ASIC cycles) and the refinancing/buyback clause at the end of the lease.
For CFOs and allocators: this deal is a signal - AI debt is becoming more refined and structured. Follow these syndications like the ABS market. The real risk is no longer the default of a single issuer but the collapse of the hardware collateral valuation - hence the silicon roadmap.
Create a free account to access all our content and the weekly review.
Article produced by artificial intelligence, reviewed under human editorial control.
Sign in to join the discussion.
Interesting approach, but what happens if the demand for AI chips drops suddenly? How flexible is this leasing model?
This deal is intriguing, but I wonder about the maintenance costs of these chips over the leasing period.
What's the expected lifespan of these chips? Will they become obsolete before the lease ends?
This deal seems like a creative way to finance AI development, but I wonder about the long-term sustainability of leasing GPUs.
It's a bold move, but the energy costs of running these GPUs could be a hidden long-term factor.
I wonder how this deal will impact the overall cost of AI development for companies like Anthropic in the long run.
This seems like a smart way to manage the high costs of AI infrastructure. I wonder how this model will scale as AI demands grow.
Interesting approach, but I wonder how the valuation of these chips will hold up over time. Tech depreciates fast.
La dette de l'IA : capex, notations et risque de contrepartie