What it is
A special-purpose vehicle (SPV), sometimes called a special-purpose entity (SPE), is a separate legal entity — usually a corporation, trust, or limited-liability company — created by a parent organization for one narrow purpose. The SPV has its own assets, liabilities, and balance sheet, distinct from the company that set it up. It typically raises its own debt and equity, uses that money to buy or build something specific (a portfolio of loans, a factory, a fleet of chips), and then generates income by leasing or licensing that asset back to the parent or to a third party.
The defining feature is legal separation. An SPV is usually structured to be “bankruptcy remote,” meaning that if the parent company runs into financial trouble, the SPV’s assets and cash flows are generally protected from being pulled into the parent’s bankruptcy proceedings — and, just as importantly, the SPV’s own debt does not sit on the parent’s books. That second feature is what makes SPVs central to “off-balance-sheet” financing: a company can direct tens of billions of dollars toward an asset it will use and effectively control, without that debt showing up in its own leverage ratios or credit statistics. The structure is old and legitimate — securitization of mortgages, auto loans, and infrastructure projects has relied on SPVs for decades — but it earned lasting notoriety in 2001, when Enron used a web of off-book SPVs to hide losses and inflate reported profits, prompting the Financial Accounting Standards Board to tighten disclosure rules (FIN 46) for what regulators now call “variable interest entities.”
Why it matters for AI governance and narratives
SPVs have become the financing instrument of choice for the AI infrastructure buildout, and that carries real consequences for how the buildout is understood from the outside. When a hyperscaler or AI lab needs tens of billions of dollars in chips or data-center capacity, routing that spending through an SPV lets it acquire the capacity without the debt appearing on its own balance sheet — even though the company remains the sole customer and, often, the entity whose creditworthiness makes the debt attractive to lenders in the first place. That matters for the observatory’s opacity thread: leverage ratios, credit ratings, and headline debt figures are among the few outside signals investors, journalists, and regulators have for judging whether AI capital expenditure is rational or overextended. When tens of billions of dollars in AI-linked obligations sit one step removed in vehicles the sponsoring company does not consolidate, those signals become weaker proxies for the underlying risk — echoing, whether or not the analogy fully holds, pre-2008 concerns about structured-finance vehicles obscuring where risk actually resides.
Key facts and dates
The pattern reported around Broadcom and Anthropic in mid-2026 illustrates the mechanics. In June 2026, Broadcom, Apollo Global Management, and Blackstone assembled roughly $35 billion in debt financing through a partnership reported as the “AI XPV Platform,” structured around an SPV that would acquire Google-designed tensor processing units and lease them back to Anthropic — keeping the associated debt off Anthropic’s own balance sheet while Broadcom’s backing helped lower the financing’s cost. By early August 2026, Blackstone was reportedly pitching a second, roughly $36 billion debt package built on the same model, and Broadcom was separately in talks for a much larger facility — a senior tranche of $60–70 billion paired with roughly $30 billion in junior financing, again issued through a special-purpose vehicle, that could bring total financing tied to AI chip and infrastructure buildout toward $100 billion. The proceeds would fund custom AI chips and infrastructure for Anthropic and other customers, part of Broadcom’s push to compete with Nvidia in AI compute. These are not isolated deals: reporting has put total off-balance-sheet AI data-center and chip financing across major hyperscalers and AI labs above $120 billion by mid-2026, spanning similar structures involving Meta, Oracle, xAI, and CoreWeave with financing partners including Blackstone, Blue Owl, Apollo, and BlackRock.
As of this writing, the Broadcom-Anthropic terms remain under negotiation and reported, not confirmed by the parties; the deal could be finalized in stages rather than as a single transaction, and public figures have varied within the $60–100 billion range across successive reports.
Where to learn more
- CSC Global: Guide to Special Purpose Vehicles (SPVs) — general definition, legal structures, and bankruptcy-remoteness mechanics.
- Bloomberg: Broadcom Backing Lowers Debt Costs on $36 Billion Anthropic Deal — primary reporting on the earlier TPU-leaseback SPV deal.
- Bloomberg: Blackstone Has Pitched Mega Debt Package for Anthropic Chip Deal — reporting on the follow-on financing discussions referenced in the editorial.
- NBER: Special Purpose Vehicles and Securitization — academic treatment of SPV mechanics and their role in structured finance.