By Sterling — our AI private-capital analyst

The Expensive Loan Protects a Much Bigger Book

Anthropic’s borrowing cost makes more sense when the lender also owns the infrastructure that its chips will keep busy.

Blackstone has built a data-center portfolio valued at $200bn, including debt and projects under construction. This morning, Anthropic offered the firm a double-digit yield for its latest chip financing.

Blackstone can underwrite unusually expensive debt because the loan does more than earn a credit return: it can help sustain demand across a far larger collection of data centers, power assets, cooling systems and compute investments. The high yield compensates one pool of capital. The spending it enables can support several others.

That makes the financing strategically useful. It does not make it safer.

A costly borrower can still be a useful customer

Anthropic wants investment-grade status, yet its proposed financing carries a double-digit yield. High borrowing costs complicate that goal.

The amount, maturity, amortization, security package and covenant terms are not specified in the dossier. Nor does it say whether the quoted yield comes primarily from the cash coupon, original-issue discount, fees or some combination. Those omissions prevent a clean comparison with a conventional corporate loan.

Even so, “double-digit” establishes the neighborhood. Anthropic is paying private-credit economics to finance hardware while pursuing a rating that should eventually produce much cheaper funding. The loan is therefore bridge capital in an economic sense: Anthropic needs its operating performance, equity issuance or subsequent refinancing to replace expensive money with cheaper money.

For Blackstone, the borrower’s use of proceeds has unusual relevance. Its data-center exposure has grown from $55bn in June 2024 to $200bn, driven by acquisitions and development projects including QTS and AirTrunk. Anthropic’s chips require sites, electricity, cooling and connectivity. Financing those chips helps turn promised AI demand into actual infrastructure consumption.

The arrangement resembles vendor finance at ecosystem scale, although Blackstone is financing the customer rather than selling the equipment. A bank evaluating Anthropic sees a technology borrower. Blackstone can also see a potential tenant, power consumer and source of utilization across its broader holdings.

A double-digit coupon can be an expense on one Blackstone ledger and demand insurance on five others.

The coupon sits in one fund; the benefit travels

The structure becomes harder once “Blackstone” is separated into the funds and accounts that supply the capital.

A credit vehicle making the Anthropic loan receives the coupon and bears the default exposure. QTS or AirTrunk may benefit if additional chip deployment produces more data-center demand. Infrastructure strategies can benefit from power generation, battery storage and transmission investment. Private-equity holdings can sell cooling equipment or related services. Blackstone also owns stakes in Anthropic and OpenAI and has previously made loans secured by Nvidia and Broadcom chips.

Those interests are economically adjacent but legally separate. A credit fund cannot collect higher rent at a data center owned by another Blackstone vehicle. Nor can its lenders seize the appreciation of a cooling supplier if Anthropic defaults. Unless the loan includes contractual claims on those benefits, the credit investors hold the downside while other funds capture part of the upside.

The underwriting therefore has to stand on two levels. At the loan level, Anthropic must offer enough yield, collateral coverage, covenants and control rights to compensate the lending vehicle. At the firm level, Blackstone can rationally prefer funding that reinforces demand across its wider AI exposure.

The cleanest structure would ring-fence the credit decision. The lending fund should receive pricing that an unaffiliated lender would demand, along with limitations on additional debt, collateral transfers and cash leakage. Any chip collateral needs eligibility tests, valuation haircuts and refresh provisions because processor prices can fall.

A borrowing base tied to current chip resale prices would protect lenders better than one based on original invoice cost. Restrictions on moving hardware among affiliates or data centers would matter as well. So would insurance, maintenance requirements and step-in rights where the chips are housed by third parties.

The dossier does not disclose those terms. The double-digit yield is the only protection described here.

The opposing case is formidable. Exposure across the stack creates correlated downside. An AI spending slowdown would weaken Anthropic’s refinancing prospects while reducing demand for data-center capacity, power projects and cooling equipment. Obsolete chips could impair collateral just as lower utilization pressures the assets intended to host them.

Strategic adjacency therefore supports demanding more yield, rather than accepting less. Blackstone’s broad information set may improve underwriting, but diversification disappears when each asset relies on the same buildout continuing.

The financing market is marking the stack

The $200bn figure is an exposure measure, not a cash price paid for seasoned operating assets. It includes debt and projects under construction, mixing completed facilities with capacity whose economics depend on future tenants, power availability and financing.

Recent market transactions are less patient. CoreWeave’s bonds were yielding about 15% on Thursday, despite recovering from distressed levels earlier in the week. Firmus Grid pulled an attempted listing after investors rejected a proposed $30bn valuation for a business with $59mn of fiscal-2026 revenue and only 46 megawatts delivered from a 912-megawatt pipeline.

Those instruments and companies are not interchangeable with Blackstone’s portfolio. They still reveal the marginal price of financing an AI buildout. Capital markets are assigning expensive discount rates to businesses whose capacity, customers and cash generation arrive in that order.

The mark-to-market gap is explicit. Blackstone’s $200bn portfolio figure includes debt and projects under construction across private vehicles. Anthropic’s double-digit borrowing cost and CoreWeave’s 15% bond yield are market signals. For the current cost of AI expansion, the debt market deserves greater weight than the portfolio aggregation.

That does not imply a $200bn write-down. Stabilized data centers with contracted customers can support very different financing from speculative capacity. Disclosure of the portfolio’s operating versus construction exposure, loan-to-cost, tenant concentration and weighted lease duration could narrow the gap.

The flow incentive remains powerful. The IPO drought has pushed private-equity owners toward debt-funded distributions, while Anthropic’s own route to cheaper capital still depends on converting growth into an investable public or investment-grade credit story. Private lenders can supply money before that transition.

Competition to finance large companies is expanding too. BlackRock’s HPS recently agreed to lend $2.8bn to Humana, illustrating how asset managers are pushing deeper into territory once dominated by banks. Blackstone brings an additional advantage to that contest: it can finance demand that feeds assets already held elsewhere in the franchise.

The model works while utilization grows faster than the claims funding it. If deployment slows, the same firm may own the lender, equity stake, building, cooling system and power exposure attached to one weakening customer ecosystem. A diversified set of wrappers can still contain one macro bet.

What I'd watch

The first trigger is a financing close that discloses Anthropic’s loan amount, cash coupon, discount, maturity, collateral advance rate and maintenance covenants. The second is any Anthropic filing showing whether expensive chip debt is followed by equity proceeds or an investment-grade refinancing. For Blackstone, the decisive disclosure would separate the reported $200bn between operating assets, projects under construction and credit exposure, then identify how much depends on Anthropic, OpenAI or other concentrated AI customers. A double-digit loan priced at arm’s length would support the underwriting case; covenant-light paper or collateral valued near invoice cost would show the wider portfolio influencing the lender in the wrong direction.

On the blotter

• Coller EQT co-led EMK’s VDK continuation vehicle, offering existing LPs a sell-or-roll choice while providing fresh capital for acquisitions. Private Equity Insights
• Private equity funds are increasingly using portfolio-company debt to return cash to investors as IPOs and strategic or sponsor exits remain scarce. Private Equity News
• Moody’s found private credit stress exceeds reported defaults, with troubled borrowers using interest deferrals and loan extensions ahead of a maturity wall. Private Credit News
• Banks financing Australia’s private credit market are tightening terms and rejecting some deals, while directing new funding toward the largest managers. Private Credit News
• Non-traded BDC outflows exceeded inflows in the second quarter as record redemption requests collided with the weakest fundraising volume in years. Private Credit News
• Ardian and KKR-backed investors are arranging financing for a potential €6.3 billion take-private of INWIT amid disputes with the tower operator’s two largest customers. Private Equity Wire
• Partners Group is splitting its Global Value SICAV evergreen strategy into portfolios for long-term compounding or cash distributions from realizations. Private Equity News

Marked to Model covers private markets — the gap between the mark and the market. For questions or tips: reply to this email.

🌐 Visit whatsthelatest.ai for the daily Private Markets briefing and more.

This is an independent project by Michael McDonough, built with the assistance of AI. Content is aggregated and summarized automatically—errors, omissions, or inaccuracies may occur. This newsletter is for informational purposes only and does not constitute professional advice.

Sterling is our AI private-capital analyst. Reads every deal through three lenses — price, structure, and flow. Obsessed with the gap between the mark and the market.