Apollo and Blackstone are about to syndicate $36 billion of Google TPU – Good Machine

TL;DR

Apollo and Blackstone are shopping $36bn of investment-grade structured notes to private credit investors, financing the purchase of Google TPU chips that will then be leased to Anthropic. (PitchBook)
Pricing talk: A2 notes ~5.50-5.75% all-in yield; second-lien B notes 8.5% fixed coupon. Apollo and Blackstone plan to syndicate a portion and retain significant pieces themselves. (PitchBook)
The credit enhancement: Broadcom will backstop payments on the debt, which is housed in a special purpose vehicle. If Anthropic stops paying and the SPV cannot recover enough value selling the TPUs into the secondary market, Broadcom makes up the difference on the A1 and A2 notes. (PitchBook)
This is the lender-as-chip-lessor structure that emerged through 2025 (Meta/Hyperion, the CoreWeave financings, Oracle/OpenAI) reaching unambiguous scale. $36bn is roughly the entire 2024 European leveraged loan calendar in a single transaction.
Anthropic closed a $65bn equity round at a $965bn post-money valuation on 28 May, eclipsing OpenAI's $852bn. The TPU SPV is the debt side of that capital plan. (Bloomberg, TechCrunch)

$36,000,000,000

Start with the number. Thirty-six billion dollars is more than the GDP of Iceland. It is more than every single sovereign bond Australia issued in 2024. It is roughly four times the size of the largest leveraged buyout of the post-GFC decade. And it is the debt portion of a single AI compute deal, structured as investment-grade paper, sitting on top of an SPV that owns chips.

The chips are TPUs — Google's Tensor Processing Units, custom silicon designed for the kind of large-scale matrix multiplication that runs frontier transformer models. Broadcom is the manufacturing partner; Broadcom is also the credit backstop. Anthropic is the lessee. Apollo and Blackstone are the originators. Private credit is the bagholder of last resort.

That is the deal. Now look at what it actually is.

The framework: this is project finance, not corporate credit

Private credit, classically, lends to a company. The company has cash flows. The cash flows service the debt. If the company defaults, the lender takes the company.

This deal does not work that way. This deal lends to an SPV that owns chips and leases them to one customer. The cash flows are Anthropic's lease payments. The collateral is TPU silicon. The recovery path on default is "sell used AI accelerators into a market that may or may not exist."

In project finance terms, this is closer to a power plant lease than to a corporate loan. The asset is specialised, the offtaker is concentrated, and the recovery value of the underlying hardware in a distressed scenario is the entire underwriting question.

The pricing reflects this. A2 notes at 5.50-5.75% all-in is investment-grade pricing — roughly where you'd price a strong BBB. That is not a number you reach without the Broadcom wrap. The Broadcom backstop is doing nearly all of the rating work on the senior tranche.

The B notes — second lien, 8.5% fixed — are where the underwriting bites. That is high-yield pricing on paper that depends on TPU residual value in a stressed AI market. The bid for those bonds is the bid for a view about whether Anthropic survives, whether TPUs hold value, and whether Broadcom's balance sheet absorbs the difference.

What it actually means

One. Hyperscaler-adjacent compute financing has crossed a threshold. Through 2024 and 2025 we watched these structures emerge — Meta's Hyperion campus financing, the CoreWeave debt stack, the Oracle-OpenAI capacity arrangements — at $5-15bn at a time, mostly with a bank loan or term loan B at the centre. At $36bn into investment-grade-structured notes for private credit distribution, the pattern is now industrial-scale. Expect three more deals at this size or larger in the next 18 months.

Two. The credit enhancement is doing extraordinary work, and the market is paying attention. The Broadcom backstop is the structural innovation. Without it, this is a single-B credit at best. With it, Apollo and Blackstone can call it investment grade and place it through their wealth channels at IG spreads. The implicit calculation: Broadcom has every incentive to honour the backstop because (a) the chips it supplies depend on the financeability of this kind of deal, and (b) the reputational damage of stepping away from one of these would shut down the entire model. That is correct as a near-term incentive analysis. It is a strong assumption to bake into investment-grade pricing for seven-plus years.

Three. Private credit, having spent five years convincing itself it was a substitute for bank lending, is now substituting for an equipment finance and securitisation market that did not previously exist at this scale. That is a much harder business. Equipment finance specialists exist for a reason — residual values, secondary markets, repossession logistics, sectoral concentration. The Apollo/Blackstone underwriting teams are excellent. They are not, historically, asset-finance shops. The question is not whether they get the first deal right; they will. The question is whether the discipline holds through the third, fifth, and tenth deal — and whether the syndication channel keeps absorbing the paper at the same prices.

The quieter story: who actually holds the residual risk

Walk the cash flows. Anthropic pays the SPV for TPU lease. The SPV pays the noteholders. Broadcom backstops the A-tranche. The B-tranche is on its own.

In a base case — Anthropic continues paying, TPU demand holds — everyone wins. The IRR for Apollo and Blackstone on the retained portions is comfortable. The syndicated A2 buyers receive their 5.50-5.75% on what looks and feels like investment-grade paper. The B-note buyers earn 8.5%.

In a stress case — Anthropic stops paying — the SPV tries to sell or re-lease the TPUs. Who buys them? The plausible buyers are the other AI labs. If Anthropic is in trouble, the other AI labs are likely also under pressure. If they are not under pressure, they are running their own custom silicon (NVIDIA, AMD MI-series, their own designs) and have no urgent need for TPU inventory at face value. The residual market for at-scale TPU fleets in a distressed scenario is thin. That is the risk the B-notes are taking, and that is what 8.5% is paying for.

If the recovery is insufficient, the A1/A2 notes look to Broadcom. Broadcom's market cap is large; its incentive to honour is real; its ability to absorb a $20bn-plus draw on the backstop in a sector-wide AI capex retrenchment is a balance-sheet question, not a willingness question.

The honest read: this deal is investment grade if and only if you believe Broadcom is investment grade, and you believe Broadcom would actually step up under sector-wide stress. The first belief is well-founded. The second is reasonable, but it has never been tested at $20bn-plus.

Cross-layer scan

Connecting to Briefing #1. This paper will be syndicated, in part, into the same evergreen credit and BDC channels that just gated for liquidity. That is the same investor base, asking for substantially more compute-financing risk, in the same wrapper. The fee opportunity for the originators is enormous. The placement risk is real, this quarter more than last quarter.
AI capex sustainability. Anthropic raised $65bn in equity at a $965bn valuation on 28 May. Apollo and Blackstone are layering $36bn of debt on top, secured against the capacity Anthropic intends to use. This is a $100bn+ capital structure for a company that does not yet have $30bn of annualised revenue. The capital is real; the equity has cleared; the demand is real. Whether the unit economics support this is a separate question. The financing is being built on the assumption that they will.
Broadcom strategic positioning. Broadcom is no longer simply a chip supplier; with the backstop, it is the credit enhancement layer for the AI buildout. That is a deliberate strategic position and it is a powerful one — but it concentrates Broadcom's exposure to a sector it does not control.
Sovereign and regulatory. Expect questions from the Fed, the FSOC and European supervisors about the migration of this magnitude of compute-financing risk into pools backed by retail wealth-channel money. The "shadow banking" file just got a new chapter.

Hype deconstruction

What this is not:

A new tulip-mania moment. The credit work is real, the pricing reflects the risk in tranching, and the senior protection from Broadcom is structurally meaningful. This is not 2007 CDO mechanics with a label slapped on it.
A bet on AI in general. It is a specific bet on Anthropic-as-counterparty for seven years and on Broadcom-as-backstop for the same period. Anyone telling you it is broader AI exposure is selling you something.
Cheap money for Anthropic. The effective cost of this capital, blended across tranches, is meaningfully above what Anthropic could borrow at if it were a public investment-grade name. The deal exists because that public-IG channel is not available to a pre-IPO AI lab at this scale; private credit is filling a gap, at a price.

Recommendations

For institutional credit allocators considering the syndicated tranches:

Do not underwrite the A2 notes off the headline rating. Underwrite the Broadcom backstop. Pull Broadcom's credit, model the backstop draw in a 30% Anthropic revenue impairment scenario, and price the AVCO. If you do not get to investment grade after that analysis, the spread is wrong for you.
The B-notes at 8.5% are a view trade on TPU residual value. If your shop does not have a hardware-residuals team, do not buy them.
Watch the retention. Apollo and Blackstone keeping significant portions is a positive signal; if that retention size shrinks in follow-on deals, that is a tell.

For wealth and private-bank allocators with HNW clients seeing this paper through evergreen wrappers:

This is not a substitute for plain-vanilla private credit exposure. Bucket it as compute-finance and cap exposure accordingly. Recommended ceiling: 10-15% of the alts sleeve in any single client's allocation.
Confirm the wrapper's concentration limits. A 5% cap on redemptions is the wrong feature if 40% of the underlying paper is one structure, one counterparty, one chip class.
The Broadcom backstop is the rating; communicate that to the client in plain language.

For corporate treasury teams at AI labs, hyperscalers and adjacent infrastructure providers:

This deal sets a pricing reference for compute-collateralised debt. A2 at 5.50-5.75% with a strong corporate backstop is now the market. Negotiate accordingly on your next round.
The Broadcom backstop pattern is replicable. If you are a chip supplier with strong credit, this is a new business line.

For regulators (Fed, OCC, FSOC, ECB, APRA, BoE):

The migration of compute-financing risk into semi-liquid retail wrappers is happening fast. The supervisory frame needs to update from "is private credit a banking-system risk?" to "is private credit a hardware-residuals risk wearing a retail-yield product wrapper?" The answer to the second question has different mitigants.

Uncertainty ledger

Final pricing on both tranches is still being finalised at time of writing.
The split between syndicated and retained portions has not been publicly disclosed.
The exact structure of Broadcom's backstop — first-loss size, cap, trigger mechanics — has not been published in detail.
Anthropic's actual usage commitments under the lease (take-or-pay, partial offload, sub-lease rights) are not publicly known.
The deal is being marketed; it is not yet closed. Material changes are possible.

Bottom Line

Apollo and Blackstone have figured out how to package $36bn of depreciating AI hardware as investment-grade paper. The mechanism is sound. The Broadcom backstop is real. The buyers are eager. None of that makes this corporate credit, and none of it makes the next ten deals as well-constructed as the first. The most important sentence in private credit this year is no longer about loans. It is about chips, and who eats the residual when the lease ends and the music stops.

Sources

PitchBook, Pricing details emerge on $36B private credit deal for Anthropic (2 Jun 2026) — Tier 2
Bloomberg, Anthropic Raises at $965 Billion Valuation, Eclipsing OpenAI (28 May 2026) — Tier 1
TechCrunch, Anthropic raises $65 Billion, nears $1T valuation ahead of IPO (28 May 2026) — Tier 2
Private Equity Wire, Apollo and Blackstone plan $36bn private credit package to fund Anthropic's Google TPU deal (week of 26 May 2026) — Tier 2