On Friday October 30, 2026, the Financial Times reported that Blackstone had told the investors in its private-credit funds to expect Anthropic to receive a high junk rating after its IPO — a double-B range stamp, the top tier of the speculative grade — rather than the investment-grade rating Blackstone had itself been projecting for the company before 2027. TechCrunch's Lucas Ropek summarised the FT piece the same afternoon at 14:52 PDT, and by Friday evening the forecast revision was a visible datapoint in every write-up of a bad week for AI-adjacent capex.

Four days after Anthropic's public S-1 landed on EDGAR, and roughly ten days before Goldman and Morgan Stanley are expected to open the road show, this is a specific kind of pre-IPO signal. The company that is leading the subordinated tranche of Anthropic's biggest debt deal has reset expectations about what the senior credit market will do with the paper. Not a downgrade — Anthropic has never had a Big Three rating to downgrade — but a walk-back on the rating Blackstone's own clients were being told to assume.

The forecast Blackstone had to walk back

The June financing was called Project Big Sky inside the deal rooms. Apollo led a $35 billion private-credit package to buy Anthropic its chips, structured around Broadcom's custom silicon and nearly 1.5 GW of Google TPU capacity. Blackstone came in on the subordinated side, alongside banking partners whose names the trade press has mostly left anonymous. The senior tranche carried a Broadcom guarantee at 5.75 per cent. A $4.4 billion junior tranche, with no guarantee and direct exposure to Anthropic's credit risk, sold at roughly 8.5 per cent. The spread — 275 basis points — is what the market was paying to hold Anthropic risk without a Broadcom back-stop in June.

Blackstone's pitch to its own investors in that structure, per the FT, was that Anthropic would be investment grade before 2027. The pitch is the kind of thing you tell a limited partner so the fund looks temporary speculative and durable prime. In October — four months later, after a Q3 run-rate of $16.8 billion, a $63 billion follow-on debt package in the works for Google TPU financing, and a public S-1 that disclosed $80.7 billion of net loss and no debt on the balance sheet — the pitch has been revised to "double-B range, top tier of the speculative junk grade." Still speculative. Investment grade in the Blackstone deck has moved from a 2027 bet to an open question.

For a trillion-dollar-neighbourhood equity story, that is a strange sentence.

What a BB+ actually says about what KBRA was given

The forecast revision has an older sibling. In June — the same month the Big Sky tranches priced — Kroll Bond Rating Agency quietly assigned Anthropic a BB+ issuer rating. BB+ is the highest tier of the speculative grade, one notch below investment grade, and the stamp did not become public until mid-October when the FT read it off a document and the trade press (CoinGape, The Deep Dive, DealRoom) carried it in the same 24-hour window. The same BB+ with a stable outlook landed on the $4.4 billion junior tranche of Big Sky.

KBRA is not Moody's, S&P or Fitch — the three agencies the market reads for post-IPO investment-grade thresholds. KBRA is smaller, newer, and is the agency a company goes to when it wants a published letter but is not ready for a Big Three. For a private company that had never carried public debt, that is the sensible first step. What is less sensible is what KBRA was handed to work with.

Per the FT's reading of the file, Anthropic gave KBRA its revenue, its adjusted operating income, an annualised Q1 figure, and internal projections for Q2. It did not hand over free cash flow. It did not hand over EBITDA. KBRA estimated both using a risk-scoring method rather than its usual full-model approach — the agency's documented workaround for issuers that will not surrender the inputs needed to run the standard model. A BB+ stamp assembled on a risk-score shortcut is still a letter. It is not a letter you would want to defend in a bond prospectus to a Vanguard fixed-income desk.

The implicit story, from the KBRA record and the Blackstone walk-back together, is that the credit side of Anthropic's capital stack is being priced on a limited data set, and the operator of the biggest private-credit tranche has, in the last four months, moved its own prediction one tier lower. Neither of those facts stops the IPO; both of them travel into the post-IPO bond market with the stock.

Why this one data point actually matters

The temptation on a story like this is to shrug and say the equity market will decide the number regardless. It almost certainly will. The S-1 is still pointing at a float somewhere in the trillion-dollar neighbourhood; the Series H was a $965 billion post-money; the market will probably clear. The reason to read the Blackstone revision carefully is not that it changes the IPO price. It is that it changes the cost of what the IPO is paying for.

Anthropic has committed to roughly $517 billion of compute over an eleven-month horizon (TheInformation's September scoop, carried by the Loop in the Series H write-up). Project Big Sky's $35 billion covered a slice; the FT's October reporting of a second package — ~$60 billion for the Google TPU buildout, carrying an $18 billion junior tranche and routed through Broadcom-guaranteed senior pieces — covers another, and leaves the balance to equity and to further debt that has not yet been raised. The whole stack is being structured around one bet: that Anthropic can get to investment-grade on the Big Three scale within one or two refinancing cycles, at which point security interests get released, covenants loosen, and the funding cost drops into the money-centre bond market where OpenAI and Microsoft already borrow.

A BB+ with a shortcut methodology is the current floor. If Moody's, S&P and Fitch rate Anthropic in the same neighbourhood after the IPO — and the Blackstone walk-back suggests the private-credit market is now bracing for exactly that — the covenant releases do not trigger, Broadcom's guarantee has to stay on, and the next $150 billion of chip paper gets priced off a double-B curve rather than a BBB one. The spread difference between BB+ and BBB- at November 2026 rates is roughly 150 basis points on a seven-year term. On a $150 billion tranche, that is $2.25 billion a year in interest Anthropic pays rather than Microsoft.

OpenAI, borrowing against Microsoft commercial-arrangement revenue that showed up as a $10 billion line in Microsoft's Q1 FY27 10-Q this week, is on the other side of that spread. The gap between how the two labs fund their compute is one of the five things that will determine which of them is still shipping frontier models in 2029.

The context around the forecast — a week when the cracks became numbers

The Blackstone revision is not landing in a calm market. On the same Friday that the FT published, TechCrunch's Rebecca Bellan ran "Cracks are emerging in the AI spending boom," a straightforward read of the four hyperscalers' Q3 earnings:

• Alphabet, Amazon, Meta and Microsoft announced capex plans this year totalling more than $740 billion.
• Alphabet printed negative $9.1 billion of free cash flow in Q3 — the second straight quarter in the red — on $51.4 billion of quarterly capex and up-to-$205 billion of guided capex for 2026.
• Amazon printed negative $42.4 billion of free cash flow over the trailing twelve months, down from +$14.8 billion a year earlier. Guided capex for 2026: $220 billion.
• Meta printed negative $2.1 billion in the quarter, on up-to-$145 billion of capex. Zuckerberg described the Muse personal AI as "very compute-intensive to run" — which Bellan accurately characterised as "a bit of a money pit."
• Microsoft was the only name with positive free cash flow at $18.6 billion, and only because it posted $45.2 billion of capex against a guided $50 billion. Microsoft told the Street the gap was supplier capacity, not restraint.
• S&P Global Market Intelligence counted $326 billion of debt raised by Microsoft, Alphabet, Amazon, Meta and Oracle year-to-date in 2026 — three times the full 2025 total.

Equity markets responded in the usual proportionate way. Nasdaq −2.9 per cent on Friday. Nvidia −4 per cent. Broadcom down more than 5 per cent. Qualcomm −11 per cent. Alphabet −12 per cent on Thursday, its worst day on record, $500 billion in market cap erased in a session. Amazon −7 per cent after earnings. Meta −10 per cent the prior week.

In that week, Blackstone's revision reads less as a company-specific worry and more as a credit-market tell on an entire asset class. The private-credit funds that are financing the AI buildout are the same funds that read hyperscaler earnings and model the second-order effects on compute pricing. If Alphabet is burning cash to buy its own TPUs, the TPU volume Anthropic needs to pay for keeps the price up; if hyperscaler-adjacent debt is now carrying triple the normal annual issuance, the spread the market demands on new AI-adjacent paper moves up with it. A double-B on Anthropic is partly a judgment on Anthropic and partly a judgment on the whole AI-compute debt curve.

The equity-versus-credit split, as a sentence

Anthropic is about to be a one-day valuation story and a decade-long credit story. The equity will land at whatever Goldman and Morgan Stanley's institutional-sales desks tell them it will land at. The debt will be re-priced every six months by whichever agency is willing to look at the file. The gap between the two numbers — the equity capitalisation and the implied credit quality of the bonds the equity is funding the growth of — is the real datapoint. On October 30, 2026, that gap widened by one tier in Blackstone's own deck.

For a company that had zero debt on its balance sheet on June 30, 2026, and is lining up roughly $100 billion of it over the next twelve months to pay for compute it has committed to using, the floor rating is the hinge on which the arithmetic works or doesn't. Blackstone can revise its forecast privately. Moody's, S&P and Fitch will do it on the public tape.

That is a sentence that costs 150 basis points.

What to watch

1. The S-1/A pricing range. When Goldman and Morgan Stanley file the amendment that sets the share-price band ahead of the November roadshow, read it against the Series H $965 billion post-money. A band well below Series H means the equity book is reading the credit market's doubt and discounting for it; a band comfortably above means institutional buyers are willing to pay for the growth narrative regardless of the bond side. The S-1 and the pricing amendment are separate documents on EDGAR; the amendment is the one that counts.

2. Moody's, S&P and Fitch, within 90 days of the IPO. Big Three ratings on Anthropic are the thing Blackstone is projecting and KBRA has pre-sketched. If the Big Three cluster around BB+/BB with stable outlooks, the Blackstone revision was the market pricing in advance. If any of them lands at BBB− with a positive outlook — a one-notch gap but a materially different cost of capital — Anthropic argued its way past the credit-market resistance in the roadshow. The timing is tight because the covenant structure in Project Big Sky depends on investment grade from two of the three, which lets security interests be released. Those are not legal fictions; they move cash.

3. The second $60 billion package. The follow-on debt stack for the Google TPU buildout — ~$42 billion of Broadcom-guaranteed senior, $18 billion of junior exposed to Anthropic's credit risk — has not fully priced. The junior tranche's yield is the real tell. If it clears above 10 per cent, the market is pricing a double-B view. If it clears closer to 8.5 per cent, Big Sky's spread held and the Blackstone revision was more cautious than the paper itself needed to be.

4. Whether Anthropic hands the Big Three what it did not hand KBRA. A BB+ assembled on a risk-score shortcut, without free cash flow or EBITDA, is a letter the Big Three will not replicate. For Moody's, S&P and Fitch, Anthropic will have to show the inputs. Whatever those inputs reveal — the real cash-conversion ratio from a $16.3 billion H1 to a committed $517 billion compute spend over eleven months — is the number that decides whether the double-B holds or gives way to a tier either side. The private disclosure to KBRA was selective; the Big Three disclosure will not be.

The IPO itself will be the equity story. The year after it will be the credit story. For now, four days after the S-1 landed and ten days before the roadshow starts, the lead subordinated lender has told its own clients to expect the second one to be messier than they had been promised.