Physical Address
304 North Cardinal St.
Dorchester Center, MA 02124
For most of this cycle, the people setting the price of artificial intelligence have been equity investors. They rewarded every larger capital budget, treated spending as proof of opportunity, and assumed the returns would arrive on schedule. In the last week of October, a different group of analysts started talking, and they ask a much narrower question. Not how big this could get, but who gets repaid, and out of what cash.
Inside roughly two trading days, a major ratings agency put one of the strongest balance sheets in technology on formal notice, the anchor lender behind the largest chip financing on record told its investors that the AI lab it is funding would remain below investment grade after its listing, and the sector’s most valuable private company cut the size of its fundraising round by a third. Stock prices reacted violently. The credit signals matter more, because credit analysts tend to move earlier and change their minds more slowly.
Meta Gets a Warning, Not a Downgrade
Start with what actually happened, because several headlines overstated it. On 30 October, S&P Global Ratings kept Meta Platforms at AA- while shifting its outlook from stable to negative. The agency’s reasoning was that Meta, unlike rivals who rent computing power to outside customers, has to earn its AI payback through its own advertising and products, and that path is less certain. A negative outlook is a formal caution that a cut could follow. It is not a cut, and Meta remains among the highest rated companies on the planet.
The arithmetic behind the caution is still remarkable. S&P now pencils in capital spending of about $140 billion this year, more than $225 billion in 2027 and around $252 billion in 2028, all far above its earlier estimates. On those figures, free operating cash flow turns negative this year, reaches a deficit of roughly $71 billion in 2027 and stays below zero in 2028. Adjusted leverage, which sat near 0.1 times in 2025, rises through the agency’s 1.0 times downgrade threshold in 2027 and heads toward 1.3 times a year later.
Meta has pushed back publicly, and its core advertising business is not the problem: third quarter revenue rose 26%. What changed is the funding model. Meta has raised about $37.5 billion through new common stock and mandatory convertible preferred shares, scheduled a further $60 billion share sale programme starting in January 2027, and carries more than $355 billion of future lease obligations. When an AA-rated company starts issuing equity on that scale, it is telling you something about how it views its own cash generation.
Alphabet’s Quarter Exposed the Earnings Mix
Alphabet dropped nearly 13% on 29 October, its worst session since listing in 2004, after reporting free cash flow of roughly negative $9 billion for the quarter while keeping full year capex guidance between $195 billion and $205 billion. Across the month, the company shed about $758 billion of market value, its weakest month on record, and the selling pushed its price to earnings multiple down to the lowest level in more than ten years.
The more instructive detail sits underneath operating income. In the second quarter, Alphabet booked net income of about $112 billion, and its own release attributed a $98 billion other income gain primarily to unrealised gains on equity securities. Coverage of that quarter traced those gains mainly to its stakes in SpaceX and Anthropic. By the third quarter, the same line had flipped to a loss of about $8.7 billion, driven largely by SpaceX.
Neither number tells you anything about search, YouTube or Google Cloud. An investor who put a market multiple on the second quarter headline was paying for a revaluation of private holdings, not for the business. That is the central earnings quality issue of this phase of the cycle: a meaningful share of reported profit at the largest technology companies now depends on what a few private AI and space companies are worth in their latest funding round.
Amazon: Record Margins, Borrowed Growth
Amazon’s results rhyme with Alphabet’s. Amazon Web Services grew 39% to $46 billion and group operating margin reached a record 13.9%, which on its own would have been a strong quarter. Then comes the cash flow statement. Capital spending hit $77.3 billion in the three months to September, trailing free cash flow swung to an outflow of $42.4 billion from an inflow of $14.8 billion a year earlier, and long term debt climbed from about $65.6 billion to $173.1 billion in nine months.
Amazon’s reported earnings this year were also lifted by its Anthropic stake. Non-operating gains on that investment added $16.8 billion in the first quarter, and Anthropic accounted for most of the $53.4 billion of other income in the second. The third quarter comparison looked weak partly because the prior year period had included a $9.5 billion Anthropic gain.
The Loop Investors Need to Untangle
This is where structure matters more than any single number. The largest cloud providers now sit on every side of the same relationships. They are equity investors in the frontier labs, suppliers of chips and data centre capacity to those labs, and beneficiaries every time a lab raises money at a higher price. A markup in a funding round flows into reported profit. Compute purchased by the lab flows into reported cloud revenue. None of this is improper, and all of it is disclosed. But it does mean a slice of headline profit and a slice of headline growth rest on the same small group of counterparties continuing to raise capital at rising valuations.
Some commentators have gone further and argued that the labs are concealing their true costs. The filings do not support that claim as stated. Anthropic’s registration statement shows first half 2026 revenue of $16.3 billion, compute and infrastructure expense of roughly $8.9 billion, and an operating loss of $693 million. The far larger $80.7 billion net loss is dominated by a $72.4 billion non-cash fair value charge on convertible notes, according to a detailed review of the S-1 by Deep Quarry. Compute is expensed as it is used. What looms over the business is not a hidden current cost but a disclosed future one.
That future bill is large. The same review found roughly $321 billion of leases that have not yet commenced and about $265.6 billion of minimum purchase commitments in the footnotes. Yet the summary table of contractual obligations in the public filing is entirely blank, and has stayed blank through every confidential draft since June. Placeholders in a preliminary prospectus are routine and must be completed before a listing becomes effective. Still, the single figure a credit analyst most wants to see is the one that has not yet been printed.
Anthropic’s Junk Rating Is Not New. The Lost Upgrade Path Is
Anthropic’s credit already sat below investment grade. Kroll Bond Rating Agency assigned it BB+ in June, one notch under the line. What changed on 30 October was the expected direction. The Financial Times reported that Blackstone, the anchor investor in the $63 billion Project North Star chip financing, told its investors it now expects a high junk rating in the BB range after the IPO, reversing earlier guidance that investment grade could arrive before 2027.
The structure of that deal explains why the rating matters so much. More than $42 billion of senior loans are guaranteed by Broadcom, which carries a strong single A profile. Around $18 billion of junior debt rests on Anthropic’s own lease payments, and Blackstone funds took about $9 billion of that slice at yields above 10%. Strip away the Broadcom guarantee and the market is pricing Anthropic’s standalone credit at double digit yields.
The practical consequence is access. Insurers and pension funds, which together control a vast share of the world’s long term capital, face limits on how much sub-investment grade debt they can hold. A company planning to spend more than $800 billion on computing infrastructure through 2045 needs that pool of money, and a BB rating keeps much of it out of reach. It also means continued reliance on better rated partners to guarantee borrowing.
The contrarian caveat runs the other way. BB is the top tier of speculative grade, not the edge of default. Anthropic holds more than $112 billion in cash, and investors expect the listing to raise around $100 billion more. This is a cost of capital story before it is a solvency story. The deeper worry is collateral. Debt secured against chips and data centre leases is only as safe as the demand for that capacity, and the demand comes largely from the same handful of borrowers. In a downturn, the collateral and the borrower would weaken at the same moment, which is precisely the kind of correlation credit committees are paid to fear.
OpenAI Finds the Edge of Private Demand
The private market delivered its own verdict in the same week. OpenAI cut its funding round to about $20 billion from $30 billion at a $1.25 trillion valuation, after first pitching $1.4 trillion. By the Friday commitment deadline, the first close was expected to come in below even the reduced target, with wealthy individuals to be offered the same price later through a separate offering. OpenAI had reportedly wanted the round finished before Anthropic lists, because the two compete for many of the same backers.
None of this makes OpenAI small. A $1.25 trillion valuation would still be the largest private mark in history. What changed is the marginal buyer. Investors who have written cheques in round after round are now reducing allocations, and the same Friday saw the Nasdaq Composite fall close to 3%, with Intel sliding by more than a quarter on its outlook.
What a Contrarian Does With This
The easy move right now is to swap one crowd narrative for another, from “AI can only go up” to “the bubble has burst.” Neither is analysis. Alphabet’s multiple has already compressed to decade lows, cloud growth at both Alphabet and Amazon remains exceptional, and Meta’s advertising engine is growing at a pace most companies would envy. Some of the damage is already in the price.
What the last week really changed is the burden of proof. The test is no longer whether AI demand exists, but whether the companies building for it can fund the gap between spending and payback without leaning ever harder on debt, leases, share issuance and gains on each other’s private stock.
That suggests a practical discipline for any portfolio with heavy technology exposure. Read the income statement in two layers, separating operating profit from mark to market gains on investments. Track free cash flow and, just as importantly, how capex is being financed. Treat long dated lease obligations as debt, because ratings agencies do. And watch credit before price, because rating outlooks and bond yields have flagged this shift weeks before many equity investors noticed.
The next markers are already on the calendar. Meta’s 2027 capex guidance with its fourth quarter results will show whether S&P’s estimates were too high. Anthropic’s amended prospectus must fill in its third quarter numbers and the total of its compute commitments. Moody’s, S&P and Fitch have yet to publish formal ratings on Anthropic. And Alphabet and Amazon have both said 2027 spending will rise again.
For two years, the equity market treated capital spending as a promise. The credit market is now treating it as a liability that has to be serviced. Investors do not have to pick a side in the AI debate to respect that distinction. They only have to notice which of the two markets has historically been right first.
