To The Moon

The Tell: SoftBank's ¥1.12 Trillion Loss Was No Accident

On November 10, $Softbank Corp.(SOBKY)$ reported a quarter the market wasn't positioned for: a net loss of ¥1.12 trillion (~$7.0 billion) for the fiscal second quarter ended September, against analyst expectations of roughly ¥129 billion in profit — and versus a ¥2.5 trillion profit a year ago.

The bleeding didn't come from OpenAI — that stake still sits on the books at a fair value of $99.6 billion, unchanged from March, with no impairment taken.

The real damage came from a corner few people were watching: power equipment for US AI data centers. Silver Bands 7, a subsidiary held with Vision Fund 2, had pre-purchased generation equipment capacity for projects in Ohio and elsewhere; when project plans were adjusted, ¥597.2 billion (~$3.73 billion) was written off, ¥457 billion of it non-refundable prepayments.

Meanwhile, SoftBank's financing costs jumped from ¥174.7 billion a year ago to ¥461.9 billion — up roughly 2.6x — on the back of a new $11.87 billion term loan, $10 billion in dollar bonds, ¥1 trillion in yen bonds, and an Arm-share-collateralized margin loan now swollen to roughly $25 billion. The most leveraged player in the AI chain is now paying for the leverage itself. The market voted: SoftBank shares are down more than 50% from their June peak.

Layer Two: The Neoclouds Can't Raise Money

If SoftBank is the most aggressive equity investor in the AI capital chain, the neoclouds are its most aggressive debtors. That layer is now being shut out of equity and credit markets simultaneously.

On the equity side: $CoreWeave, Inc.(CRWV)$ closed at $44.13 on November 4, an all-time low since listing, down ~43% in a month; IREN fell ~45% and Nebius ~39% over the same stretch.

Credit looks worse. CoreWeave's 9.75% coupon senior bonds now trade at an implied yield above 15% — that's distressed-debt pricing, not growth-stock pricing.

It carries $35.1 billion of debt against just $5.0 billion of equity, and spent 24.9% of quarterly revenue on interest alone.

Gartner's October forecast reads like an obituary for the category: by 2030, over 80% of neocloud providers will no longer operate independently.

Firmus pulled its IPO; AirTrunk shelved its Singapore listing — the primary market door is closed too.

Layer Three: Big Tech's Free Cash Flow Has Gone Negative

The neoclouds' money was borrowed. Big Tech's money was supposed to be its own. That's why the real signal this earnings season was in the cash flow statement, not the income statement:

• $Alphabet(GOOG)$ : Q3 capex of $51.36 billion — double year-on-year — against $42.29 billion of operating cash flow, for FCF of negative $9.07 billion; trailing-twelve-month FCF shrank 73% to $19.7 billion.

• $Meta Platforms(META)$ : Q3 capex including finance-lease principal of $34.91 billion exceeded $32.79 billion of operating cash flow — FCF negative $2.12 billion, its first negative print in a decade. Buybacks went to zero, replaced by a $12.5 billion mandatory convertible preferred, a $25 billion common-stock program, and a $60 billion non-voting Class C shelf — a potential $97.5 billion financing stack.

• $Amazon.com(AMZN)$ : Q3 capex of $77.3 billion drove trailing-twelve-month free cash flow to a negative $42.4 billion outflow, with 2026 cash capex guidance still at ~$220 billion.

Combined, the four majors' TTM free cash flow fell from $209.1 billion a year ago to $62.4 billion — down 70%.

$Microsoft(MSFT)$ s cash flow hasn't gone negative yet, but it trimmed FY2026 capex guidance from $175 billion to $170 billion

— the first time in this cycle a hyperscaler has revised the number down.

Layer Four: The People Writing the Checks Are Walking Away

At the very end of the chain — the end-demand that all of this financing ultimately rests on — the signals are even blunter.

Microsoft borrows roughly ten percentage points cheaper than CoreWeave, and it is now backing away from neocloud deals.

The logic is simple: when self-build is cheaper than subleasing, and the customer's credit quality alone moves your funding rate by 225–300 basis points, the middleman's business model loses its reason to exist.

OpenAI was shopping a $30 billion round in late September; Bloomberg's November 6 report says it has raised $11 billion so far and the round will likely close around $20 billion, at a $1.25 trillion valuation.

A $30 billion target, and even $20 billion is a stretch.

Anthropic's expected IPO valuation is being marked down by hundreds of billions per week: the conversation in early October was near $2 trillion; on November 2 the WSJ reported institutional feedback at $1.5–1.6 trillion; by November 6 Bloomberg's range was $1.1–1.3 trillion — just a 14–35% premium to its $965 billion private round in May.

One number in the S-1 explains buyers' hesitation: purchase and lease commitments for future compute total roughly $895.4 billion, against $131.8 billion of cash on hand.

The Last Straw: December's Yen

One more thing that doesn't sit on any Silicon Valley balance sheet but weighs on the whole chain: the Bank of Japan.

The BOJ held rates at 1.25% on October 30 but, for the first time in 13 years, declared its 2% inflation target met; markets now price roughly an 80% chance of a 25bp hike to 1.5% at the December 17–18 meeting.

Former BOJ chief economist Kameda put it bluntly: short of a total collapse of the AI trade, nothing stops a December hike.

The yen is the cheapest funding currency in this system — SoftBank's margin loans, the global carry trade, the liability side of everything connects to it. Doubled financing costs are already printed in SoftBank's earnings. One more BOJ hike is one more tax on everyone who bought GPUs with cheap yen.

Conclusion: CAPEX Has Topped, and the Semis' Excess Return Is Over

String the layers together and the conclusion is hard to dodge:

1. Equity capital (SoftBank, OpenAI's new round, Anthropic's IPO) is shrinking in size and price;

2. Debt capital (CoreWeave at a 15% implied yield, Firmus pulling its IPO) is closing;

3. Internal capital (negative FCF at GOOG/META/AMZN, Microsoft cutting capex guidance) is being overdrawn;

4. The master valve of cheap money (yen rates) is being tightened.

Each layer alone could be dismissed as an idiosyncratic event. Together, in the same quarter, pointing the same direction — this isn't demand disappearing (Google Cloud is still doubling year-on-year). It's that the willingness to fund that demand has peaked. AI capex, at least for this cycle, tops out here.

The positioning implication is straightforward: once the marginal buyer of compute shifts from "secure capacity at any cost" to "prioritize within cash-flow constraints," semiconductors no longer offer sufficient excess return. For three years, buying NVDA and the supply chain was a bet on capex revisions going up. That variable is now flattening, or turning down — what's left is a valuation and crowding problem. The cycle money has been made. Don't fight the balance sheet.

All data as of November 10, 2026, from public filings and press reports. This is an opinion piece, not investment advice.