Neocloud Debt Stress Meets Record Equity: Meta $95B, CoreWeave, Week of October 21, 2026

Meta's reported record $95 billion equity raise and CoreWeave's bond yields near 15% show that AI capex is now funded by the strongest balance sheets while the weakest neoclouds face a debt-market squeeze.

The through-line is that AI infrastructure has become a capital-markets story, and the funding source now separates winners from losers. Meta's reported plan to sell up to $95 billion of non-voting Class C shares, reportedly a record equity offering, comes as its capex rose from $69.69 billion in FY2025 to guidance as high as $145 billion in FY2026 (C3). Meta's Q3 showed revenue up 26% to $64.6 billion but its first-ever negative free cash flow, which is the balance sheet this raise is funding (items 4 and 5). Microsoft's FY2026 capex reached $115.95 billion, up 79.6% (C6). Demand is now being disclosed in public for the first time: Anthropic's S-1 shows $417 billion in compute commitments amortized over 20 years, set against $16.3 billion of first-half 2026 revenue and an $80.7 billion net loss that is largely non-cash (C4).

What is actually shifting is that anchor-tenant contracts have become the collateral for the whole stack. Nscale's reported $44.6 billion compute deal with Anthropic in West Virginia (C31), TeraWulf's roughly $19 billion 20-year Anthropic lease (item 11), and SpaceX's $40 billion GPU lease with Apollo (item 22) are the contracts that lenders and equity buyers now have to underwrite. The weak link sits in the same chain. OpenAI's annualized revenue run rate of about $50 billion falls roughly $20 billion short of earlier estimates, and its roughly $30 billion raise at a $1.25 trillion valuation was trimmed after investor pushback (C25, item 21). A contract is only as good as the revenue behind its counterparty, and the market is now testing that.

The exposure is concentrated in neoclouds that have levered capex against those contracts. Applied Digital's FY2026 capex of $2.87 billion is 469% of revenue against a reported $6.4 billion debt load (C23). IREN's FY2026 capex of $3.00 billion is 424% of $707 million in revenue (C26), and Nebius's FY2025 capex intensity reached 767% of revenue (C9). CoreWeave is the stress test. Five-year credit-default swaps imply roughly 60% default odds, and 2032 bonds have traded above 15% (C18; item 6 cites 63% and 16%). The stock fell 10.2% to $55.5 on October 21 (C18). Gartner's forecast that 80% of neocloud providers lose their independence by 2030, framed by a $4.9 trillion revenue gap (item 2), is the sector's bear case in a single number. The bull case is real but narrow. Goldman Sachs upgraded CoreWeave to a $139 target and Rosenblatt kept $250 (C18), and Nebius raised B300 GPU-hour pricing 21% to $9.50 (item 9).

The October 21 session sharpens the point. Arista fell 7.5% to $170.52 with no disclosed catalyst, despite quarterly revenue above $3 billion growing 37.7% (C2). Applied Materials fell 8.2% to $375.46 after record Q3 results (C24). Bloom Energy fell 9.1% to $216.77 that day and closed at $193.30 on October 23, down 11.2%, despite record Q3 results, raised guidance, and gigawatt contracts with Oracle and Google (C10, C12). Lam Research fell 11.8% with supply constraints cited (C11), KLA fell 9.5% without an explanation (C7), and Intel fell 8% to $85.50 (C16). TeraWulf fell 11.3% as its Muskie power contract doubled to 1 GW (C17), so even expansion is being punished. When companies with record results and rising contracts fall 8% to 12% without a stated cause, the market is repricing the financing chain and the multiple rather than the operating numbers. No column confirms a single cause, so this is an inference, but it is the most defensible reading of the pattern.

Power and policy are tightening the constraint. Trump's statement that data centers must supply their own electricity (item 19) favors builders with on-site generation or dedicated power, which is Bloom's pitch and why its Oracle 2.4 GW commitment and 702 MW Google-contracted Wyoming project matter (C12). The UK shows the downside. Microsoft withdrew from a Nscale UK data center deal as grid connections slipped into the 2030s (C6, item 15), which puts Nscale's planned IPO at risk. On chips, Reuters reported that ByteDance, Alibaba, and Tencent placed more than $35 billion of Nvidia orders, but secondary reports suggest deliverable supply covers only about one-tenth of that (C30). TSMC's reported 69% profit jump and $2 billion CoWoS interposer agreement with GlobalFoundries (C8) show packaging capacity being locked up by contract.

Watch four things next. First, whether Meta's Class C offering clears and sets the price for every hyperscaler raise that follows. Second, whether CoreWeave's 2032 bonds hold below 15% and its CDS stops widening. Third, whether OpenAI's $30 billion round closes at $1.25 trillion and whether Anthropic's reported $2 trillion Nasdaq listing proceeds on the S-1 timetable (C31). Fourth, whether Nvidia's China deliveries track the $35 billion order book. Before publication, reconcile conflicting inputs. CoreWeave's bond yield is 15.5% in one item and 16% in another, and its CDS-implied default odds are 60% in one column and 63% in another. Applied Digital is reported down 9% to $19.10 on October 21 in C19 but up 8.5% to $20.98 on a Wednesday session in C27, and October 21, 2026 was a Wednesday.