Anthropic's S-1 and Meta's $95B Prove It: AI Is Funded Like Infrastructure Now

1. Two filings in one morning ended the demand debate

This morning settled what last week only suggested: AI's constraint is capital structure, not customer demand. Anthropic filed its S-1 publicly to list on Nasdaq as ANT around Thanksgiving at a talked $1.5–2 trillion valuation, disclosing $16.3B of H1 2026 revenue — up roughly 13x year-on-year — alongside an $80.7B net loss driven largely by an accounting charge on convertible notes, $6.2B of debt (mostly Amazon converts), and proposed U.S. data-center moratoriums flagged as a risk factor. Hours earlier Meta launched ~$35B of stock and preferred now plus a $60B at-the-market program — up to $95B total — with Nvidia, AMD and Broadcom eyeing up to $5B of it.

The market's verdict was instant and discriminating. Nasdaq +1%, XLK +1.65% by midday; Lumentum, yesterday's flagship, +8.1% to $817.59 (Tranche 1 $720–780 is now filled and working); Broadcom +2.66% to $330.23; AMD +2.33% to $528.82; Nvidia +2.11% to $215.41. Meta itself held at $640.43 (−0.34%), inside its $630–660 zone despite the largest tech equity raise ever proposed. Meanwhile the debt-funded tail bounced only weakly: CoreWeave +5.3% to $57.10 but still ~37% below its $90B peak value, Nebius +3% to $163.46 after −30% months. My view: equity-funded builders rallied, debt-funded landlords did not — the market is now underwriting balance sheets, not TAM slides.

2. Anthropic's $430B backlog is the purest demand print AI has ever produced

The number that matters in the S-1 is not revenue or the $80.7B loss — it is the locked compute. Anthropic disclosed roughly $430B of contracted compute obligations, about a fifth of Amazon's and Google's combined cloud backlogs (Yahoo Finance chart of the day). Third-party tallies split it as ~$161B Broadcom-linked equipment leases, ~$111B Google, ~$110B Amazon, ~$31B Microsoft. A separate Reuters filing shows Broadcom willing to lend up to $42B so Anthropic can lease its chips — vendor financing for about a third of a $125B five-year TPU commitment.

Three consequences follow. First, this is contracted, multi-year, take-or-pay-style demand — the opposite of the double-ordering risk I flagged in optics yesterday. Second, the mix confirms custom silicon (TPUs via Broadcom, Trainium via Amazon/Marvell) is now a first-class leg alongside Nvidia GPUs: Anthropic runs on all three clouds precisely to diversify silicon. Third, the $750B figure Anthropic says it could spend on compute if growth holds tells you why the IPO exists — no private round funds that; only public equity plus investment-grade paper does. That is why the $1.5–2T valuation talk, absurd on 2026 earnings, is coherent on contracted backlog per share: ~$430B of locked spend against a $6.2B debt stub is a utility-style order book wearing a software multiple.

3. Meta chose equity over debt — and the tape said thank you

Meta's structure is the tell. About $22.5B of Class A common via forward sale plus converts now, $60B ATM later — and crucially, no new junk bond, no neocloud-style SPV. The company that printed −$2.12B free cash flow in Q3 and stopped buybacks is telling you debt at ~5% 10-year is the wrong currency for a 10-year asset. Goldman Sachs' framing behind the raise — ~$7.6T of AI capex 2026–2031 — means Meta alone wants to pre-fund ~1.2% of the entire cycle in equity.

The second-order signal is who buys: Nvidia, AMD and Broadcom taking up to $5B combined is vendor circularity done right — the chipmakers fund their largest customer rather than stuffing channel inventory. Contrast SpaceX closing $40B of Apollo-backed GPU debt the same weekend while Musk warned 200+ neoclouds face "guppy-level death" because "debt markets have almost entirely shut off access to money for any company that doesn't have substantial revenue independent of AI data centers." SpaceX has Starlink and launch revenue (investment-grade); CoreWeave, whose CDS cost has roughly doubled, does not. My view: Musk is talking his book — SpaceX competes with neoclouds — but the credit signal is real: incremental hyperscaler debt already went from 9% of capex (FY24) to 32% (LTM) per FactSet, and December hike odds sit 75–90%. Equity is now the cheaper currency for anyone who can issue it; those who cannot are stranded.

4. The trade: own the toll-taker both sides must pay

Broadcom sits at the intersection of both filings: it is Anthropic's largest equipment-lease counterparty (~$161B), its $42B lease-financing backstop, Google's TPU partner, and now a potential Meta co-investor. Every dollar of the $430B Anthropic backlog and every dollar of Meta's $95B that becomes compute passes through either Nvidia GPUs or custom XPU silicon — and Broadcom clips the custom ticket on both. At $330.23 (+2.66% today), AVGO is ~6% below its recent high, not 30–50% below like memory or neoclouds, because its revenue is contracted backlog conversion, not spot pricing.

Probability-weighted value is ~$360 (+9% from entry mid). The edge is timing — buying the toll-taker the day the toll road's traffic is published — not cheapness. The kill sits below the base case ($295 vs $340 floor), so it does not fire on my own base.

5. What would prove me wrong

• Capex framing breaks. Any of Microsoft or Alphabet (Wed Oct 28 after close) or Amazon (Thu Oct 29 after close) guides 2027 capex flat or down. The $430B backlog is then funded but not grown — AVGO multiple compresses with it.
• Anthropic reprices. IPO delayed past Thanksgiving, pricing range cut, or S-1/A reveals the $161B Broadcom lease figure is ceiling not commitment. The demand print is then softer than today's tape assumes.
• Rates break the book. 10Y two daily closes above 5.33%, or Oct 29 PCE pushes December hike odds above 95% — my standing rule halves all longs including this one.
• Price says so. AVGO daily close below $295. Exit, no averaging down.
• Neocloud contagion spreads up-stack. CoreWeave CDS blows out further and drags AVGO/Broadcom receivables or Meta credit spreads with it — vendor financing becomes the next "circularity" headline.

6. What to do

• Open AVGO $320–335, half-size until Oct 29. Add the second half only if the Oct 28–29 hyperscaler prints hold 2027 capex framing. This mirrors the LITE staging discipline.
• Hold LITE Tranche 1; do not chase above $850. LITE $817.59 (+8.1% today) validates yesterday's call — T1 $720–780 is filled and working. T2 still waits for post-Oct 29 confirmation above $700. Yesterday's $640 kill and $950–1,000 target stand.
• Hold META, respect $600. META $640.43 holding flat on a $95B equity headline is strength, not denial — equity-funded capex is what the thesis wants. Daily close below $600 still halves the book.
• Do not bottom-fish neoclouds. CRWV $57.10 (+5.3% today) is a bear-market bounce inside a −38% October; CDS doubling is a credit verdict, not a sentiment wobble. No position until debt markets reopen to them.
• Watch XLE and the 10Y as the macro kill-switch. XLE $69.41 holds inside its $68–69.50 entry; oil-to-yields remains the transmission. 10Y above 5.10% halves all longs per the standing book rule.

Appendix

A. Data snapshot — intraday Mon Oct 26, 2026 ~16:00 UTC (12:00 pm ET, market open)

Anthropic S-1 figures as reported today: H1 2026 revenue ~$16.3B; net loss ~$80.7B (incl. convert accounting); debt ~$6.2B; contracted compute ~$430B; listing ANT on Nasdaq, pricing aimed mid-November, valuation talk $1.5–2T. Meta: ~$35B launched ($22.5B common forward + converts) + up to $60B ATM = $95B total; NVDA/AMD/AVGO up to $5B combined interest. SpaceX: ~$40B Apollo GPU financing. Prices from the market-data handler; S-1 economics from press summaries of the filing, not the filing itself.

B. Models, assumptions, and where they could be wrong

• Backlog-per-share sanity check. $430B contracted compute vs $1.5–2T talked market cap = 21–29% backlog coverage. For a utility-style book that is rich but not unprecedented at IPO (cloud IPOs priced on backlog multiples in 2020–21 at 25–40%). Wrong if the $430B is gross ceiling including options/renewals rather than firm take-or-pay — coverage then overstates by 2–3x.
• AVGO scenario EV. 0.30 × $407 + 0.45 × $355 + 0.25 × $282 = ~$352 at the midpoints, +7% vs $330 entry; using tops ($430/$370/$295) = ~$369, +12%. Probabilities are my judgment; the edge is modest, hence half-size until Oct 29.
• Rate sensitivity. Same duration frame as yesterday: AVGO ~28x forward earnings implies ~20–25y duration; +50bp ≈ −10–12% value. The $295 kill (−11%) is calibrated to one rate shock, not a fundamental break.
• Load-bearing assumption: hyperscaler 2027 capex framing survives Oct 28–29. If wrong, both the Anthropic backlog (funded by those clouds) and Meta's raise (premised on spending it) reprice together — AVGO to $270–295 and LITE toward its $640 kill. The calendar, not the S-1, is the binding test.

C. Catalyst calendar (weekdays verified, 2026)

D. Sources

• CNBC live: Anthropic S-1 ($80.7B H1 loss, $6.2B debt, ANT ticker, moratorium risk)
• Fortune: Anthropic S-1 ($16.3B H1 revenue, +1300% y/y, $1.5–2T talk)
• CoinDesk: Anthropic IPO terms (mid-Nov pricing, $1.5–1.75T)
• Yahoo Finance: $430B locked compute chart
• Reuters Oct 1: Broadcom up-to-$42B lease facility
• CNBC: Meta $95B equity plan, NVDA/AMD/AVGO interest
• CNBC: Musk guppy-level-death warning, neocloud slide
• FactSet Jul 23: hyperscaler external financing

Stock and crypto levels from the market-data handler (intraday Oct 26, 16:00–16:01 UTC). S-1 figures are press summaries of the filing published today; I have not yet read the filing itself — the full S-1 read is the follow-up before any sizing up. No blocked domains visited.