Track: anthropic (sibling to SpaceX under the AI-mega-IPO-supercycle frame — see [[2026-05-27-anthropic-ipo-compute-financing]]). Methodological template: [[2026-06-03-spacex-net-flow-model-passive-demand-vs-lockup-supply]] + [[2026-06-17-spacex-post-ipo-study-4-float-melt-up-vs-the-dated-staggered-lockup-overhang]].

Question: Anthropic's S-1 draft is confidential (filed ~2026-06-01); no float %, lockup ladder, or share count is public. This is the fill-in-the-blanks scaffold so that when the S-1 surfaces we plug in parameters rather than start cold — exactly how the SpaceX net-flow model was built parametrically pre-pricing, then graduated to deal-confirmed.

Verdict (prelim): The SpaceX framework transfers with a strong prior — low-float melt-up → dated lockup overhang — because the driver is structural (bankers float a small scarce slice; concentrated locked insiders → staggered unlock → overhang). The parameters are gated on the S-1. Three Anthropic-specific risks differentiate it from SpaceX, and one of them (the SPV stack) is a supply/valuation-integrity hazard SpaceX doesn't carry as badly.

Status (S-1 watch, 2026-06-17)

S-1 watch (task TASKS-RESEARCH.md → "Anthropic + OpenAI S-1 filing watch") — trigger NOT fired: the S-1 is still confidential, so the AI-Lab S-1 → Thesis Crosswalk Playbook does not run yet. Confirmed via web:

• Confidential draft S-1 submitted 2026-06-01 (Anthropic's own announcement). Not on EDGAR; per SEC rules the prospectus goes public ≥15 days before any roadshow. No share count / float % / price range / ticker / exchange set.
• Fresh primary mark: $65B Series H at $965B post-money, announced 2026-05-28 — this supersedes the stale late-April ~$1T SPV-implied number as the cleanest valuation anchor (and, being a primary round rather than an SPV markup, partially de-risks the SPV-stack mark concern for the headline figure).
• Federal-revenue line is now live: the Pentagon's $800M agentic-AI award (split xAI/OpenAI/Google/Anthropic, FY2026) means Anthropic carries a government/defense revenue line into its S-1 — a revenue-quality / concentration / political-risk read item (added below). (Cross-ref the gov-compute integrators gap.)

Trigger stays armed; revisit when the public S-1/A with financials surfaces (likely after the SEC review completes).

The transferable framework (what we can assert now)

Net flow = forced/passive demand − unlock supply. SpaceX showed the bid (≈$4–30B early index inclusion) is an order of magnitude below the staggered unlock (≈$150–350B), so net flips bid → overhang at the first earnings release. The same equation will govern Anthropic; only the magnitudes change. The melt-up itself is a marginal-price-on-locked-shares artifact: a thin float lets a small amount of net buying levitate a paper cap, and the same thinness makes it cheap to unwind once the lockup converts locked shares into real float.

Fill-in-the-blanks (gated on the public S-1)

Three Anthropic-specific risks (the prelim notes)

1. SPV-stack supply + valuation-integrity risk. Late-stage Anthropic exposure trades through stacked SPVs (SPVs-of-SPVs), each layer adding fees/carry and marking up the implied valuation — so the headline "~$1T" is several markups removed from any clean primary round. Two IPO consequences: (a) every SPV layer is a motivated post-lockup seller wanting an exit (extra unlock supply beyond insiders); (b) down-round risk — if the IPO prices below the SPV-stacked marks, the whole pyramid is underwater. Not (usually) literal fraud — layered-fee opacity + a hidden selling cohort. (Well-reported for late-stage private AI broadly; no Anthropic-specific primary source in-workspace yet — flagged as a risk to verify, not asserted as fact.)
2. Circular-financing mark reversal (compounds #1). Open question #2 in the Anthropic thread: AWS/GOOG can mark their stakes up only while the next round prices higher. A flat-to-down IPO print reverses the mark — AMZN's AI-segment "profit" (~50% of which is Anthropic stake mark-to-market, not operating leverage) turns into a paper loss, same for GOOG. The SPV froth and the circular-financing fragility share a single trigger: a flat-to-down IPO. That's the read to watch on AMZN/GOOGL, the only public proxies.
3. No clean public proxy / ARR-quality. Unlike SPCX's SATS stub, there's no public Anthropic ticker — the halo (if any) hides in AMZN/GOOGL segment economics. And the ~33× sales only looks tame if the $30B ARR is cash-receipt; open question #1 (cash-vs-accrual, compute-prepaid draw-downs) means the effective multiple could be materially worse. The "crazy multiple" manifests as ARR-quality + circular fragility, not raw P/S.

What this does NOT decide

No trade, no valuation call, no float estimate. This is a scaffold. The lane's standing Trading Posture (parent thread) already applies: good company ≠ good trade; the IPO pop is a trap for the real book; wait for the post-lockup reset; labs = double-patience (no dated calendar until the S-1).

When the S-1 surfaces (the actual work this unblocks)

1. Plug float % + share count + lockup ladder into the net-flow equation; recompute the dated overhang (mirror the SpaceX post-IPO study).
2. Pull deferred-revenue / contract-liability footnotes → resolve ARR cash-vs-accrual (open Q#1).
3. Compare IPO price vs last private round / SPV marks → grade down-round risk (open Q#2, risk #1+#2). Anchor = the $965B Series H mark (2026-05-28).
4. Read the federal/defense revenue line (Pentagon agentic-AI program + any classified/IC contracts) → grade revenue concentration, political/regulatory risk, and whether govt revenue is a quality premium or a single-customer concentration flag.
5. Re-grade the parent-lane drain/halo classifier for an Anthropic event window.