Capital Layer
← all notes

2026-09-02 · Pre-IPO & private markets

Anthropic pre-IPO and the Lighter perp — a price the venue draws itself

Data as of 2 Sep 2026 (Lighter UI and docs, public sources on Anthropic).

Bottom line: the ANTHROPIC perp on Lighter is the worst way to get exposure to the company. The docs state outright that price is determined "only by the internal pricing mechanism": it's an average of the venue's own order book, trading about $317k a day, with no external source and no price caps — and the venue's liability under its own terms is capped at $100. The price isn't cheap either: a mark of 1,928.9 × 1bn assumed shares = $1.93tn — double the last real round ($965bn, May 2026) and 60–80% above brokered secondaries. The perp is paying the full price of the $2tn IPO-target rumour before the IPO has happened.

Conflict of interest, stated up front: this breakdown was prepared with the help of Claude — a model made by Anthropic itself. The offset: every number carries an external source and date, and the central claims were checked by reading primary documents directly.

What it is

Two objects in one question. The ANTHROPIC perp on Lighter is a perpetual with no equity in the company: a cash-settled derivative against the venue's clearing entity (Elliot Technologies, Inc., arbitration in Miami) that "tracks" the price of one share out of an assumed 1bn shares. Anthropic is a private AI lab (Delaware PBC): last round $65bn at a $965bn valuation (28 May 2026), confidential S-1 filed 1 Jun 2026, reportedly targeting around $2tn and a Nasdaq listing in October 2026.

Series H round $965bn · 28.05.26 — real money Secondaries (brokers) $1.05–1.2tn · NPM/Caplight, thin trading Perps (18 venues) ≈$1.93–1.96tn · Lighter mark 1,928.9 × 1bn shares Rumoured IPO target ~$2tn · October 2026, Nasdaq — not confirmed by the company scale: $225bn per tick; the perp already pays the rumour price — ~4% upside to target, long funding ~3.5%/yr, plus venue risk on top

Four levels of "Anthropic's price" on 2 Sep 2026. Sources: Anthropic newsroom (round) · NPM/Caplight via Qz and Yahoo (secondaries) · DefiLlama via CryptoBriefing and the Lighter UI (perps).

1. The perp's index price is self-referential — and the docs say so

"Pricing is determined only by the internal pricing mechanism" — no external source: not Forge, not Caplight, not Chainlink (Chainlink's August expansion to 100+ pre-IPO markets doesn't cover this one). The index is a 30-minute average of the venue's own book; pre-IPO markets have no price caps; funding is dampened 100x; and the venue's own liquidity pool is likely the market maker in that book (an inference from the RWA section, not a direct quote). At $317k a day — against $12bn a month for the sector, where Binance holds 83% — a handful of traders sets the price.

Source: docs.lighter.xyz, pre-IPO markets and RWA pricing sections, 2 Sep 2026.

2. The perp pays the rumour price — and the asset class shows who controls settlement

$1.93tn against a $965bn real round is a 100% premium; against secondaries, +60–80%. The class has six precedents:

  • FTX, when converting its Coinbase contract, changed the divisor itself (250 → 261.3m shares);
  • the Notice.co oracle mishandled a SpaceX split — the perp fell 45% in 30 minutes, $1.5m liquidated;
  • Ventuals, while shutting down, settled everyone at its own TWAP;
  • Robinhood's OpenAI tokens were disavowed by the issuer the next day.

On Lighter, conversion or settlement at IPO is "at discretion, with 1 day's notice", and the "IPO doesn't happen" case is absent from the docs entirely. The fair other side: close to a dated listing, perps converge well — Cerebras within 1.3–3% of the Nasdaq open, SpaceX almost exactly ($154–172 versus a $160.95 close) — but three weeks before listing, the same SpaceX perp dropped 27% on no news.

Sources: The Block 12.2020 · CoinDesk 28 May 2026 · The Defiant 06.2026 · Talos/Coin Metrics 06.2026 · CNBC 12 Jun 2026.

3. Buying the real stock pre-IPO is practically impossible

On 12 May 2026 Anthropic declared transfers through 8 platforms (Forge and Hiive named) "void and will not be recognized on our books and records", and banned SPVs and forwards; Hiive delisted it. The only real liquidity event of 2026 — a $6bn employee tender at a $350bn valuation — was undersubscribed: insiders declined to sell. For an outside investor there is effectively no route into the stock before IPO, and what is being sold carries documented risk of the transfer being void.

Sources: TechCrunch, Axios 12–13 May 2026 · Bloomberg 4 Feb and 8 Apr 2026 · Hiive, 2 Sep 2026.

4. The company — unprecedented growth with two structural caveats

Revenue run-rate: $9bn (December 2025) → $65bn (July 2026), ahead of OpenAI ($40bn); investor expectations for year-end — $100–120bn. The copyright overhang is gone ($1.5bn, final on 20 Jul 2026).

0 35 70 Dec 25 · 9 Feb · 14 Mar · 19 Apr · 30 May · 47 Jul · 65 $bn annualised run-rate (not GAAP)

Anthropic's stated annualised revenue run-rate, December 2025 – July 2026. Run-rate is worth discounting 30–40% to an annual equivalent. Sources: CNBC, Bloomberg, TechCrunch for the respective months.

Caveat one: about $330bn of compute commitments (Google ~$200bn over 5 years · Azure $30bn · AWS ~$100bn over 10 years). That's roughly $50–55bn a year, or 0.8x the current run-rate; margins (estimates of 44–60%) and burn are known only at blog level — nobody has seen them before a public S-1.

Caveat two: public shareholders won't have a say — Delaware PBC, a trust with Class T shares and founder super-votes. You buy the economics, not control.

Sources: Anthropic newsroom 12 Feb and 28 May 2026 · Bloomberg/CNBC 17 Aug 2026 · Microsoft blog 18 Nov 2025 · The Information 5 May 2026 · Authors Guild 20 Jul 2026.

5. If synthetic exposure is needed anyway, Lighter is the worst option available

Binance (the sector leader) and Coinbase (index expressed as company valuation) are deeper with fairer mechanics — though all three work off 1bn assumed shares. Lighter: an unlicensed counterparty, a Stage-0 rollup with a single-operator sequencer and instant upgrades, a 4.5-hour outage on 10 Oct 2025 with about $25m in user losses, loss socialisation in the rules, $100 liability.

Sources: lighter.xyz/terms 2 Sep 2026 · L2BEAT 2 Sep 2026 · The Defiant 10–11.2025.

What tests the view

  • 31 Oct 2026 — has the listing happened (rumoured for October, Nasdaq)? If so: compare the opening price with the perp price the day before — convergence test: gap ≤5% (Cerebras and SpaceX precedents). A bigger gap means the "perp predicts the IPO" class loses its only advantage.
  • 31 Mar 2027 — call: Anthropic is public by this date. If not, perps stay in "no anchor" mode, where every documented blow-up in the class happened.
  • Until listing — an external pre-IPO price oracle appearing on Lighter would overturn finding #1.

Unknowns

Obtainable: the venue pool's share of the ANTHROPIC book on Lighter; Coinbase's primary contract spec (the page returned 403); open interest — $2.19m in the Lighter UI versus $8.6m in press reports in June (probably different venues or dates; not reconciled). Long funding was about 3.5% annualised per the UI, and floats.

Unknowable from outside: Anthropic's real margins, burn and cap table — estimates until a public S-1; what happens to the perp if the IPO doesn't happen — the Lighter rules don't cover it; the real share count — "1bn" is an assumption at every venue, and the FTX precedent shows the divisor can change at settlement.

Who it suits

The Lighter perp combines three weaknesses: a self-referential price on a thin book, a 100% premium to the last real round (the rumour upside is already paid — about 4% is left to the target), and a counterparty with $100 of liability and the right to rewrite settlement on a day's notice. Anthropic as an asset is a separate question: if the IPO happens, buying listed shares with shareholder rights is a normal route. A typical frame for a single AI lab is a small slice of a portfolio (a common ceiling is up to 2%), stated run-rate discounted 30–40%, and attention to how much revenue flows through cloud investors. Anyone who still wants to pay now for the conviction that the IPO will land well above $2tn would at least do it on a deeper venue, in a size they are prepared to lose entirely.

Sources are listed next to each section above.

Research for information only. Not investment advice.