Anthropic IPO Research: The First Audited Look at Frontier-Lab Economics

TL;DR

Status verified: Anthropic confirmed a confidential draft S-1 on June 1, 2026; as of August 12, 2026 there is no public S-1, no audited financials, no ticker, price range or confirmed date on EDGAR. The last primary valuation is the $65B Series H at $965B post-money (May 28, 2026); secondary markets imply $1.1–1.2 trillion. Reported run-rate revenue crossed $47B in May 2026, and the company projected its first quarterly operating profit (~$559M) on $10.9B Q2 revenue. Reuters reports bankers are underwriting a possible ~$2 trillion valuation off an internal 2028 revenue forecast of ~$190–200B — a level that sits ~$700B above our AVOID threshold (see August Update). Reported run-rate crossed $47B in May 2026, with a first quarterly operating profit (~$559M) on $10.9B Q2 revenue. Treat the profit, secondary marks, and the 2028 figure as reported/unaudited.

Rovenin signal is price-dependent: BUY below ~$700B implied market cap, HOLD $700B–$1.1T, TRIM $1.1T–$1.3T, AVOID above ~$1.3T. At the rumored ~$1T+ IPO, we rate the direct deal a HOLD/PASS for most public investors , extraordinary growth is real, but the multiple already prices in flawless execution against a compute-cost structure that is only just turning positive and a governance model that transfers control away from common shareholders.

Best risk-adjusted exposure today is via proxies, ranked by capital efficiency: SK Telecom (SKM) offers the highest stake-to-market-cap leverage (~20%+ of a ~$14B cap), Zoom (~5–14%) and Salesforce (~3%) are moderate, and Alphabet/Amazon are diluted (~3–4% of a multi-trillion cap). Direct pre-IPO access via SPVs is possible but discouraged by Anthropic itself.

Key Findings

The IPO is real but early-stage. Anthropic's own June 1st newsroom post confirms a confidential Form S-1 under Rule 135. Bloomberg (June 3, 2026) reported Anthropic selected Morgan Stanley, Goldman Sachs and JPMorgan as lead underwriters, with Wilson Sonsini (Google's 2004 IPO counsel) as legal counsel, targeting an October 2026 Nasdaq listing; none of these are confirmed by Anthropic. The raise is reported at "more than $60B," which would rank among the largest IPOs ever. No ticker or price range exists. (AnthropicBitMEX)

Valuation has repriced ~16x in 14 months — Series E ~$61.5B (Mar 2025) → Series F $183B (Sep 2025) → ~$350B (Nov 2025 Microsoft/Nvidia round) → Series G $380B (Feb 2026) → Series H $965B (May 2026), with secondary marks at $1.1–1.2T. This is the single fastest large-cap private repricing on record and a core bear signal. (Seeking Alpha)

Revenue trajectory is historically unprecedented but definitionally slippery. Run-rate (last-month-annualized) went ~$1B (Dec 2024) → ~$9B (end-2025) → $14B (Feb) → $30B (Apr) → $47B (May 2026). But trailing revenue was only $4.8B in Q1 and a projected $10.9B in Q2 — meaning calendar-2026 revenue likely lands ~$20–26B, not $47B. Investors must not conflate run-rate with realized revenue. (VentureBeatSaaStrAI)

The margin turn is the whole thesis. Per WSJ (May 20, 2026), confirmed by CNBC/Bloomberg, compute cost per revenue dollar fell from 71¢ (Q1) to a projected 56¢ (Q2), producing the first ~$559M operating profit and reversing prior guidance of no full-year profit before 2028. This is remarkable versus OpenAI's structural losses, but it is a single unaudited quarter, revenue is booked gross (inflating the top line), and compute obligations are enormous.

Compute footprint is both moat and dependency. Commitments span Amazon (>$100B over 10 years, up to 5GW Trainium, >1M Trainium2 chips, Project Rainier), Google+Broadcom (~3.5GW TPU from 2027), Microsoft/Nvidia ($30B Azure, up to 1GW, $15B combined investment), and SpaceX/Colossus ($1.25B/month, ~$45B total through 2029, ~220,000 GPUs/300MW). Multi-cloud neutrality is a genuine edge; the SpaceX and hyperscaler bills are a genuine risk. (Anthropic)

Enterprise-first is the differentiator. ~80–85% of revenue is enterprise/developer (vs. OpenAI's ~85% consumer). Per Menlo Ventures' "2025: The State of Generative AI in the Enterprise," Anthropic now earns 40% of enterprise LLM spend, up from 24% in 2024 and 12% in 2023, while OpenAI fell to 27% from 50% in 2023 (Google reached 21%). In coding, Menlo estimates Claude commands ~54% market share vs. 21% for OpenAI (up from 42% six months earlier), driven by Claude Code, which reached ~$8B annualized run-rate by May 2026 (up from $2.5B in February). 8 of the Fortune 10 and >1,000 customers spend >$1M/yr; >300,000 business customers; >100,000 on Bedrock. (Forbes)

Governance is a double-edged sword. Delaware PBC + Long-Term Benefit Trust holding Class T stock that elects a growing (now majority) share of the board. Mission-aligned and safety-credible, but a structural governance discount for public common holders who have limited control.

National-security conflict is a live, material risk. The Pentagon designated Anthropic a "supply-chain risk" (Feb/Mar 2026) after Anthropic refused unrestricted ("any lawful use") access over autonomous-weapons and mass-surveillance concerns; President Trump ordered federal agencies to cease use. Litigation is ongoing. This both burnishes the safety brand and closes a large government market. (CNBC)

Details

1. IPO Mechanics & Snapshot (what is knowable vs. undisclosed)

Confirmed by Anthropic: On June 1, 2026, "Anthropic, PBC confidentially submitted a draft registration statement on Form S-1 to the U.S. Securities and Exchange Commission for a proposed initial public offering of our common stock," under Rule 135, with share count and price "not yet set." As of August 12, 2026, no public S-1/A appears on EDGAR, so no audited financials are yet public. (Anthropic)

Reported but unconfirmed by Anthropic (label rumored/analytical):

Underwriters: Morgan Stanley, Goldman Sachs, JPMorgan (Bloomberg, June 3, 2026). Counsel: Wilson Sonsini (adviser since 2022, per FT/Reuters). (TechCrunch)

Exchange/ticker: Nasdaq reported; no ticker disclosed. (TradingView's "Oct 23, 2026" date and a "Jane Street acquired Anthropic" line are data-provider artifacts, not fact.)

Timing: October 2026 window widely cited; CNBC (July 15, 2026) confirmed bankers are lining up investor meetings, indicating preparations are advancing but no roadshow date. (CNBC)

Raise size: ">$60B" reported. An earlier-2026 report cited a hypothetical $400–500B IPO valuation target — now clearly stale given the $965B round.

Use of proceeds / share class / lockups / insider selling: Not yet disclosed (awaits public S-1).

2. Valuation History & Private Financing Timeline

Anthopic's Evaluation Snapshot

2022: Series B $580M (incl. FTX/Alameda ~$500M).

2023: Google up to $2B (initial $300M for ~10%); Amazon begins staged investment; Series C ($450M, May 2023) with Salesforce, Zoom, Google, at ~$4.1B.

Amazon: staged to $8B total, with up to $5B more added in the April 2026 AWS expansion (up to ~$25B potential).

Series E: ~$3.5B at $61.5B post (Mar 2025), led by Lightspeed. (Seeking Alpha)

Series F: $13B at $183B (Sep 2025), led by ICONIQ. (Sacra)

Nov 2025: Microsoft (up to $5B) + Nvidia (up to $10B) strategic investment alongside $30B Azure commitment; valuation reported ~$350B.

Series G: $30B at $380B (Feb 12, 2026), led by GIC & Coatue; concurrent $5–6B employee tender at ~$350B. (SacraSacra)

Series H: Per Anthropic's May 28, 2026 release, "$65 billion in Series H funding led by Altimeter Capital, Dragoneer, Greenoaks, and Sequoia Capital, valuing the company at $965 billion post-money," with Capital Group, Coatue, D1, GIC, ICONIQ and XN co-leading and strategic infra partners Samsung, SK hynix and Micron participating. The round included ~$15B of previously committed hyperscaler investment ($5B from Amazon) and was widely described as the "last private round."

Reconciliation of conflicting reports: The FT-reported ~$900B round earlier in 2026 was the in-progress figure for what closed as the $965B Series H, the same round, not two separate events.

Secondary markets: Forge ~$1T (May 2026), Caplight ~$1.2T (a reported +550% YoY), with willingness-to-sell quotes as high as $1.15T. These are illiquid minority marks, not primary valuations. (The Next Web)

3. Financial Profile & Unit Economics

Revenue: Run-rate crossed $47B (May 2026), per Anthropic's Series H release. But trailing quarterly revenue was $4.8B (Q1 2026) and a projected $10.9B (Q2 2026) (WSJ/CNBC, reviewed investor projections). SaaStr estimates calendar-2026 revenue lands ~$20–26B. Revenue is booked gross, sales through AWS/Google/Microsoft count full end-customer spend with the partner cut as a cost, inflating the top line vs. a net-reporting SaaS peer. (Anthropic)

Margins & profitability: Q1 compute-to-revenue ratio 71¢; Q2 projected 56¢, yielding the first ~$559M operating profit (WSJ, May 20, 2026). Anthropic's internal model reportedly targets ~77% gross margin and ~$17B positive cash flow on ~$70B revenue by 2028. Skeptics (e.g., Ed Zitron) argue the "profitability" is a run-rate/ARR artifact and that gross reporting overstates scale. We treat the Q2 profit as reported/unaudited and a genuine but fragile inflection. (Perplexityaimagazine)

Customer metrics: >1,000 customers >$1M/yr (doubled from 500+ in <2 months); >300,000 business customers; >100,000 on Bedrock; 8 of Fortune 10; ~70% of Fortune 100. Enterprise mix ~80–85%. (Anthropic)

Compute intensity is the central margin question: Anthropic's aggregate compute commitments run to the hundreds of billions across AWS, Google/Broadcom, Microsoft/Nvidia and SpaceX. Revenue per GW is improving as inference efficiency rises, but the SpaceX deal alone ($15B/yr) consumed roughly half of April run-rate revenue at signing, a ratio without recent precedent.

4. 2028: hype or real? and why run-rate ≠ revenue

Run-rate is not revenue Annualized run-rate vs. actual booked revenue vs. the 2028 internal projection. Rovenin.

Three numbers are circulating, and they are not interchangeable:

~$47B "run-rate": last month's revenue × 12. An exit-pace snapshot, not money booked.

~$20–26B actual 2026 revenue (our estimate): what Anthropic likely books this calendar year. Run-rate captures only the fastest recent month, so in a hypergrowth curve it sits well above trailing revenue.

~$190–200B 2028 projection: internal, unaudited; ~4x the current run-rate and ~8–10x actual 2026 revenue.

Our take, "real or hype" is the wrong frame; it's conditional. The near-term run-rate is real and checkable. The 2028 figure is a forecast that needs three things to all hold: demand keeps compounding with no competitive air-pocket; compute cost/revenue keeps falling (the 71¢→56¢ trend continues); and open-weight/Chinese models don't commoditize the mid-market. Any one slipping breaks the multiple. Treat $190–200B as a bull-case input, not a base case, and never pay a price today that only works if 2028 lands. The audited S-1's booked revenue and gross margin will show how much of the run-rate is durable recurring revenue vs. burst adoption.

5. Compute & Capex Footprint

Amazon/AWS: Per Anthropic's April 2026 announcement, ">$100 billion over the next ten years to AWS technologies, securing up to 5GW of new capacity… over one million Trainium2 chips to train and serve Claude." Project Rainier (Indiana, ~$11B site, ~500K Trainium2 at launch, now >1M chips); Amazon added $5B (up to ~$25B potential total). AWS remains "primary cloud and training partner," spanning Trainium2 through Trainium4. (Data Center Dynamics)

Google + Broadcom: October 2025 deal for up to 1M TPUs / >1GW; April 2026 expansion to ~3.5GW of next-gen TPU from 2027; Broadcom 8-K confirms long-term Google TPU roadmap through 2031; ~$21B in Broadcom orders reported (Mizuho estimate). (Silicon Republic)

Microsoft + Nvidia: $30B Azure commitment + up to 1GW (Grace Blackwell/Vera Rubin); Nvidia up to $10B and Microsoft up to $5B investment; Claude on Azure Foundry, the only frontier model on all three hyperscalers. (anthropic)

SpaceX/Colossus: $1.25B/month through May 2029 (~$45B total), full Colossus 1 (~220,000 GPUs, >300MW) in Memphis, primarily inference; 90-day termination clause; concentration risk given the facility is xAI-linked. (Crypto Briefing)

Fluidstack: part of a $50B US infrastructure commitment (Nov 2025).

Meta: reportedly in talks to lease up to $10B of compute to Anthropic over ~two years (July 2026).

Implication: The multi-vendor strategy reduces single-supplier lock-in and lets Anthropic match workloads to the cheapest suitable silicon, a real structural advantage. But the take-or-pay nature of these commitments means margins are hostage to utilization; a demand air-pocket would be brutal.

6. Product Mix & Monetization (segment mix not officially disclosed, estimates labeled)

API/Claude platform: Opus 4.x / Sonnet 4.5 / Haiku 4.5 token pricing (Opus ~$5/$25 per M in/out; Sonnet cheaper); up to 1M-token context; Claude Code (~$8B run-rate, ~54% coding share per Menlo).

Cloud channel: Bedrock (>100K customers), Vertex AI, Azure Foundry.

Enterprise workflow: Claude, Claude Code, Claude Cowork, Claude in Chrome/Excel/PowerPoint, Claude Security/Mythos.

Consumer: Free/Pro/Max subscriptions (~20% of revenue; Claude ~30M weekly users vs ChatGPT ~800M–900M).

Analytical mix estimate (Rovenin): ~70–80% API + enterprise, ~20% consumer, small but fast-growing services/strategic.

7. Moat & Competitive Landscape

Anthropic's moat rests on three pillars: (1) safety/constitutional-AI brand that de-risks enterprise procurement (non-training on customer data; ~70% of head-to-head enterprise wins per Ramp data); (2) enterprise-first product depth, especially coding (Claude Code) and MCP (the open standard it created, now adopted by OpenAI, Google, Microsoft); (3) cross-cloud neutrality (only frontier model on all three hyperscalers).

Competitive read:

OpenAI (key comp): ~800M–900M weekly users, stronger consumer/brand, but ~85% consumer-mix, structural losses (~$7B Q1 2026 loss reported), and enterprise share falling from ~50% (2023) to ~27% (2026). Last raised a $122B round in March 2026 at an $852B post-money valuation on ~$24B annualized revenue (~35x). Anthropic leads enterprise/coding; OpenAI leads scale/consumer. (Axis Intelligence)

Google/Gemini: ~21% enterprise share, Gemini 3 competitive on benchmarks/price, deep pockets and TPU cost advantage, and an Anthropic investor.

Meta/Llama: open-weight pricing pressure, but only ~9–13% enterprise share and declining.

xAI/Grok: compute-rich (Colossus) but consumer-troubled; ironically now Anthropic's landlord.

Chinese models (DeepSeek, Qwen): the real long-run pricing threat, dramatically cheaper, closing the quality gap, pressuring API margins on commodity workloads.

Durability verdict: Model-quality leads are compressing across the board; Anthropic's durable edge is the enterprise trust + coding workflow + neutrality bundle, not raw model IQ. Pricing power is real in coding/regulated verticals, thinner in commodity chat.

Claude compared to different LLMs. Image credit: Coatue Management

Claude is spanning across enterprise work. Credit: Coatue

Github PRs since Opus 4.5 Launch. Credit: Coatue

8. Acquisitions & Strategic M&A

Acquisitions (4 in ~6 months): Bun (JS runtime, Dec 2025), Vercept (computer-use, Feb 2026), Coefficient Bio (biotech, Apr 2026), Stainless (SDK/MCP tooling, >$300M, May 2026, strategically cuts off tooling used by OpenAI/Google). Pattern: buying the developer-tooling layer beneath rivals.

Enterprise AI services company (May 4, 2026): with Blackstone, Hellman & Friedman, Goldman Sachs, $1.5B combined ($300M each from Anthropic/Blackstone/H&F, $150M Goldman); backed also by General Atlantic, Apollo, GIC, Sequoia, Leonard Green; acquired Fractional AI as its delivery core.

KPMG global alliance (May 19, 2026): Claude embedded across 276,000 employees + Digital Gateway; KPMG preferred PE consultant. PwC struck a parallel deal.

Gates Foundation: ~$200M partnership (health/development).

Project Glasswing (Apr 7, 2026): $104M cyber-defense coalition (AWS, Apple, Broadcom, Cisco, CrowdStrike, Google, JPMorganChase, Linux Foundation, Microsoft, Nvidia, Palo Alto Networks) using Claude Mythos; expanding to ~200 partners; >10,000 high/critical vulnerabilities found.

Defense/national security: $200M DoD prototype agreement (July 2025); Claude Gov; then the supply-chain-risk designation dispute (see Risks).

9. Similarities with the SpaceX IPO, and what it signals

SpaceX went public June 12, 2026, the largest IPO ever (~$75B raised, ~$1.8T valuation), spiked past $2 trillion intraday, then sold off hard within days. The parallels are close: both filed confidentially within weeks of each other, both are capital-intensive "own-the-future" stories, both are scarcity-driven trophy assets.

The differences favor SpaceX on durability: a 24-year history, a near-monopoly in launch, and a real subscription base (Starlink ~58% of revenue); Anthropic is four years old, effectively single-product, and fights OpenAI, Google, Meta and cheaper Chinese models daily. SpaceX also gave ~30% of the deal to retail; Anthropic's split is unknown but likely more institutional.

What it signals: appetite for trophy AI/tech names is enormous but fragile. SpaceX is the live cautionary tale, even a hugely oversubscribed mega-IPO can pop and then correct sharply once scarcity fades. Expect the same volatility for Anthropic; the first-day pop is not the price that matters.

Two trophy mega-IPOs SpaceX (public June 2026) vs. Anthropic (rumored): a side-by-side on the same question. Rovenin.

10. Walmart vs. Anthropic: the scale reality check

At ~$2T, Anthropic would be worth ~2x Walmart (~$900B market cap) on a fraction of the revenue. Walmart booked ~$713B in FY26 revenue and ~$22B net income, serving 280M customers weekly with 2.1M staff. Anthropic's ~$47B run-rate is ~7% of Walmart's sales; its ~$20–26B actual revenue is ~3%; even the $190–200B 2028 forecast is ~27% of what Walmart sells today.

Our take: not a knock on quality, software scales as retail never can, and Anthropic's growth and margins live in a different universe. But it clarifies what a $2T buyer pays for: the entire value sits in future revenue and margins that don't exist yet. Walmart is the anchor to remember when the run-rate charts start to feel like gravity is optional.

A ~$2 trillion company with ~7% of Walmart's sales Walmart vs. Anthropic, revenue (left) and market value (right). Rovenin.

11. The same dollars, going in a circle?

Much of Anthropic's compute spend flows back to its own investors. Nvidia (up to $10B), Microsoft (up to $5B), Amazon ($8B→$25B) and Google (~$3B / ~14%) put equity in, and Anthropic commits >$150B back to those same firms for chips and cloud ($100B+ AWS, $30B Azure, ~$21B Google/Broadcom TPU, plus Nvidia GPUs throughout). Critics hear an echo of dot-com "vendor financing," where a supplier funds its own customer and both sides' numbers flatter the real economics.

Our take, real concern, not a closed loop. Outbound commitments dwarf the equity coming in (>$150B committed vs. ~$25–45B invested), so net cash genuinely leaves Anthropic and lands on partners' P&Ls, this isn't pure round-tripping, and Anthropic has large third-party enterprise/coding revenue unrelated to its investors. The subtler risk: it inflates the apparent health of the whole AI-infrastructure complex, and a demand slowdown would hit investment marks, chip orders and cloud revenue at once, correlated, not diversified. Read the S-1's related-party disclosures closely.

The same dollars, going in a circle? Equity in from chipmakers/clouds vs. compute Anthropic commits back to them. Rovenin.

12. . Governance & Ownership

Anthropic is a Delaware PBC ("responsibly develop and maintain advanced AI for the long-term benefit of humanity"). The Long-Term Benefit Trust (LTBT), five financially disinterested trustees holding Class T stock, elects a growing share of the board, reaching a majority within ~4 years of establishment. Board includes Dario Amodei (CEO), Daniela Amodei (President), Yasmin Razavi, Reed Hastings, Chris Liddell (ex-Microsoft CFO, Feb 2026), and Vas Narasimhan (Novartis CEO, LTBT-appointed, Apr 2026). Google holds ~14%; Amazon second-largest. (harvard)

Klover.ai analysis, critical assessment: The Klover piece correctly identifies the core structural fact, Class T carries governance rights but no economic value, so public Class A holders get economic participation while sitting below founders (Class B) and the LTBT (Class T) in control. Its claim that ESG/mission-aligned capital may reward the structure is plausible but speculative; the more reliable market read (echoed by the Snap/Lyft dual-class precedent) is that public markets apply a governance discount to trust/founder-controlled entities. Klover's framing that "the discount is relocated" rather than eliminated is a fair characterization. Rovenin view: expect a mid-single-digit to low-double-digit governance discount vs. a hypothetical one-share-one-vote structure.

13. Risk Framework

Compute concentration/dependency: hundreds of billions in take-or-pay commitments; SpaceX/xAI facility risk; hyperscaler-as-investor-and-supplier circularity.

Margin compression: the 56¢ compute ratio must keep falling; Chinese/open-weight pricing pressure on commodity tokens.

Governance complexity: LTBT majority control; limited common-holder power.

Policy/national-security conflict: Pentagon supply-chain-risk designation; federal-use ban; ongoing litigation; the June 2026 export-control episode forcing suspension of certain models for foreign nationals. (SmartAsset)

Reliability/capacity bottlenecks: repeated Claude Code capacity strain in 2025–26.

Safety-policy drift: RSP changes and external criticism cut both ways with the safety brand.

Key-person risk: heavy dependence on Dario and Daniela Amodei.

AI-bubble/macro: frontier valuations, data-center/energy constraints, community pushback, circular financing concerns raised by Wall Street.

14. AI-Space Macro Context

Anthropic's S-1 will be the first audited look at frontier-lab unit economics, potentially a repricing event for the entire private AI complex (OpenAI ~$852B–$1T+ on ~$24B ARR; SpaceX already public via SPCX). The bull case says durable fundamentals (real enterprise revenue, a genuine margin turn) distinguish this from 1999; the bear case notes the same circular vendor-investor financing, energy/data-center constraints, and reflexive secondary marks that characterize bubbles. Rovenin's read: this is neither pure 1999 nor risk-free, it is a genuine platform shift financed on partly circular, capacity-constrained terms, where a handful of names may justify their marks and the long tail will not.

15. Proxy Plays (ranked by capital-allocation efficiency = stake value ÷ market cap)

SK Telecom (SKM), highest leverage. Invested $100M (Aug 2023) at ~$5B (~2%, since diluted to an estimated ~0.3%). At $965B that stake is worth ~$2.9B (~21% of SKT's ~$14B market cap); at ~$1.2T secondary, ~$3.6B (~26%). SKT trades at a high trailing P/E ~53x (depressed 2025 earnings from a security incident/restructuring) but forward P/E ~13–15x; the ~0.24 PEG claim is unverified (only plausible on aggressive forward-EPS-growth assumptions of +283% 2026 EPS growth). AI data-center revenue grew +89.3% YoY to ₩131.4B (Q1) and +92.5% to ₩136.2B (Q2 2026); SKT fully consolidated SK Broadband (to 100%, completed May 29, 2026, cash-based so no dilution) and formed "SK Hyper" (July 2026) for a 15GW data-center build. Dividend ~4.3% local (reinstated ₩830/quarter in 2026 after a 2025 suspension). Analyst targets range from JPMorgan Neutral (₩80,000, "overheated") to Korea Investment Buy (₩121,000); consensus "Hold." SKT participated in the Series H (undisclosed size) and CEO Jung Jae-heon says it will not sell, so it is a permanent optionality holding booked in OCI, not an IPO-liquidity event. "Korea discount"/SOTP narrative supports upside but may persist. Highest stake-to-cap ratio, but wrapped in a legacy telco with FX and chaebol-governance risk. (Substack)

Zoom (ZM). ~$51M in Series C (May 2023) at ~$4.1B; stake ~0.38%, marked ~$1.27B at $380B and ~$3.8B at $1T, vs. a ~$27B market cap (~5–14% depending on mark). High reflexivity; the stock jumped ~10% on the June 1 stake headline. Meaningful but not company-defining. (Substack)

Salesforce (CRM). Repeated investments since Series C (2023); stake ~$5B (~two-thirds of its strategic portfolio) vs. ~$165B cap (~3%). Also a deep product partner. Rose ~10% on June 1. (Sacra)

Alphabet (GOOGL) & Amazon (AMZN). Largest absolute stakes (Google ~14%; Amazon $8B+ and up to ~$25B potential) but only ~3–4% of multi-trillion caps, diluted exposure. You buy these for the core business, with Anthropic as a free option. (Capacity)

Pre-IPO access routes: secondary markets (Forge, Caplight) and SPVs/access funds exist but are illiquid, accredited-only, opaque, and Anthropic has publicly warned against SPV exposure ("invest at your own risk… assume" caution). For ordinary investors, the public proxies are the cleaner route.

16. Valuation Framework & Scenarios (fresh, current-market)

Anchor points. Series H $965B on ~$47B run-rate ≈ ~20x run-rate (but ~40–48x trailing 2026 revenue of ~$20–26B). Secondary ~$1.1–1.2T ≈ ~23–26x run-rate. OpenAI cross-check: ~$852B on ~$24B ARR ≈ ~35x. So Anthropic is cheaper per unit of revenue than OpenAI and profitable sooner, the relative-value bull argument.

Moonshot drivers modeled explicitly:

(a) Claude Code / coding agents: the proven engine (~$8B run-rate, ~54% share). Base: scales to $25–40B by 2030; bull: coding becomes the default agentic dev platform, $60B+.

(b) Enterprise workflow/agents (Cowork, Office, Security): base $15–30B by 2030; bull: Anthropic becomes the enterprise "agent layer," $50B+.

(c) Healthcare/biology: early (Coefficient Bio, Gates Foundation); optionality, base ~$3–8B, bull $20B+ if Claude becomes core to drug discovery/clinical workflows.

(d) Defense/government: currently impaired by the supply-chain-risk designation; base near-zero to modest, bull (post-resolution) $5–15B.

Scenarios:

Valuation Scenarios

Bear assumes margin compression, Chinese/open-weight repricing, and share loss; base assumes durable enterprise/coding leadership with a holding margin turn; bull assumes Anthropic becomes the dominant enterprise agent platform with healthcare/defense optionality paying off.

Near-term (IPO/early public): at the rumored ~$1T IPO the stock would price at ~40–48x trailing 2026 revenue and ~9x forward 2030 base-case revenue, i.e., you are paying today for the 2030 base case with little margin of safety.

Rovenin Price-Dependent Bands (fresh, anchored to today's environment)

BUY — implied market cap below ~$700B (≈ <15x run-rate / <30x trailing 2026 revenue): growth and the margin turn give a real margin of safety; base/bull skew favorable.

HOLD — ~$700B to ~$1.1T: fair value for the fundamentals; execution-dependent; the likely IPO zone.

TRIM — ~$1.1T to ~$1.3T: priced for near-flawless execution; take gains/lighten.

AVOID — above ~$1.3T: discounts the bull case as base; asymmetric downside if the audited S-1 disappoints or compute costs re-expand.

17. Recommendations

Wait for the public S-1. Do not commit at the rumored IPO price on run-rate math. The first audited financials will reveal trailing revenue, true gross margin (net of gross-booking), the real compute-obligation schedule, dilution, and lockup terms. Benchmark that changes the call: audited gross margin >60% and a second profitable quarter would justify moving up a band; audited gross margin <45% or a Q3 slip back to operating loss would push us toward AVOID at ≥$1T.

For exposure now, prefer proxies in this order of efficiency: SK Telecom for maximum stake-to-cap leverage (accepting telco/FX/governance baggage), then Zoom/Salesforce for cleaner US-listed operating businesses with embedded upside, then Alphabet/Amazon only if you want the core franchise with Anthropic as a free option. Threshold: if SKT's Anthropic-attributable value exceeds ~30% of its market cap purely on secondary marks, the proxy is discounting the IPO — trim.

At IPO: BUY only if it prices below ~$700B implied cap; HOLD/participate modestly in the $700B–$1.1T zone; do not chase above ~$1.3T. Size for volatility — expect a violent first-90-day trading range.

Avoid SPVs/secondary access funds for non-institutional investors — fee-laden, illiquid, opaque, and explicitly discouraged by Anthropic.

Monitor three catalysts: (i) resolution of the Pentagon supply-chain-risk dispute (re-opens a large market), (ii) the audited compute-cost ratio trend, (iii) Chinese/open-weight pricing pressure on API margins.

18. Caveats

Unaudited/unconfirmed figures: run-rate revenue, the $559M Q2 operating profit, secondary valuations, underwriters, exchange, ticker, timing and raise size are all reported/rumored pending the public S-1. Run-rate ≠ trailing revenue; revenue is booked gross.

Conflicting data resolved: the "$900B round" (FT) and "$965B Series H" are the same round at different reporting moments; "$30B (April)" and "$47B (May)" run-rates are both Anthropic disclosures at different dates; SKT ownership (~0.3%) and PEG (~0.24) are estimates, not confirmed.

Data-provider artifacts (e.g., TradingView's "Jane Street acquired Anthropic," fixed Oct 23 date) are errors, not facts.

Forward scenarios are Rovenin analytical estimates, not company guidance, and carry wide error bars given the pace of change and the absence of audited financials.

This is investment research for informational purposes, not personalized investment advice. Treat all figures as of August 12, 2026.

19. How institutions vs. retail are likely to act

Institutions split in two. The insiders, Coatue (co-leading, and, in their words, "tripling down"), Altimeter, Sequoia, GIC, the hyperscalers, are underwriting the bull case and want a high benchmark; a rich mark flatters existing positions. The generalist long-onlys and pensions buying at IPO will be more disciplined, anchoring to the audited S-1, gross margin, and the credibility of the 2028 forecast; many start small and add only after a print or two. Expect heavy oversubscription regardless (scarcity guarantees it), but fast repricing if the numbers disappoint.

Retail will chase. The Anthropic/OpenAI/SpaceX trio is the most-anticipated IPO cluster in years, and, as the ZM/SKM "proxy barbell" chatter shows, retail has already been front-running access via proxies. If Anthropic mirrors SpaceX's retail-friendly allocation, expect a violent day-one pop from demand that can't get shares, then (as with SpaceX) a sharp give-back.

Our guidance for both: the opening price is set by scarcity, not fundamentals. Discipline means waiting for the audited S-1 and post-lockup reality, not buying the open. This is exactly where our price bands earn their keep, and where the proxies (especially SK Telecom's stake-to-cap leverage) remain the more efficient pre-IPO expression.