The following is FutureSearch's forecast for Anthropic's financial and technological path through May 2027, first published May 28, 2026 and re-run on October 26, 2026, the day Anthropic filed its S-1 publicly. My central case has ARR reaching $115B by May 2027 (p10 $85B, p90 $162B), with Claude Code contributing about $18.5B, and the IPO pricing around November 24, 2026.
| Anthropic | Latest |
|---|---|
| Public S-1 | Filed October 26 to list on Nasdaq as ANT; the price range and offering size are blank (SEC) |
| Second-quarter 2026 revenue | $11.6B, with $974M of GAAP operating income, its first operating profit (S-1) |
| Third-quarter 2026 revenue | $16.8B, up about 45% from the second quarter, and a second straight quarter of adjusted profit (CNBC). The S-1 states it only as 86 times fourth-quarter 2024 revenue, about $16.9B |
| Customers | More than 11,900 paying $100,000 or more a year, up from 950 a year earlier, and net revenue retention of about 900% (S-1) |
| Adjusted operating margin, revenue per compute dollar | 11% and $1.70 in the third quarter (Bloomberg via Euronews) |
| Run rate | $65B at the end of July (Reuters); the S-1 leaves its October 31 figure blank |
| Last private round | $65B Series H at $965B post-money, May 28 (Anthropic) |
| IPO timing | A Thanksgiving-week debut is reported (Fortune). Details and odds: IPO dates and valuations |
| IPO valuation talk | About $1.5T at the October 14 investor day, down from about $2T floated earlier (Reuters) |
At the $1.5 to $2 trillion discussed for the IPO, Anthropic would list at roughly 22 to 30 times its third-quarter revenue annualized, and 13 to 17 times my median May 2027 ARR. My median for the 90-day post-IPO market cap is $1.65 trillion.
Third-quarter revenue reached $16.8 billion, up about 45% from the second quarter, with a second straight quarter of adjusted profit (CNBC). The S-1, filed October 26, has the audited history behind it. Revenue was $4.6 billion in 2025, then $4.7 billion in the first quarter of 2026 and $11.6 billion in the second, when GAAP operating income reached $974 million, Anthropic's first quarterly operating profit (S-1). The run rate reached $65 billion at the end of July (Reuters). The S-1 defines run-rate revenue as 28 days of usage fees times 13 plus a month of subscription fees times 12, and leaves its October 31 figure blank.
The re-run p50 of $115B by May 2027 (p10 $85B, p90 $162B) is up from $112B on October 16, with most of the change in the upper tail. The S-1 pulls both ways. Sequential growth slowed from more than doubling in the second quarter to about 45% in the third, about 5% to 7% of this year's revenue came from customers later found violating usage policy, whom Anthropic is now cutting off, and the company says it expects its growth rate "to decline over time." Against that, Anthropic says it has operated compute-constrained, and new capacity is arriving fast, with SpaceXAI GPUs costing about $3.1 billion a month once fully delivered in December, against $5.6 billion of total compute spending in the second quarter. The structural drivers are unchanged: enterprise concentration (more than 1,900 customers spending $1M+ a year and about 250 above $10M, per the S-1), the coding moat (54% of coding-specific LLM spend per Menlo), and cross-cloud distribution no competitor has matched.
Claude Code is the live ammunition. It went from $500M ARR in September 2025 to $8B in May 2026, and trackers put it near $15B by August. The S-1 gives no Claude Code figure. The nearest it comes is that the top 5% of Claude Enterprise users who use Claude Code spend about 20 times the under-$75 monthly average of Enterprise users who don't, and that subscription revenue, which covers the Pro, Max, Team and Enterprise plans that bundle Claude Code, grew from $180 million in the first half of 2025 to $3.2 billion in the first half of 2026. The p50 of $18.5B by May 2027 (p10 $9.8B, p90 $33B), up from $17.5B on October 16, assumes the doubling-every-six-weeks cadence keeps cooling toward the single-digit monthly growth it showed in late summer. The case for the right tail is that agentic workloads consume 10 to 100 times more compute per developer-day than chat, and Anthropic owns the coding category. The Claude 5 family's launch theme reinforces it, with the flagship tier built around orchestrating hundreds of parallel subagents and running codebase-scale migrations from kickoff to merge.
Cursor is the live counter, now inside SpaceX after an all-stock acquisition that closed August 14 (SEC filing). It launched its own model in November 2025 and is now near $4B ARR with a 60% enterprise mix (MLQ), double where it sat at launch. The disintermediation risk is real, and I no longer wave it off.
The biggest tail risk I carried on the revenue line was the gross-revenue accounting question, and the S-1 largely settles it. Anthropic recognizes sales through the AWS, Google Cloud and Azure marketplaces, 48% of first-half revenue, on a gross basis as principal, so full customer spend hits revenue and the cloud providers' fees hit expenses. That policy, and the 2023 to 2025 revenue figures, are unchanged from the first confidential draft in June through five amended drafts (S-1). A forced restatement to net, which would have cut headline ARR by 20 to 40% in one print, was a single-digit risk in my model and is now a remote one.
Anthropic filed its S-1 publicly on October 26, after a confidential draft on June 1 and five amendments. From here the rules set the calendar. The roadshow can start 15 days after the public filing, so November 10 at the earliest, and the June 30 financials go stale after November 12, so an amended filing with third-quarter numbers comes first. Among 22 large US IPOs since 2019, the median gap from public filing to pricing was 27.5 days, which would land on November 22; SpaceX took 22 days in June. Fortune reports a Thanksgiving-week target (Fortune), after Bloomberg had put formal marketing as soon as the week of November 9 (Bloomberg).
The re-run distribution lands at p10 November 18, 2026, p25 November 20, p50 November 24, p75 December 15, and p90 February 15, 2027. The median did not move. What the filing changed is the tail. On October 16 a quarter of the distribution sat after February 9, 2027, much of it the risk that Anthropic would not file publicly in time. Now a late date needs a postponement after a public filing, which among those 22 IPOs happened three times, to Klarna, StubHub and Cerebras.
Valuation is the more important number, and the cleanest comp has now moved twice. In June, SpaceX had priced at $1.77T and traded above $2T, the exuberance case for mega-cap listings. It fell to about $1.4 trillion after its first earnings report repriced the AI capital spending underneath it, then recovered to roughly $1.9 trillion by mid-October (see our updated SpaceX valuation). Both halves of that round trip matter: public investors have now watched the year's biggest IPO surrender its scarcity premium and earn much of it back. My re-run median for Anthropic's 90-day post-IPO market cap is $1.65 trillion (p10 $1.05T, p90 $2.51T), the same median as October 16 with a wider range. The S-1's lock-ups add supply inside that window, since some holders can sell up to 6% of their locked shares on each of days 27 and 48 after the prospectus and 7% on each of days 62 and 76. Valuation talk among prospective IPO investors slipped from about $2 trillion to about $1.5 trillion in October (Reuters). At $1.65T against $115B of forward ARR, that is about 14x forward revenue.
The 20x weekly Sonnet hour cap is the most-watched consumer proxy for whether Anthropic has compute headroom. There is a measurement wrinkle the re-run surfaced: Anthropic's current plan documentation leans on relative multipliers and dynamic counters rather than published hour-denominated caps (Claude Help Center), so the resolved number may key off a conservative republished figure. The forecast sits at a p50 of 490 hours (p10 260, p90 590), down from 545 in August.
Pricing tells the sharper story. The Mythos-class tier shipped at $50 per million output tokens ($10 input), double the $25 Opus anchor that held across four consecutive generations, and Fable 5.1 still lists at $50 today. My re-run forecast for the top GA model's output price in May 2027 keeps the p50 at $50 with a p90 of $65, while the lower quartile (p25 $25) covers the worlds where the cheaper Opus line, now Opus 5.5 at $20 since September 22, carries the top GA slot or competitive pressure forces the flagship price down. Even at $50 Anthropic is pricing on capability rather than rationing demand through price.
Capacity is arriving. Anthropic announced fresh commitments alongside the Series H: AWS expansion of up to 5GW, Google and Broadcom for 5GW of TPUs, and SpaceX access to Colossus 1 and 2. The S-1 puts prices on it, with commitments of at least $44.9B to Microsoft, $114.5B to Google and $110B to Amazon, about $161B of Broadcom TPU leases, and SpaceXAI capacity at about $3.1B a month once it is all delivered in December (S-1).
On June 9, Anthropic brought Mythos-class to general availability as Fable 5 and Mythos 5, the first public tier above the Opus line. Fable 5 posts 95.0% on SWE-bench Verified and 80.3% on the harder SWE-bench Pro (Axios).
SWE-bench Verified is saturated, so the live frontier metric is now SWE-bench Pro. My re-run forecast for the top GA model's Pro score by May 2027 is a p50 of 91.8% (p10 86%, p90 97.2%), up from 84.3% in August. Opus 5.5, released September 22, reports 89.9% in its system card (Anthropic).
Commerce lifted the June 12 export order on June 30 and Fable 5 and Mythos 5 returned worldwide on July 1. The full grading lives on the Fable forecast page. What remains forward-looking is durability, and the re-run puts the probability that a Mythos-class model is available and unsuspended on May 21, 2027 at 88%, down from 90% in August, with the residual 12% covering a renewed order or a licensing regime that re-gates the tier.
The Claude 5 family system cards formally assessed Fable 5, Mythos 5, Sonnet 5, and Opus 5 as below the CB-2, Cyber Tier 2, and automated AI R&D thresholds, while describing Mythos 5 as "near the border" on CB-2 (Anthropic), and the Opus 5.5 card did the same. I now put the probability of an ASL-4-equivalent determination by May 2027 at 48%, up from 21%.
The Department of War designated Anthropic a supply-chain risk on March 3-4 under 10 U.S.C. § 3252 and FASCSA. The litigation runs on two tracks. On August 27 the Northern District of California vacated and permanently enjoined the § 3252 designation. On September 25 the D.C. Circuit upheld the parallel FASCSA designation 2-1, and Anthropic asked for rehearing on October 13 (CNBC). I now put the supply-chain designation vacated by May 2027 at 30%, down from 46%.
The contract question went the other way. The $200M CDAO agreement that anchored the bull case for federal revenue was cancelled in early 2026 amid the dispute over "all lawful use" clauses (Wikipedia), so a qualifying award now needs a fresh prime contract with a publicly confirmed $100M+ value while the procurement ban is still being enforced. The NSA relationship is real (Axios; FT), but classified awards rarely surface a public figure. I put a $100M+ DoD or IC prime contract by May 2027 at 2%, down from 11%.
None of this changes the revenue base by more than a few percent. Anthropic is on pace for $115B ARR without federal procurement, where a $100M contract rounds to noise. What the federal track sets is the disclosure surface and the tail risk, not the trajectory.
The three scenarios come out at decoupling with friction 56%, convergence 30%, and entanglement 14%, re-derived from the August re-runs. The restoration moved weight out of the bear case.
Decoupling with friction is the base case, and it is what the summer actually looked like. Commercial and intelligence-community trajectories proceed, punctuated by federal disruptions, with the nineteen-day Fable suspension as the template. ARR lands $98.5-136B, the Mythos-class tier stays available under a compliance regime, and the IPO prices between November and December 2026.
Convergence is the bull case. Federal détente meets a capability bull: the designation vacated (now a 30% marginal on its own), an IC prime contract surfaces despite the odds, and the IPO prices clean above $1.3T.
Entanglement is the bear case, and it is lighter than it was in June. Federal friction institutionalizes, a fresh order or licensing regime re-gates the Mythos tier, the IPO slips past mid-2027, and accounting or litigation overhang bites.
Five predictions, updated against the re-runs. The second has already resolved.
Anthropic prices its IPO above $1T between November 2026 and March 2027. The S-1 is public, press reports point at Thanksgiving week, and the Series H mutual-fund anchor is the price floor. Probability: 89%, with the miss risk split between a slip past March and a first-day value under $1T.
A Mythos-class model is back in general availability before the end of 2026. Resolved YES on July 1, 2026, six months early. The June probability was 80%.
The top GA model holds its $50 output price rather than reverting to the $25-30 Opus anchor. The Mythos-class tier set the new ceiling, and Anthropic prices on capability, though Opus 5.5's $20 price is a live path back down. Probability: 50%.
The 90-day post-IPO market cap clears $1T. The Series H floor and frontier-AI appetite align, with the deflated SpaceX comp as the new caution. Probability: 91%.
No $100M+ DoD or IC prime contract surfaces publicly before May 2027. The $200M agreement is cancelled, the procurement ban is being enforced, and classified awards stay opaque. Probability: 98%.
Pulling it together: the commercial case the public-market skeptics underrated is intact and compounding faster than my June model had it, with ARR tracking toward $115B, Claude Code holding its trajectory through the suspension, and a full Claude 5 family shipped from Sonnet to Mythos. June's lesson stands in softened form. The government showed it can switch the flagship off, then showed it can switch it back on under pressure, and both facts now live in the S-1's risk factors rather than in the central numbers.
Changelog
-
May 28, 2026 — First published, on the Series H close and the $47B run-rate disclosure.
-
June 18 — The Fable 5 / Mythos 5 launch and export suspension: ARR $93B, IPO December 15 with an April 2028 tail, 90-day cap $1.09T.
-
July 1 — The export order was lifted; noted, with its direction of effect, ahead of the next re-run.
-
August 6 — Re-ran every forecast on the summer tracker baseline: ARR $93B to $115B, the IPO median in to November 30, the 90-day cap to $1.14T, SWE-bench Pro down to 84.3% after Opus 5 shipped below the flagship, and ASL-4 down from 38% to 21%.
-
August 20 — Anthropic's $65B end-July run rate, its first operating profit and reports of an October window: ARR to $140B, the pricing date in to October 26, 2026, and the 90-day cap to $1.85T.
-
August 29 — Held against every other forecast we publish on AI: ARR eased to $138B, the Sonnet-hour cap to 545, and Claude Code held at $23.5B. The IPO date and the 90-day cap were revised too; I did not take those and they stand at their August 20 values.
-
October 16 — Re-ran every forecast after third-quarter revenue of $16.8B, Opus 5.5 and the D.C. Circuit ruling, then held them against every other forecast we publish: ARR $138B to $112B, Claude Code $23.5B to $17.5B, the IPO median to November 24, the 90-day cap to $1.65T, SWE-bench Pro to 91.8%, ASL-4 to 48%, the designation vacated to 30% and a $100M+ contract to 2%.
-
October 26 — Anthropic filed its S-1 publicly. Re-ran the four forecasts here that it bears on: ARR $112B to $115B, Claude Code $17.5B to $18.5B, the IPO median held at November 24 with the p90 in from May 14 to February 15, 2027, and the 90-day cap held at $1.65T with a wider range ($1.05T to $2.51T). The other forecasts stand at their October 16 values.
Run this forecast yourself in the FutureSearch app. Ask it to refresh the numbers any time the news cycle moves.