The editorial argues that 220x trailing revenue and 55x forward revenue multiples are historically reserved for companies with 80%+ gross margins and defensible moats — neither of which Anthropic currently has. With Claude API gross margins in the low 40s and $11.3B in compute costs flowing to Amazon and Google, the pricing assumes a future Anthropic has not yet earned.
The editorial notes revenue grew from $2.1B to $9.1B in a single year, with a diversified mix across API (61%), consumer (24%), and marquee enterprise contracts including the DoD, Deloitte, and three of the four largest US banks. That growth profile, combined with a deliberately thin 7% float, explains why the market pushed the stock above $290 on debut.
The editorial highlights that the lion's share of Anthropic's $11.3B cost of revenue flows back to its two largest strategic investors under compute-for-equity arrangements. This circular capital flow — where cloud providers fund the AI lab that then pays them for compute — complicates any clean read of the company's economics and raises questions about true independent gross margin.
Reuters frames Anthropic's debut as Wall Street's referendum on whether AI's commercial promise can withstand scrutiny at public-market scale. The headline 'AI's promise, peril on Wall St test' signals that the listing is being read as a bellwether for the entire sector rather than a single-company event.
By submitting the Reuters framing to Hacker News and drawing 296 points and 415 comments, the submitter amplified the view that this debut matters beyond Anthropic itself as a sector-defining moment. The strong engagement suggests the developer community treats this as a signal about AI's broader commercial standing.
Anthropic rang the Nasdaq bell this morning under the ticker ANTH, closing its first day at a fully-diluted market cap of $2.03 trillion. The offering priced at $215, opened at $268, and spent most of the session north of $290 before settling. It is now the largest technology IPO in history by first-day market cap, eclipsing Saudi Aramco's 2019 debut in nominal terms and dwarfing Alibaba's 2014 listing by more than 6x.
The raise itself was comparatively modest — $14.2B in primary proceeds against a float of roughly 7% of shares outstanding — which is the usual trick for keeping scarcity high on day one. Early investors (Google, Amazon, Spark, Lightspeed, Menlo, and the Series E cohort led by ICONIQ) are locked up for 180 days. Dario Amodei retains a dual-class voting structure that gives founders and the Long-Term Benefit Trust combined control well past the public float threshold.
The S-1, filed six weeks ago and amended twice, discloses $9.1B in trailing-twelve-month revenue, up from $2.1B a year prior. Claude API revenue accounts for roughly 61% of the top line, with Claude.ai consumer subscriptions at 24% and enterprise contracts (including the headline deals with Deloitte, the US Department of Defense, and three of the four largest US banks) making up the rest. Net loss was $4.8B on the year, driven almost entirely by compute spend — an eye-watering $11.3B in cost of revenue, of which the lion's share is paid to Amazon and Google under the much-discussed compute-for-equity arrangements.
Two things got priced today, and they are worth separating.
The first is Anthropic itself. At $2T, the market is paying roughly 220x trailing revenue and about 55x forward revenue if you accept the sell-side consensus of ~$36B next year. That is a multiple reserved, historically, for companies with 80%+ gross margins and a credible moat. Anthropic has neither — not yet. Gross margin on Claude API is somewhere in the low 40s depending on how you allocate training compute, and the moat is a function of model quality that competitors catch up to on a predictable cadence. The S-1 is unusually candid about this: the risk factors section explicitly names "compression of frontier model lead time from 12 months to under 6 months" as a material risk.
The second thing priced is the entire AI capex thesis. Anthropic's listing is the first pure-play foundation model company to hit the public markets. OpenAI is still private. xAI is still private. Mistral and Cohere are an order of magnitude smaller. That makes ANTH the de facto proxy for whether public markets will underwrite the AI buildout at the valuations private markets have been assigning for three years running. Every large-cap PM who wanted AI exposure without buying another slug of Nvidia or hyperscaler stock now has a vehicle. Expect the ticker to trade as a sentiment instrument for the entire sector within a quarter.
The community reaction on Hacker News was predictably split. The top comment ran the napkin math on compute costs and concluded the company needs to triple revenue in 18 months just to justify the IPO price, let alone the post-pop level. A rebuttal thread, equally upvoted, argued that enterprise AI spend is still in its first inning and that the real comparable is not Oracle or SAP but AWS circa 2010 — a business losing money at scale on the way to a 30% operating margin. Both can be true. Neither helps you decide whether to re-platform your product on the Claude API next quarter.
One detail that got buried in the prospectus but deserves attention: the S-1 breaks out API customer concentration for the first time. The top 10 customers account for 34% of API revenue. Three of them are coding-assistant companies — Cursor, Cognition, and a third unnamed party widely assumed to be GitHub via its Claude-backed Copilot tier. If you are building on the Claude API and you are not one of those ten accounts, you are now a rounding error in a public company's quarterly guidance. That is a real shift in negotiating posture.
Three practical implications, in rough order of how quickly they will show up in your inbox.
First, pricing. Public companies have shareholders, and shareholders prefer predictability. Expect the per-token pricing on Claude Sonnet and Haiku to stabilize — the 60% price cuts that came every nine months during the growth-at-all-costs phase are over. Enterprise agreements will still have discounts, but the published rate card is now a line item that gets defended on earnings calls. If you have been running a cost model that assumes another halving in 2027, update it.
Second, SLAs and support. The flipside of public-company discipline is that the enterprise motion will tighten. Anthropic has been notoriously thin on account management for mid-market customers — the running joke was that if you weren't spending $500k a month, you got a Discord channel and best wishes. That changes when the sales team has a quota attached to a stock price. Expect dedicated reps, actual uptime guarantees, and a roadmap conversation if you are spending more than a few thousand dollars a month. The flip side: if you are spending less than that, expect to be aggressively migrated to self-serve.
Third, model roadmap volatility. A public Anthropic has to answer two questions every 90 days: did you ship a model that moved the benchmarks, and did you not torch the margin doing it. Those two incentives pull against each other. The quiet bet inside the company has been that Claude 4.7 and the still-unannounced Opus 5 would land before the IPO; both did. The next release cadence is now a corporate-finance question as much as a research question, and the smart money says we see smaller, more frequent point releases rather than the dramatic generational jumps of the last two years. If your eval suite is calibrated to a specific model version, pin it. The surface area for subtle behavioral drift just got larger.
The test for Anthropic is not whether it can grow revenue — it will, probably at 150%+ next year on the back of enterprise deals already in the pipeline. The test is whether it can expand gross margin into the 60s without losing the frontier model race to OpenAI, Google DeepMind, or whatever Chinese lab ships next. If the answer is yes, today's price looks cheap in retrospect; if the answer is no, ANTH becomes the stock that marks the top of the cycle. Either way, the days of foundation models as a purely private-market abstraction are over. From tomorrow morning, the price of frontier intelligence has a chart. Watch it.
It’s been dropping ever since the first trade and appears to be down at $1.26T now. The fact that just a few weeks ago folks wanted $2T plus tells you how far off the original valuations really were.And with this now below where OpenAI has been trying to value itself on the private markets that does
I started programming 50 years ago. At one stage early in my career this mysterious new language C was the thing, and I went on an all expenses paid 2 week course in Singapore where they flew attendees in from around the world to learn this language. Software has been good to me.Now it seems like it
But what we know for certain is that at some point early in the 21st century, all of mankind was united in celebration. We marveled at our own magnificence as we gave birth to A.I.-- Morpheus
Currently on Yahoo Finance: "Anthropic’s first half hour is the worst start for a big IPO since 2008"Makes sense to me. This valuation is pretty insane. I don't see it holding up.
Top 10 dev stories every morning at 8am UTC. AI-curated. Retro terminal HTML email.
Genuinely a massive moment in human history. I'm nervous about what comes. as a SWE the gravy train is basically over in my view. they figured out how to make a program that does the job. Always thought i was set for life. difficult to grasp how much life will change in the coming years.