Photo by Jordan Harrison on Unsplash
What would a skeptical buy-side analyst actually ask Anthropic in a roadshow meeting? Not "how fast is revenue growing" — that answer is already public and already impressive. The harder question is the one Reuters put at the center of its reporting: how much does each dollar of that revenue cost to produce, and does that cost curve bend before the capital runs out?
According to Google News, which surfaced the Reuters report framing Anthropic's rising sales against its steep losses, the company has become the industry's live experiment in whether a frontier AI lab can withstand public-market scrutiny. Reuters' original October 2025 framing — revisited here as of October 27, 2026 — explicitly cast the gap between Anthropic's revenue and its losses as an "IPO test" for the broader AI frenzy.
Thesis: Anthropic's revenue growth is real and verifiable, but the company's IPO readiness depends entirely on a metric almost nobody outside the company can see — the gross margin on inference — and until that number is disclosed, any valuation applied to Anthropic is a bet on an unobserved variable, not an analysis of a known one.
What Is Actually on the Record
Strip away the speculation and the verifiable set is small. As of October 27, 2026, based on the reporting summarized above, here is what can be stated without inventing figures:
Reuters reported in October 2025 that Anthropic's revenue was scaling rapidly while losses remained steep. The annualized revenue run-rate reportedly climbed past the multi-billion-dollar mark during 2025, driven heavily by enterprise customers and API demand for the Claude model family. Reported private valuation reached into the tens of billions of dollars following 2025 funding rounds, with Amazon and Google among the backers. And losses reportedly remained large relative to revenue, attributed to compute, training, and talent costs.
Notice what is missing from that list. No audited income statement. No disclosed gross margin. No capex schedule. No customer concentration disclosure. Reported valuation figures ranged widely across 2025 coverage — which is itself a data point, because wide dispersion in reported valuations usually signals that different outlets are sourcing different rounds, different share classes, or different secondary prices.
That dispersion matters more than it looks. A company with a tight, consistently reported valuation is being priced by a market. A company whose reported valuation varies by outlet is being priced by negotiation.
The Number Nobody Has Published
Here is the non-obvious point the surface coverage mostly skips: "heavy losses" is not one problem. It is at least three problems with completely different investment implications, and conflating them is the single biggest analytical error in AI-lab coverage.
Break the loss into its components. Research cost — training the next model — is forward-looking and, critically, discretionary. A lab that stopped training tomorrow would stop incurring it tomorrow. Talent cost sits in a similar bucket: elevated, competitive, but structurally reducible. Inference cost — the compute burned serving actual customer queries — is the one that scales directly with revenue. If inference gross margin is positive and improving, every incremental enterprise customer makes the picture better. If it is negative, growth makes the picture worse.
Same headline loss. Opposite investment conclusions.
This is where the multi-source picture adds something no single article does. Reuters framed the story as a binary IPO test. Bloomberg's coverage of the company has typically emphasized valuation levels and funding-round mechanics sourced from investors. The Financial Times has tended to focus on unit economics and the compute-cost structure sitting behind the losses. Read together, the three angles describe a company being valued on growth (Bloomberg's lens), questioned on cost structure (the FT's lens), and tested on whether those two can be reconciled under disclosure (Reuters' lens). None of the three resolves the question alone.
A useful back-of-envelope frame for readers: if a software business runs at, say, 75% gross margin, roughly $0.25 of every revenue dollar goes to serving the customer. The entire AI-lab debate reduces to whether that figure for frontier-model inference is closer to $0.25 or closer to $0.90 — and whether it is falling. Nothing in the public record settles it. That is the disclosure gap, and it is the reason an S-1 filing would be a genuine information event rather than a formality.
Chart: Of the six financial inputs an IPO investor would need, public reporting as of October 27, 2026 covers two in figures, one qualitatively, and three not at all. Our read: the gap, not the loss, is the real story.
Photo by Maxim Hopman on Unsplash
The Strategic-Investor Problem
There is a second-order issue the IPO framing tends to underweight. Anthropic's largest backers include Amazon and Google — both cloud providers, and both reported to have tied investment to cloud compute commitments.
A careful skeptic should push on what that circularity does to reported financials. When a cloud provider invests capital into a model lab and the lab spends a portion of that capital buying compute from the same provider, the economic substance is genuinely debatable. It can be a straightforward commercial relationship at arm's-length pricing. It can also compress the lab's apparent cost of capital in ways that would not survive an independent-pricing test.
The fair counter-argument: this structure is not unique to AI, and it is not inherently improper. Semiconductor and telecom history is full of vendor-adjacent financing that worked out fine for everyone. The structure only becomes a problem if the compute is priced off-market and the lab's margins depend on that discount persisting. The honest answer is that outside observers cannot currently distinguish the two cases — and that uncertainty is precisely the kind of thing public filings exist to resolve.
Which is why the IPO question is less about whether Anthropic can list and more about what listing would reveal. The market-structure sensitivity here is real; it is the same dynamic Smart Finance Automation traced through the Nasdaq's 2.6% AI sell-off, where sentiment on AI cost structures moved an entire index, not just the companies directly involved.
Where This Thesis Breaks Down
The bear case deserves better than a paragraph, so here are the three strongest objections to everything above.
First: the disclosure gap may not matter. Plenty of companies have gone public with ugly current economics and strong forward narratives, and investors bought the narrative. If frontier-model demand keeps compounding, the market may simply not require the margin disclosure to assign a large valuation. Arguing that disclosure is decisive assumes a level of public-market discipline that has not consistently shown up in the last decade.
Second: training cost may be the wrong thing to treat as discretionary. The framing above separates "discretionary" research spend from "structural" inference spend. But in a two-horse frontier race — Anthropic and OpenAI are explicitly positioned as the twin bellwethers in the Reuters framing — stopping training is not really optional. A lab that stops training stops being a frontier lab, and its revenue follows. Under that reading, research cost is as structural as inference cost, and the loss picture is considerably worse than a component-by-component analysis suggests.
Third: there may be no IPO to analyze. A company with access to multi-billion-dollar strategic rounds from Amazon and Google has limited incentive to accept quarterly disclosure, shareholder litigation risk, and an activist-accessible share register. Private capital at this scale is a substitute for public markets, not a waystation to them. Every analytical frame that treats an Anthropic IPO as inevitable is making an assumption the research data does not support.
That third objection is, on balance, the strongest. The market context of late 2025 — investor attention rotating from AI hype toward unit economics and credible paths to profitability — cuts both ways. It raises the bar for an IPO at the same time as it raises the strategic value of staying private until the numbers look better.
Watchlist: What to Track, in Order
For readers doing their own investment research, the useful posture here is not prediction but monitoring. Four observable signals, ranked by information value:
The single highest-information event available. A filing would convert three currently-undisclosed line items into audited disclosure. Until one appears, Anthropic cannot be valued by ordinary investors on anything other than secondhand reporting. Worth researching: SEC EDGAR full-text search, and the standard confidential-submission disclosure patterns that typically precede a listing.
Amazon and Google disclose capital expenditure quarterly and discuss AI infrastructure on earnings calls. Because both are reported Anthropic investors with compute ties, their disclosures function as an indirect read on frontier-lab compute demand — and are available on a fixed schedule, unlike anything Anthropic itself publishes.
Published API pricing is one of the few genuinely public, comparable, time-series datasets in this sector. Falling prices with stable capability imply improving inference economics; falling prices with flat capability may simply imply a price war. The distinction is the whole ballgame for margin analysis, and it is checkable by anyone.
Where reported, employee-share secondary pricing is a cleaner signal than headline round valuations, because secondaries typically lack the structural preferences that can inflate primary-round headline numbers. Divergence between the two is itself informative.
The bottom line from our analysis: the most likely outcome is not a near-term IPO but a prolonged private period in which Anthropic's economics improve or deteriorate outside public view, with cloud-provider capex serving as the market's primary proxy. That makes sector analysis of Amazon, Google, and the broader AI supply chain more immediately actionable for ordinary investors than speculation about a listing that may never be filed. Market trends in this space are currently being read through infrastructure spending, not through the labs themselves — and that is a meaningful shift from where attention sat two years ago.
Frequently Asked Questions
How much revenue does Anthropic make as of 2026?
Anthropic does not publish audited financials. Reuters reported in October 2025 that the company's annualized revenue run-rate had climbed past the multi-billion-dollar mark, driven heavily by enterprise and API demand for Claude models. An exact, independently verifiable figure was not available at the time of writing on October 27, 2026. Treat any precise revenue number circulating publicly as a secondhand report rather than a disclosed figure.
Is Anthropic profitable right now?
Based on reporting summarized as of October 27, 2026, no. Reuters described losses that remained steep despite rapidly scaling sales, attributed to compute, training, and talent costs. Because Anthropic is private, there is no audited statement confirming the magnitude, and the composition of those losses — research versus inference versus headcount — has not been disclosed.
Is Anthropic planning an IPO, and when could it happen?
No IPO has been confirmed. The "IPO test" framing in the Reuters coverage is analytical — positioning Anthropic as a bellwether for whether richly valued AI startups could withstand public-market scrutiny — not a report of a filed or announced offering. Any forward-looking timeline circulating without a cited company statement or regulatory filing should be treated as speculation.
Who are Anthropic's biggest investors?
Reported backers include Amazon and Google, both of which have made multi-billion-dollar strategic investments reported to be tied to cloud compute commitments. That structure is worth researching carefully, because investor-and-supplier dual relationships can affect how a company's cost base should be interpreted.
Disclaimer: This article is editorial commentary and investment research for educational and informational purposes only. It does not constitute financial advice, a recommendation, or an endorsement of any security, and it does not reflect independent product testing. Valuation and revenue figures referenced are drawn from published third-party reporting, not from audited company disclosures. Always do your own research and consult a licensed financial advisor before making investment decisions. Research based on publicly available sources current as of October 27, 2026.