Reporting from the New York Times on September 18, 2026, echoed by Bloomberg and other outlets, said Anthropic's annualized revenue run rate is on track to top $100 billion sometime in 2026, and that the company is targeting a possible November 2026 initial public offering. Investors involved in early discussions have reportedly floated a valuation around $2 trillion. None of these figures is confirmed by Anthropic itself, audited, or drawn from a regulatory filing. The IPO is the headline. The more useful thing to understand, whether or not the listing happens on that timeline, is the number doing the heavy lifting in every one of these stories: annualized revenue run rate.
Anthropic
OpenAIRun rate is a real and useful metric. It is also not the same thing as a year of revenue, and the gap between the two widens exactly when a company is growing fastest, which is exactly the situation Anthropic is reportedly in. This piece is a plain explanation of what the metric is, why it gets used for companies like Anthropic, why it draws scrutiny from analysts and journalists, and how to read the specific numbers attached to Anthropic's reported ramp with that context in mind.
What "annualized revenue run rate" actually means
Run rate is a projection built from a short recent period. A company takes its most recent month or quarter of actual billed revenue and multiplies it up, by 12 for a month or by 4 for a quarter, to describe what a full year would look like if that pace held steady. The output is usually labeled an "annualized run rate" or "ARR" (not to be confused with annual recurring revenue, a related but distinct SaaS metric that also gets shortened to ARR).
How an annualized run rate is built
The mechanical steps behind a headline like "$100 billion run rate," and how that differs from a year of audited revenue.
For a company growing as fast as the one in this piece, the two numbers can differ by tens of billions of dollars at the same moment in time.
The appeal of the metric is obvious for a company whose revenue is changing quickly. A full-year revenue figure for a fast-growing business blends months of small numbers with a final month of much larger ones, so the annual total understates how big the business has actually become by the time anyone reads the report. Run rate strips that lag out and answers a narrower, more current question: how much is this business billing right now, extrapolated forward. That makes it a genuinely informative growth signal, and it is why venture investors, and now IPO bankers, lean on it so heavily when the underlying business is a matter of months old at its current scale.
Why the metric draws scrutiny
The same property that makes run rate useful, its speed, is also the source of every legitimate criticism of it.
A run rate assumes the most recent period repeats forever. For a business adding tens of billions of dollars in a matter of months, that assumption is doing an enormous amount of work.
Metir AI analysis
Four specific objections come up repeatedly in coverage of Anthropic's numbers and of hyper-growth AI companies generally. First, a run rate assumes the latest period continues indefinitely, when in practice growth curves bend, whether upward through further acceleration or downward through market saturation, competition, or a slowdown in a single large customer's usage. Second, a single strong month or quarter, driven by a large one-time contract, a pricing change, or a promotional push, can inflate a run rate well above the business's steady-state pace. Third, run rate is not an audited figure. It is typically self-reported, drawn from internal billing systems and shared with investors or press, without the verification a public company's financial statements go through. Fourth, and perhaps most consequential for valuing a company like Anthropic, run rate says nothing on its own about margins, customer concentration, contract duration, or churn. A dollar of revenue that costs more than a dollar in compute to serve is very different from a dollar of high-margin software revenue, and the run-rate headline does not distinguish between them.
None of this means the growth is not real. Multiple independent outlets, citing different sourcing, have converged on a similar trajectory for Anthropic, which is itself a form of corroboration. It means the number is a snapshot of momentum, not a settled measure of the business's size.
Reading Anthropic's reported ramp
With that framing in place, Anthropic's reported figures are a useful worked example precisely because the growth is so fast that the run-rate caveat matters more than usual.
Anthropic's reported annualized run rate, 2025 to 2026
Each bar is a snapshot annualized run rate at a point in time, as reported by press coverage citing investor materials. None of these figures is an audited or company-confirmed annual revenue total.
The dashed bar is a forward trajectory reported by the press, not a closed period. Source: Bloomberg/NYT (Sept 18, 2026), PYMNTS, GraniteShares, Gulf News.
| Date | Reported annualized run rate |
|---|---|
| End of 2025 | ~$9 billion |
| May 2026 | ~$47 billion |
| End of July 2026 | ~$65 billion |
| End of 2026 (on track, per reporting) | ~$100 billion |
Every figure in that table is a run rate, not a booked annual total, and none of them is confirmed by Anthropic. Reporting attributes the growth largely to enterprise and developer adoption of Anthropic's Claude Code coding assistant and its Cowork workplace product, both of which scale with usage rather than with a fixed subscriber count, which is consistent with a metric that can move this quickly. Read the table as four separate snapshots of velocity at four separate moments, not as four data points on a single audited revenue line, and the ten-and-a-half-fold jump from the end of 2025 to the projected end of 2026 becomes easier to hold alongside the caveats above rather than in place of them.
Run rate vs audited revenue: what each one actually measures
The two figures answer different questions, and an IPO is the event that forces a company to publish both side by side for the first time.
| Run rate | Audited trailing revenue | |
|---|---|---|
| What it measures | Current pace, annualized | Revenue actually recognized over a closed period |
| Verified by | Usually self-reported, unaudited | External auditors, filed with regulators |
| Reacts to | The most recent month or quarter | The full period, smoothed by slower earlier months |
| Captures margin or cost | No | Only alongside separate cost and margin disclosures |
| What a fast grower shows | Run rate typically exceeds trailing revenue | Trailing revenue is typically lower, by a widening amount as growth accelerates |
For a company growing as fast as Anthropic is reported to be, the gap between the two rows can be tens of billions of dollars at the same moment in time. That gap is not evidence of anything improper. It is the mechanical consequence of annualizing a recent period for a business whose recent period looks nothing like its earlier ones.
What a ~$2 trillion valuation implies
Reported investor discussions have floated a valuation of roughly $2 trillion for a possible Anthropic listing. Against a $100 billion run rate, that works out to roughly 20 times run-rate revenue, a very high multiple by the standards of established public companies, though one that has precedent among the fastest-growing private technology companies of this AI buildout. Some reporting on 2028 revenue projections for Anthropic, in the range of $190 billion to $200 billion, frames the same valuation as closer to 10 times a multi-year-out revenue projection rather than 20 times a current-moment run rate, which is a materially different way of describing the same number.
Which framing is the fairer one is exactly the kind of question a public listing exists to resolve. A multiple against a snapshot run rate treats today's pace as durable. A multiple against a multi-year projection treats a specific forecast, itself unaudited and unconfirmed, as reliable. Neither framing is wrong on its face, and this piece takes no position on which one the market should use, or on what Anthropic is actually worth. The point worth taking away is narrower: the same $2 trillion figure can be described as either a very aggressive multiple or a more moderate one, purely by choosing which revenue denominator to divide it by, and both denominators are currently unaudited numbers rather than filed financial statements.
What an S-1 filing would force into the open
If Anthropic does file publicly and lists, the filing itself changes the nature of every number discussed here. A registration statement requires audited financial statements covering defined historical periods, not a projected annualized pace. It requires disclosure of gross margin, which for an AI lab means revenue net of the cost of the compute used to serve inference, a figure that has never been publicly disclosed for Anthropic at the level of detail a prospectus demands. It typically requires disclosure of customer concentration, meaning how much of revenue comes from a small number of large enterprise contracts versus a broad base, which matters a great deal for how durable a growth curve actually is. None of the reporting so far includes any of these figures for Anthropic. They are also, taken together, close to the actual list of open questions serious observers of the offering have raised.
The takeaway
A $100 billion run rate is a real and remarkable number, and the reported trajectory behind it, from roughly $9 billion at the end of 2025 to that pace less than a year later, is one of the fastest revenue ramps reported anywhere in the current AI buildout. It is also, mechanically, a projection built from a recent snapshot rather than a year of booked, audited revenue, and the gap between the two grows wider precisely as growth accelerates. The healthiest way to read a headline like this one, for Anthropic or for any fast-growing AI company reporting similar figures, is to hold the growth signal and the accounting caveat in the same sentence rather than picking one and discarding the other.
That same discipline, reading a vendor's own growth or performance claims with a little skepticism rather than taking a headline number at face value, is worth applying broadly across the AI industry, where self-reported benchmarks and run-rate figures are common currency. It is also part of the case for not building a product or a workflow around a single model provider's roadmap or pricing. A platform like Metir AI, which gives teams access to Claude alongside GPT, Gemini, Grok and other leading models in one workspace, is one way to treat any single vendor's numbers, whatever they eventually turn out to mean, as something to evaluate rather than something to depend on.
Sources:
- Anthropic's Annualized Revenue to Top $100 Billion in 2026, NYT Reports | Bloomberg
- Anthropic Targets November IPO as Revenue Surges | PYMNTS
- Anthropic IPO 2026 Explained: From $965 Billion to a Possible $2 Trillion Listing | GraniteShares
- AI, Anthropic, Run Rate, and the IPO Filing Revenue Metric | FourWeekMBA
- Anthropic on Track to Top $100 Billion Revenue Pace Ahead of Potential IPO | Gulf News
Image credits
Header image: the New York Stock Exchange building facade at 11 Wall Street, New York City, photographed by 颐园居 via Wikimedia Commons, licensed under CC BY-SA 4.0. In-body photograph of the New York Stock Exchange trading floor by Carol M. Highsmith, via the Library of Congress and Wikimedia Commons, public domain. Neither photo depicts Anthropic, its offices, or any Anthropic IPO event, which has not taken place; both illustrate the New York Stock Exchange and the public markets generally.