Is $2 trillion for Anthropic believable? Three analysts pull apart the number

Anthropic is reportedly targeting a November Nasdaq listing at a valuation investors are discussing near $2 trillion.

That figure would more than double the $965 billion post-money valuation set in the company’s May 2026 Series H round, and it would make Anthropic’s IPO the largest by valuation, eclipsing SpaceX’s reported $1.77 trillion debut earlier this year.

The headline number invites an obvious question: is it justified?

Invezz put that question, and two others probing where the real risk sits, to three market professionals.

Their answers converge on a single theme. The $2 trillion figure is not really a valuation of what Anthropic is today.

It values what the market believes Anthropic’s revenue mix will be over the next two to three years, and the details of that mix, not model capability, are where the argument for or against $2 trillion actually lives.

The growth the number is built on

Anthropic’s annualised revenue run rate exceeded $65 billion by the end of July 2026, up from roughly $47 billion in May and about $9 billion at the end of 2025, a more than sevenfold increase in under a year.

Some investors expect Anthropic’s annualised revenue to reach $100 billion to $120 billion by year-end, while the company is reportedly projecting $190 billion to $200 billion in revenue in 2028.

That trajectory is a central part of the case for a $2 trillion price tag on a company that only recently turned adjusted operating income positive, after burning roughly $5.6 billion in cash in 2024.

Kate Leaman, chief market analyst at AvaTrade, argues the growth curve is precisely why sceptics may end up wrong about the price being too rich rather than too cheap.

“A $2 trillion price tag sounds enormous until you remember revenue went from a $9 billion run-rate to over $100 billion in about a year,” Leaman told Invezz.

Leaman is referring to the projected annualised revenue trajectory rather than reported full-year revenue.

There is a scarcity argument sitting alongside the growth argument, and Leaman thinks it matters more than investors are currently pricing in. OpenAI is not publicly tradeable.

If Anthropic lists, it becomes the only large, liquid, pure-play route into frontier AI available to public market investors, a position that “does strange things to price discovery when a trade this size has nowhere else to go.”

Combine that with the conventional logic of hot tech IPOs, where underwriters sometimes accept a lower offer price to support a strong first-day performance rather than risk a listing that limps, and Leaman’s case is that undershooting fair value is at least as plausible as overshooting it.

But she is explicit that the re-rating case depends on several conditions holding simultaneously, not just growth continuing.

Revenue needs to remain enterprise- and API-driven rather than concentrated in a handful of large compute deals, because public market investors treat recurring revenue and one-off contracts very differently.

Anthropic’s safety spending, and partnerships built around it such as the Accenture tie-up, need to demonstrate a genuine competitive advantage rather than a cost centre, evidence that enterprises are choosing Anthropic partly because of its governance story rather than in spite of it.

And the macro backdrop needs to cooperate.

Leaman points to the Federal Reserve’s first rate hike in more than three years as a complicating factor: a market discounting future cash flows at a higher rate is “a much harder room for any re-rating story to land in, however good the growth numbers look.”

Revenue quality is the whole ballgame

If Leaman’s framing is about what could push the valuation higher, Veni Dhir, director of corporate venture capital at ADP, is focused on what determines whether the $2 trillion figure is actually supported by the underlying economics.

Dhir shared her views in a personal capacity and not on behalf of her employer.

Her starting point is that investors need to look past the headline valuation and examine what sits beneath the revenue growth.

That distinction matters because Anthropic’s rapid growth has been driven in significant part by usage-based AI consumption.

API revenue can scale quickly as customers increase workloads, but it can also fall if enterprises reduce inference spending, switch to cheaper models or move workloads elsewhere.

The more important test, Dhir argues, is whether Anthropic can turn that usage into something stickier by embedding Claude into enterprise workflows.

Its push into enterprise applications, including Claude Code and agent deployments, could make the relationship harder to displace than a model API alone.

“Anthropic’s move into enterprise applications changes the picture in its favor,” Dhir said.

“Selling ‘cognition’ is a race to the bottom — but selling a workflow outcome inside the enterprise has pricing power and switching costs. The catch: it puts them in a street fight with the enterprise software incumbents who already own the distribution.”

That creates a different competitive challenge. Anthropic’s model capabilities may give it an opening, but companies such as Salesforce, Microsoft and SAP already control deep enterprise relationships, procurement channels and software workflows.

For Anthropic, the question is whether it can become embedded deeply enough that customers continue expanding their spending even as the cost of accessing AI falls.

Dhir’s central warning is that the commoditization risk may eventually move beyond the model layer and into the applications built around it.

“The biggest risk to the IPO thesis is commodity cognition: if models keep converging, margins compress toward zero and the $2T is really a bet on Anthropic becoming the default enterprise AI layer. For Anthropic to outperform, it needs to win distribution — the workflow layer, not the model layer, is where the durable value accrues. That’s exactly what I underwrite as an investor,” Dhir told Invezz.

That makes the quality of Anthropic’s growth more important than the growth rate itself.

The key evidence would be whether enterprise customers expand their spending after initial deployments, how much of that revenue comes from durable contracts rather than variable usage, and whether inference costs fall faster than pricing.

Dhir also points to the economics behind the headline revenue numbers.

Gross margins after inference costs, contracted compute obligations, and the gap between adjusted and full profitability will determine how much of Anthropic’s growth ultimately translates into cash flow.

At $2 trillion, Anthropic therefore does not need to prove that AI demand is enormous.

It needs to show that it can capture enough of that demand in enterprise workflows that customers do not easily abandon, while improving inference economics fast enough to prevent model commoditization from eroding margins.

Pricing a risk with no historical playbook

The third question was less about valuation mechanics and more about how a rational investor should even think about catastrophic or extinction-level AI risk when pricing a single company’s equity.

Mohanad Yakout, senior market analyst at Scope Markets, drew a sharp distinction that has practical implications for how the $2 trillion figure should be read.

“Extinction risk can’t be priced into one stock, but catastrophic-yet-survivable liability can,” Yakout said.

In other words, an investor who genuinely believes advanced AI carries a non-trivial probability of a civilisation-scale outcome should express that belief through their overall sector allocation to AI equities, not through a discount applied specifically to Anthropic’s share price relative to its peers, because in that scenario the distinction between Anthropic and any other AI company stops mattering.

Catastrophic but survivable liability, the kind more relevant to how public equity markets actually function, is a different exercise entirely.

“Catastrophic liability is priced firm by firm,” Yakout said. “Estimate probability times cost times the share the company would actually bear, then adjust for deployment scale, safety controls, insurance and legal shields, and governance.”

This is closer to how markets already price liability risk in other high-stakes industries, such as pharmaceuticals facing product liability and energy companies exposed to environmental disasters, where the size of the potential loss is weighed against the probability and the portion of that loss the company itself, rather than insurers, regulators or the wider industry, would actually absorb.

Yakout flagged two specific traps investors should avoid when attempting this exercise.

The first is multiplying a very small probability by an unbounded or “infinite” loss figure, a calculation that produces an artificially large risk number despite the fact that, as he put it, “shareholders can lose at most 100%.”

Equity holders’ downside is capped at the value of their investment regardless of how catastrophic the underlying event might be for society at large, and treating catastrophic scenarios as an open-ended liability overstates the specific financial risk to shareholders.

The second trap is false precision, since AI catastrophes, unlike defect rates in aviation or pharmaceuticals, have no real historical base rate to draw on.

Any probability an analyst attaches to such an event is necessarily a judgement call rather than an actuarial calculation, and Yakout’s implicit warning is against dressing that judgement up as something more rigorous than it is.

What the $2 trillion number is actually testing

Put together, the three perspectives point to the same underlying tension.

The bull case for $2 trillion, articulated most directly by Leaman, rests on the growth rate continuing to outpace what conservative bankers are willing to write into a prospectus, combined with Anthropic’s scarcity value as the only pure-play route into frontier AI available to public investors.

The more cautious framing from Dhir is that the headline number is close to meaningless without decomposing it into revenue quality, inference unit economics and enterprise conversion, because a valuation built on usage-based API revenue behaves very differently from one built on durable enterprise contracts, even if the trailing twelve-month figures look identical.

And Yakout’s framework is a reminder that whatever risk premium or discount investors attach to Anthropic specifically should reflect survivable, firm-level liability exposure, not an attempt to price an industry-wide tail risk into a single equity.

None of the three analysts dismisses the $2 trillion figure outright, and none endorses it unconditionally either.

What they agree on, from three different angles, is that the number is a proxy for a set of much narrower and more testable questions: whether enterprise spend on Anthropic behaves like recurring software revenue or volatile compute consumption, whether the company can build a distribution moat outside pure model capability, and whether its governance and safety spending function as a genuine differentiator rather than a cost line.

Those are the metrics likely to move the share price in the quarters after listing, far more than the headline valuation itself.

(With inputs from Devesh Kumar and Utkarsh Roshan)

The post Is $2 trillion for Anthropic believable? Three analysts pull apart the number appeared first on Invezz