Anthropic Won’t Sustain a $1 Trillion Valuation by End of 2027

My call: the trillion will not stick

The consensus story writes itself. Anthropic sprints from obscure AI lab to $65 billion annualized revenue, files to go public, and rings the bell as a trillion plus national champion. In the PowerPoint, this is how the AI age becomes real: the Claude maker graduates into the same weight class as Apple and Microsoft in roughly the time it usually takes a startup to hire HR.

My call: by 31 December 2027, Anthropic will not have a sustained public valuation at or above $1 trillion. Measured cleanly, that means no 20 trading day stretch in its first post IPO year where the market cap averages $1 trillion or more.

A trillion dollar print is absolutely in play, especially on day one. The problem is not touching the line. It is staying above it once hype, rates, and revenue quality all get to vote at the same time.

The IPO fantasy vs. the math on the table

Start with the fairy tale. The underwriters, led by Morgan Stanley and Goldman Sachs, are floating an offering that could raise more than $100 billion at a valuation whispered as high as $2 trillion. That would top SpaceX’s recent record IPO, hand public markets a new pure play frontier AI bet, and cement Anthropic as the AI age’s first true mega cap idol.

The pitch has three headline stats.

First, growth. Anthropic’s annualized revenue run rate, by leaked investor updates, jumped from roughly $9 billion at the end of last year to more than $65 billion, with about $11.6 billion in Q2 revenue. That is not a curve, it is a wall.

Second, the story of stability. The revenue mix leans on Claude API usage, resale through AWS and Google, and multi year contracts with blue chip customers like JPMorgan, Netflix, and Salesforce. This is marketed as “infrastructure grade” business, less flaky than OpenAI’s consumer heavy base.

Third, positioning. OpenAI has quietly stepped back from its own listing and is bleeding cash. Anthropic’s bankers would like you to see this as a passing of the torch: one lab retreats from public scrutiny while the other bravely offers itself to your pension fund.

Now the part that ruins the roadshow slides. Projected net losses for 2025 are said to be near $42 billion, roughly a fivefold increase from 2024. Compute and data centers eat capital. GPUs do not care about your TAM slide. To keep the model frontier class, Anthropic must keep pouring money into chips, power, and research at a scale more reminiscent of a sovereign wealth fund than a software company.

In private markets, this is a feature. In public markets, it is a discount factor.

Why $1 trillion is a ceiling, not a floor

If Anthropic lists anywhere near the rumored $2 trillion, the IPO will test three fragile beliefs at once.

Belief one: AI revenue is software, not infrastructure. The whole valuation case hinges on investors treating that $65 billion run rate like high margin, high retention software revenue. The reality is murkier. A meaningful chunk of revenue flows through cloud resale with AWS and Google, where economics will look more like a channel partnership. Another chunk is early enterprise experimentation and committed usage that can be repriced or renegotiated once CFOs sober up.

If public investors decide Anthropic’s revenue profile rhymes more with an infrastructure provider than with pure SaaS, the multiple falls fast. A trillion dollar valuation on anything less than software style margins starts to look like performance art.

Belief two: pricing power survives open weight reality. In the IPO pitch, Anthropic keeps charging a premium for best in class models while cheaper, open weight systems nibble around the edges. In the market Anthropic actually lives in, open weight models and hyperscaler native offerings are already “good enough” for many workloads. That does not wipe out Anthropic, but it squeezes average selling prices and muddies the story that these models are a scarce luxury good.

The more regulators, enterprises, and engineers normalize open weight alternatives, the less room Anthropic has to act like a monopoly on intelligence. The moment discounted enterprise renewals start to show up in the filings, the valuation math gets redrawn.

Belief three: public markets will underwrite frontier losses indefinitely. Private investors can tell themselves the losses are a rational land grab. Public investors have to live with quarterly marks and rising real rates. A business burning tens of billions a year on speculative frontier capability, while also flagging regulatory and public backlash as formal risk factors, is not some cozy toll road.

For a $1 trillion valuation to hold, investors have to believe in years of blazing growth, improving unit economics, and relatively frictionless data center expansion. That is an everything goes right scenario, priced at essentially zero discount. Not impossible, just rare.

How this most likely breaks

There are three main paths from here to a scored outcome.

The cleanest is that bankers quietly walk expectations back. The formal IPO range tops out below a trillion, or only brushes it on fully diluted math. Anthropic still goes public in blaze of AI glory headlines, but never actually sustains a trillion dollar market cap once the stock starts trading freely.

The spicier variant is that they lean into the myth. Keep the float tight, dangle a $2 trillion whisper, lean hard on the scarcity of a pure play frontier model asset. In that world, I would fully expect a day one or week one spike where the tape briefly reads “$1T+”. Then stabilization support fades, lockups start to matter, macro wobbles a bit, and the stock settles into something meaningfully smaller.

The third is the least talked about: timeline slippage. A bruising regulatory fight over data centers, a sudden reversal in AI sentiment, or a general market accident could push the IPO into 2028 or force Anthropic into some hybrid listing or restructuring that makes the neat $1 trillion test inapplicable. That still resolves my claim. “No sustained $1T+ by end 2027” includes “no public listing to hang it on.”

To be wrong, I need a world where AI euphoria stays hot, rates stay friendly enough, open weight competition does not gut pricing, regulators stage a lot of hearings but impose few binding constraints, and Anthropic’s S 1 manages to look closer to Nvidia than to a supercharged telecom. That world is coherent. It is just not the median one.

What to watch between now and the bell

Over the next four months, a few signals will tell you whether my skepticism is about to be steamrolled.

First, the S 1. The revenue mix, gross margins, and contract lengths are the whole ballgame. If a big share of that $65 billion run rate is low margin resale, promo heavy usage, or short tenor experiments, the trillion case leaks air fast.

Second, the pricing range and order book quality. If the initial talk is $2 trillion and the formal range crawls out shy of $1 trillion, that is the market saying the fantasy and the appetite do not match. If the range holds but the coverage is thin and skewed to fast money, expect a choppy first year.

Third, early trading behavior. A violent spike then drift lower is the classic pattern of a story that sold better in the green room than on the actual show.

Finally, the politics. Anthropic is planning to list public backlash against AI and data centers as a risk factor in its prospectus. Seven in ten Americans have already told Gallup they do not want those facilities anywhere near them. Even if Congress moves slowly, local permitting boards do not, and bond markets understand NIMBY.

The satirical verdict

Put it together and the picture looks less like “the Nvidia of intelligence” and more like “a very large margin call disguised as a growth story.” The IPO can still be historic, still make early investors rich, still produce a my God it is full of zeros moment on CNBC.

What it probably will not do is convince public markets to treat a five year old, hyper capital intensive research lab as a permanent member of the trillion dollar club.

If I am wrong, you will get to tell your grandkids you bought Anthropic at a trillion and it worked. If I am right, you will get to tell them you watched the world’s smartest people build AGI, then discover that even superintelligence cannot save a busted multiple.