NextFin News - Anthropic is about to face the question private markets never asked. Ahead of the AI lab's Oct. 14 pre-IPO investor day at its San Francisco headquarters, Momei Qu, managing director of PSP Growth at Canada's Public Sector Pension Investment Board, framed the tension that will decide whether the landmark offering clears its reported $1.8 trillion-to-$2 trillion valuation: public-market investors may accept low margins and heavy capital spending for fast-growing AI companies, or they may demand proof of returns and profitability instead. The answer will not just price one stock. It will set the template for how public markets value the entire frontier-AI buildout, and it will tell the pension funds, endowments, and sovereign wealth vehicles that have poured capital into private AI whether the exit door is actually open.
The Situation: A $2 Trillion Lab Meets a Colder IPO Window
Anthropic's investor day, scheduled for Oct. 14, is effectively the starting gun on what could become the largest initial public offering in history. The company confidentially submitted a draft registration statement on Form S-1 to the Securities and Exchange Commission on June 1, four days after it closed a $65 billion Series H round on May 28 that valued it at $965 billion post-money. Prospective institutional investors are now pegging a fair public valuation between $1.8 trillion and $2 trillion - roughly double the private mark set only months earlier. The event gives a select group of buyers their first formal opportunity to quiz senior management before a roadshow that could begin as early as November.
The growth story behind that re-rating is extraordinary. Reported annual revenue surged roughly twelve-fold in 2025 to about $4.6 billion from $386 million the prior year, and by July 2026 annualized revenue had reportedly crossed $65 billion, with expectations above $110 billion by year-end. In the second quarter of 2026, the company turned positive on an adjusted operating-income basis - the milestone its investor pitch is built around. The company has also said revenue has grown tenfold year over year for three straight years, a claim that tracks the same compounding trajectory from a different base.
But the margin for error is thin, and Qu's framing lands on a market that has already begun to push back. High-profile listings have been scrapped or delayed in recent months: wearable maker Oura postponed its U.S. offering in late September, citing uncertainty in the IPO market, and Australia's Firmus, an Nvidia-backed AI data-center operator, abandoned a planned listing that would have valued it at about $30.6 billion. OpenAI has postponed its own listing, with its chief executive calling a 2026 debut ill-advised, and is pursuing a private round at a reported $1.4 trillion valuation instead. The mixed signals are visible even in the IPO exchange-traded funds that track the new-issue market: the Renaissance IPO ETF was up roughly 17 percent year to date as of early October, but it fell about 11.6 percent over the three months ended in September - a sharp reversal that shows how quickly the window can narrow.
Qu's vantage point matters. PSP Investments ended its fiscal year on March 31, 2026, with C$320.6 billion in net assets under management, up 7 percent from the prior year, and posted a one-year net return of 6.5 percent. Its Growth platform invests in earlier-stage, high-growth B2B software companies - the same category Anthropic is trying to reposition as a public-market compounder. When a pension investor that measures performance in decades, not quarters, asks whether heavy capex deserves a premium or a discount, the question carries the weight of the capital that funds the AI buildout in the first place.
The tension Qu identified is not a side issue. It is the central valuation question of the AI era: does the market pay for growth and strategic positioning, or does it pay for cash returns? The next few weeks will produce the first real answer.
The Mechanism: Why Capex-Heavy AI Breaks the Old Software Valuation Model
Traditional software companies scale with near-zero marginal cost. Frontier AI labs do not. Every additional user of a large language model consumes inference compute; every model generation requires training clusters that cost billions. The transmission channel is simple: revenue growth in AI is mechanically tied to capital expenditure growth. A lab cannot grow faster than the compute it can secure, power, and cool.
That linkage is why Anthropic's reported $518 billion in future cloud, computing, and infrastructure obligations - nearly all non-cancellable - matters more than the headline valuation. Those commitments are not optional capacity; they are a fixed-cost stack that must be serviced regardless of demand. In private markets, that stack was framed as strategic positioning, the price of securing scarce chips and power ahead of rivals. In public markets, it will be framed as operating leverage - and operating leverage cuts both ways.
Here is the mechanism in plain terms. If revenue keeps compounding at the reported pace - from $65 billion annualized to beyond $110 billion - the fixed compute base becomes a moat, and the $518 billion in obligations looks cheap relative to the cash it helps generate. If revenue growth decelerates while those obligations remain, the same fixed costs become a margin compressor that no amount of pricing power can fully offset. Public-market investors price that asymmetry differently than private-growth funds do, because public funds cannot mark illiquid positions forward; they must mark to a daily price set by marginal sellers. A private fund can hold through a bad quarter and re-underwrite in three years. A public investor faces a price every day, and that daily price disciplines the multiple.
This is the structural break from the 2020-2021 software boom. Then, capital intensity was low and margins expanded with scale. Now, scale requires more capital, not less. The question is whether the market treats that as a permanent feature of the industry or a transitional phase of the race.
Anthropic's own leadership has argued for the disciplined path. Daniela Amodei, the company's president and co-founder, has said the lab has aimed to be "as judicious with the resources that we have while still operating in this space where it's just a lot of compute," adding that Anthropic "has always had a fraction of what our competitors have had in terms of compute and capital, and yet, pretty consistently, we've had the most powerful, most performant models for the majority of the past several years." That pitch - do more with less - is the counter-narrative to the outbuild-everyone strategy, and it is the story the Oct. 14 investor day must sell. But even Anthropic is not operating on a shoestring: the company has said it carries roughly $100 billion in compute commitments of its own, with requirements expected to keep rising.
Anthropic has always had a fraction of what our competitors have had in terms of compute and capital, and yet, pretty consistently, we've had the most powerful, most performant models for the majority of the past several years.
The quote captures both sides of the valuation debate. If efficiency can keep Anthropic at the frontier with less capital, the capex burden is smaller than the race implies and the premium is justified. If the race is won by whoever spends the most, the lab with the smaller balance sheet loses - and the $518 billion obligation stack becomes a trap rather than a moat. Amodei's own words frame the bet the market is being asked to make.
Cyclical or Structural: The Call That Decides the Valuation
My judgment: the capital intensity of frontier AI is cyclical in its severity but structural in its existence. Getting this call right decides the valuation, and getting it wrong flips the conclusion.
The severity is cyclical. Today's race-condition capex, with multiple labs overbuilding to avoid being left behind, will normalize as model performance converges and inference becomes commoditized. History offers three comparable cycles. The fiber-optic buildout of the late 1990s saw carriers lay enough dark fiber to carry global traffic many times over; when demand failed to keep pace, the capex collapsed and the sector re-rated violently, with hundreds of carriers failing before traffic eventually caught up with capacity a decade later. The shale boom of the 2010s saw producers drill at losses to hold acreage; when capital discipline returned, output kept growing but equity valuations never recovered to boom multiples - the sector produced more oil and less value. The cloud buildout of 2020-2022 saw hyperscalers guide to sustained high capital spending, then pull back sharply in 2023 as utilization lagged - a mean-reversion that took only two quarters to materialize.
But the existence of high compute intensity is structural. Unlike fiber or shale, AI compute is not a single-commodity pipe; it is a general-purpose input that compounds in value as models improve and agents multiply. A regime change is visible in the rules of the industry: compute has become the binding constraint on output, model quality scales with compute spend, and the largest labs have turned infrastructure ownership into a competitive advantage rather than a cost center. That driver will not self-correct the way an inventory glut does. Even if the race cools, the baseline level of compute intensity remains permanently above the pre-2023 norm, because the applications being built - agents, coding assistants, enterprise workflow automation - consume compute continuously rather than in one-time transactions.
So the correct read separates the two legs rather than blending them. The short-term cyclical leg says capex growth will slow and multiple compression will follow if revenue growth disappoints - and the $1.8 trillion-to-$2 trillion valuation being discussed prices the structural leg almost entirely and the cyclical leg not at all. The long-term structural leg says the winner that survives the shakeout owns a utility-like asset base that public markets will eventually re-rate upward. The near-term trade and the long-term investment are pointing in opposite directions, and the IPO pricing will force them into a single number.
The Second-Order Question the Market Is Not Asking
The first-order read is obvious: a hot IPO, a big valuation, a volatile debut. The second-order question is different, and it is the one Qu's framing points toward: what happens to the hyperscalers that are financing this buildout?
Anthropic's $518 billion in obligations flows largely to cloud providers. If the public market rewards Anthropic's capex-heavy model, the hyperscalers' own capital-expenditure guidance gets validated and their multiples expand - the IPO becomes proof that the buildout is economically sound, not just strategically necessary. If the public market rejects it, the transmission runs backward: AI labs cut or renegotiate commitments, hyperscaler capex plans get marked down, and the semiconductor supply chain that has priced a decade of AI demand growth faces a growth revision. The IPO is not just a listing; it is a referendum on the entire capex chain that has carried the equity market for three years. A $2 trillion Anthropic validates trillion-dollar capex cycles. A $1.2 trillion Anthropic calls them into question.
There is also an expectation gap embedded in the timing. Because of securities-law windows, the company can list in late October while showing audited financials only through the second quarter of 2026. That means public investors are underwriting a year-end revenue run-rate above $110 billion on second-quarter data - a gap of more than a full quarter between what the prospectus will show and what the market is pricing. The first-day volatility will come from that gap, not from the headline numbers, which are already widely reported. It also means the first post-IPO earnings release will be a binary event: a print that confirms the trajectory extends the premium; a print that misses it forces an immediate repricing because there is no private-market bid to absorb the shock.
The Counter-Thesis: Why the Bulls May Be Right
The strongest case against my skepticism is straightforward and deserves weight. Anthropic is the first pure-play frontier AI lab available to public investors - a scarce asset in a category that may define the next decade of productivity growth. Claude's adoption is compounding, the company has already reached positive adjusted operating income, and the reported $65 billion-to-$110 billion revenue trajectory implies growth rates that public-market software investors have not seen at this scale in a generation. Prospective investors are already signaling a fair-value range of $1.8 trillion to $2 trillion; the demand is real, not theoretical, and scarcity value alone can sustain a premium for longer than fundamentals would justify.
Bulls also argue that the capex is not a burden but the moat itself. In a world where model quality is a function of compute, the lab that has already contracted the capacity wins the next generation. The $518 billion in obligations, on this view, is the price of a durable competitive position - and if revenue merely keeps pace, the fixed-cost base generates operating leverage that private markets never got to price. The counter-thesis is backed by the most relevant authority in the room: the institutional investors being courted at the Oct. 14 event, who are reportedly willing to underwrite a $2 trillion valuation. They have the capital, the mandate, and the conviction.
This case rests on one assumption - that revenue growth holds - rather than on a mechanism that works in both directions. That is not a weakness if the assumption is right. It is a concentration of risk, and the market is being asked to pay a premium for that concentration.
The falsifying signal is quantifiable. If Anthropic's reported revenue fails to approach the $110 billion year-end trajectory while its non-cancellable obligations remain in place - specifically, if quarterly revenue growth decelerates below the pace needed to double the run-rate within twelve months - the structural-premium thesis breaks and the valuation re-rates toward the private-market anchor of $965 billion. That is a specific, observable threshold: revenue trajectory versus committed fixed costs. It can be checked against the first two public quarterly prints, and it does not require a judgment call about model quality or market share.
What Comes Next: Beneficiaries, the Exposed, and the Signals to Watch
The mechanism cashes out into a clear asymmetry. In the base case - revenue holds near the reported trajectory and the IPO prices in the lower half of the $1.8 trillion-to-$2 trillion range - the beneficiaries are the frontier-AI lab itself, the cloud providers carrying its obligations, and the semiconductor supply chain that has priced sustained capex growth. The exposed are public-market investors who buy at the top of the range on the assumption that private-market marks are a reliable anchor; they are not. Private marks are set in negotiated rounds with preferred terms and downside protection. Public prices carry none of that protection and all of that liquidity.
Split by time horizon, the paths diverge. In the short term, sentiment and IPO-window liquidity will dominate, and a successful debut could lift the entire AI complex regardless of fundamentals - this is the momentum leg, and it is real while the window stays open. In the medium term, fundamentals take over: the next two quarterly revenue prints will matter more than the debut pop, because they test whether the growth story survives contact with public disclosure. In the long term, the structural question resolves on whether AI compute demand keeps compounding faster than the capital required to serve it - and on that question, the fiber-optic analogy is the warning, not the shale analogy, because compute, like bandwidth, becomes cheaper and more abundant over time, which is good for adopters and bad for suppliers who overcommitted at peak prices.
Three signals decide which path the story takes. First, the pricing: a debut at or above $2 trillion validates the structural-premium read; a debut below $1.5 trillion signals that public markets are discounting the capex burden. Second, the first post-IPO revenue print against the reported $110 billion trajectory - the single most important number in the story. Third, whether the hyperscalers revise their own capex guidance after the offering - the second-order transmission that tells you whether the market is rewarding or punishing the buildout.
Scenarios: the base case is a successful listing in the $1.6 trillion-to-$1.8 trillion range with first-day volatility, followed by a range-bound first year as quarterly revenue prints test the growth story. The upside case is revenue acceleration past $110 billion, which re-rates the stock toward $2 trillion and pulls the AI complex higher with it. The downside case is a growth deceleration that forces the market to price the $518 billion obligation stack as a liability rather than an asset - a re-rate back toward the $965 billion private anchor, with spillover into hyperscaler and semiconductor multiples.
Data as of Oct. 13, 2026. Reported Anthropic figures are based on disclosures ahead of a public filing; the draft registration statement remains confidential and all valuation ranges reflect prospective-investor estimates reported ahead of the investor day. PSP figures are in Canadian dollars as of March 31, 2026.
Anthropic's IPO will not just price a company - it will price the market's verdict on whether AI's capital intensity is the moat or the trap. Right now, the price assumes the moat. The next revenue print decides which it is.
Explore more exclusive insights at nextfin.ai.
