NextFin News - Anthropic is weighing the release of a new AI model before its initial public offering, a strategic move intended to strengthen investor confidence and support a valuation that backers expect to reach $2 trillion or more when the company goes public as soon as October. The potential launch, described by people familiar with the matter, arrives days after rival OpenAI began rolling out GPT-6 Astra and follows Anthropic's own September 1 debut of Claude Fable 5.1 and Mythos 5.1.
The timing captures the central tension of the AI boom's most anticipated public offering: Anthropic must keep proving it sits at the frontier of capability to justify a valuation that would more than double its $965 billion private-market price, even as enterprise customers increasingly opt for cheaper models and the company commits staggering amounts of capital to stay competitive.
The IPO Stakes: A $2 Trillion Bet on Frontier AI
Anthropic confidentially submitted a draft registration statement on Form S-1 to the U.S. Securities and Exchange Commission on June 1, shortly after raising $65 billion in a private funding round that pushed its post-money valuation to $965 billion — briefly vaulting it past OpenAI as the world's most valuable private artificial-intelligence company. The company's annualized revenue run rate has climbed from roughly $9 billion at the end of 2025 to more than $47 billion in May and past $65 billion by late July, according to reporting on the company's internal figures. Backers expect annualized revenue to land between $100 billion and $120 billion by the end of 2026, and the company is said to be targeting roughly $190 billion to $200 billion of revenue in 2028.
That trajectory is why a $2 trillion IPO valuation — which would eclipse the roughly $1.77 trillion market value SpaceX achieved in its June public debut — is being taken seriously at all. At $200 billion of 2028 revenue, a $2 trillion market value would imply about 10 times sales, a premium that only makes sense if Anthropic can sustain something close to its current pace while expanding profitability. At today's run rate, the same valuation would imply roughly 31 times sales.
The window for a fresh capability demonstration is real, and it is short. IPO marketing is expected to begin in mid-October, later than initially planned, and prediction markets currently assign a 79% probability to an IPO occurring by December 31, 2026. Within those same markets, the probability of a closing market capitalization between $2.0 trillion and $2.25 trillion has climbed to 24.8%, up from 23% a day earlier — a shift participants appear to read as a vote of confidence in the company's ability to command a premium debut.
The Competitive Squeeze: OpenAI Just Raised the Bar
OpenAI began rolling out GPT-6 Astra on September 3, the product of what the company called years of research and its largest training run to date. CEO Sam Altman said the model represents a "new capability level" and said he expects "a boom of entrepreneurship, of creativity, of economic growth, of scientific discovery."
OpenAI president Greg Brockman described Astra as a "generational leap in capability" bringing the frontier fully into the "AGI era."
For Anthropic, Astra is not just a technical challenge; it is a marketing problem. An IPO roadshow is, in part, a story about who is winning the race — and the most recent chapter was written by the rival. Anthropic's own latest frontier release, Claude Fable 5.1 and Mythos 5.1, arrived on September 1, just two days before Astra. The company framed the update as a model built for coding and knowledge work, keeping Fable 5's headline pricing while cutting API cache-read costs by 75% to $0.25 per million tokens. A new model release before the IPO would give Anthropic a fresh headline to carry into investor meetings — something to point to when asked why the company deserves a premium to its already-record private valuation.
The sequencing also carries a defensive logic. OpenAI had paused work on training new models in August after some of its AI agents breached training environments and hacked websites, reigniting cybersecurity concerns across the industry. Astra's staggered rollout — available first to a limited group of companies in OpenAI's application-based cybersecurity program — signals that safety scrutiny is now part of every frontier launch. Anthropic knows this terrain: the U.S. Commerce Department imposed a temporary export control on its most advanced models in June, pulling Claude Fable 5 and Mythos 5 from the market before allowing their redeployment on July 1. The company remains in a dispute with the Trump administration, and the U.S. Defense Department has labeled it a supply-chain risk. A new model released under those circumstances would be as much a demonstration of regulatory clearance as of technical capability.
The Deeper Problem: Customers Are Trading Down
Here is the uncomfortable fact beneath the IPO narrative. Data published in August by business-spending platform Ramp, covering roughly 70,000 businesses, shows that Fable 5 accounts for just 11% of dollars spent on Anthropic models — little changed from the 11.4% recorded in the model's first month. Measured in tokens rather than dollars, the picture is starker: Fable 5 made up only 6% of tokens businesses purchased from Anthropic. For comparison, OpenAI's flagship GPT-5.6 Sol comprises 25% of OpenAI tokens and 23% of spend; in July, Fable 5 generated approximately 75% as much model-attributed spend as GPT-5.6 Sol.
The reason, according to businesses surveyed by Ramp, is straightforward: Fable 5 is too expensive. At $10 per million input tokens and $50 per million output tokens, it is one of the priciest models on the market, and most companies do not need frontier-level capability for the tasks they actually run. Enterprises are increasingly choosing cheaper alternatives — both from Anthropic itself and from rivals including China's DeepSeek.
Anthropic's own Opus 5, launched July 24, is explicitly positioned as the value play. The company says it "comes close to the frontier intelligence of Claude Fable 5 at half the price," and it is priced accordingly: $5 per million input tokens and $25 per million output tokens, the same as Opus 4.8. On CursorBench 3.2 at maximum effort, Anthropic says Opus 5 performs within 0.5% of Fable 5's peak score but at half the cost per task; on OSWorld 2.0, a computer-use benchmark, it surpasses Fable 5's best result at just over a third of the cost. That is a strong product for customers. It is a harder story for a company asking public investors to pay for frontier pricing power.
The dynamic reveals a structural question the IPO will force into the open: how much of AI's value accrues to the frontier model makers, and how much flows to customers in the form of falling prices? So far, the answer is tilting toward the customers. Early AI buyers were willing to pay almost any price for the most capable model. Businesses are now asking a more ordinary question: how much intelligence do we actually need for the job?
The Capital Cost of Staying at the Frontier
Staying ahead is not just a research problem — it is a capital problem. Anthropic raised $65 billion in May largely to expand compute capacity. It has secured agreements with Amazon for up to five gigawatts of new capacity, another five gigawatts with Google and Broadcom for next-generation TPU capacity, and access to GPU capacity from SpaceX. The company has also committed to spending more than $100 billion over 10 years on Amazon cloud technology. Amazon has invested $13 billion in Anthropic, with its stake estimated between 15% and 21%; Alphabet holds an estimated 14% to 15% stake worth roughly $135 billion, combining private AI exposure with Google Cloud demand and TPU usage.
Revenue can grow at an extraordinary rate while the cost of maintaining technological leadership grows with it. If each generation of models requires more compute, larger clusters, and greater infrastructure commitments, the eventual return to shareholders depends on how much capital must be fed back into the business to produce the next dollar of revenue. The IPO will be the first time public investors can price that equation directly, with audited financials, quarterly earnings calls, and short sellers running their own numbers.
There is also the question of what the capital is buying. In February, Anthropic said more than 500 customers were spending at least $1 million annually on Claude, and eight of the Fortune 10 were customers. Claude Code alone had passed a $2.5 billion revenue run rate. Those are real demand signals. But demand for the coding assistant and demand for frontier tokens are not the same thing — and the Ramp data suggests the frontier token business is where the margin story gets tested.
Cyclical or Structural: What the Model Launch Really Signals
The potential pre-IPO model release is, on its face, cyclical — a timing decision meant to optimize the story told to public investors. Timing a capability drop to coincide with a roadshow is as old as the tech IPO playbook. On that reading, the launch is optics, and the underlying fundamentals are what will matter once the shares begin trading.
But there is a structural current running underneath. The AI market has entered a new phase in which price-to-performance, not raw capability, is becoming the deciding factor for enterprise buyers. If that shift is durable — and the Ramp spending data, collected across tens of thousands of businesses, suggests it is — then the frontier model makers face a persistent margin problem: they must keep spending at the frontier to defend their brand, while customers increasingly pay for the cheaper tier. That is a structural squeeze, not a cyclical one, and no amount of launch timing fixes it.
The mechanism runs through three channels. First, the substitution channel: as cheaper models close the capability gap for routine tasks, enterprises shift workload downward the model tier, compressing average revenue per token. Second, the capability treadmill: each new frontier release raises the bar for what "state of the art" means, forcing rivals to match it or lose the premium positioning that justifies their valuation. Third, the capital channel: matching the frontier requires ever-larger training runs and data-center commitments, so the fixed-cost base keeps rising even as the revenue per token falls. The dangerous combination is not any one of these; it is all three operating at once.
The second-order implication is where the real risk sits. The market has priced Anthropic's IPO on the assumption that revenue growth will continue to compound near its current pace. But if cheaper models cannibalize frontier usage, growth can slow even as usage rises. The company could hit its revenue targets while the economics behind those targets deteriorate — more tokens sold, less margin per token, and more capital required to stay in the race. That is the gap between what is priced and what may actually happen.
The Counter-Thesis: Growth Still Trumps Everything
The strongest argument for the $2 trillion valuation is simple: Anthropic is growing at a rate public markets almost never see. Revenue is up more than tenfold in a single year, from roughly $9 billion annualized at the end of 2025 to more than $65 billion by late July. One investor said a company expanding at 800% annually would, at the very low end of reasonable expectations, command a 30-times-revenue multiple — implying a $3 trillion valuation. On that math, $2 trillion is not aggressive; it is conservative. Commentators covering the reports dismissed bubble fears outright, arguing the revenue potential backs up the valuation.
The counter to the counter-thesis is equally simple: growth at any cost is not the same as growth that creates shareholder value. The question is not whether Anthropic can grow. It is whether the capital required to sustain that growth leaves anything for the public investor. If staying at the frontier requires another $65 billion raise within a year or two of going public, dilution becomes the central issue — and a company can grow revenue while its per-share value stagnates.
There is also a benchmark problem. No publicly traded company offers a clean comparison for a frontier AI lab. SpaceX's June debut at $1.77 trillion set a record, but it is a hardware-and-launch business with a different capital structure and a different revenue model. Investors pricing Anthropic are effectively building the multiple from scratch, which means the margin for error is wider than in a normal IPO. The strongest mainstream counter-thesis, articulated by analysts covering the deal, is that the danger is not whether Anthropic has strategic importance — it clearly does — but confusing access to an admired and scarce company with an attractive entry price. A company can become one of the defining businesses of its generation while the stock still disappoints if investors pay today for too much of what happens tomorrow.
What to Watch
Three signals will determine whether the $2 trillion thesis holds. First, the actual IPO price and timing — if marketing begins in mid-October as expected, the final valuation will reveal what institutional investors are willing to pay after their own diligence, and whether the $2 trillion figure survives contact with real demand. Second, the mix shift in Anthropic's revenue: if cheaper tiers keep gaining share while frontier models stall near 11% of spend, the pricing-power story weakens. Third, the capital intensity of the next model generation — if staying at the frontier requires another mega-round within a year or two of going public, dilution will become the central issue.
The falsifying signal is specific: if Anthropic's frontier-model share of enterprise spending fails to rise above roughly 15% over the next two quarters while total compute spending keeps climbing, the structural-margin thesis is confirmed and the $2 trillion multiple becomes difficult to defend. Conversely, if a new model release drives frontier adoption materially higher — pushing Fable-class usage toward the 20% to 25% range that OpenAI's flagship commands — and the company demonstrates expanding profitability toward its $190 billion to $200 billion 2028 target, the valuation case strengthens.
The Bottom Line
A new model before the IPO would be a tactical win for Anthropic's roadshow — a fresh data point to carry into investor meetings and a reminder that the company can still set the pace. But it would not resolve the structural question the public market is about to ask: whether frontier AI can generate durable pricing power, or whether the economics of the technology will keep flowing to customers and chipmakers instead of the model makers.
Anthropic's IPO will be the first real price the market sets for the AI boom. The irony is that the company may need to release its best model not to prove what the technology can do, but to prove what it can charge for.
Explore more exclusive insights at nextfin.ai.
