Written by Ralph Sun
The lower valuation some investors are discussing for Anthropic’s prospective IPO is not proof that artificial-intelligence demand is slowing. It is evidence that the market is becoming less willing to award one laboratory the whole future of AI.
Reuters reported on October 15 that some current and prospective Anthropic investors were discussing a valuation of about $1.5 trillion for a prospective IPO, versus a prior expectation of roughly $2 trillion. The distinction matters. This is not a completed $500 billion reduction in a publicly observed company valuation, and it is not a transfer of $500 billion into Meta. It is a change in what some investors are prepared to pay for Anthropic’s future cash flows at a potential share sale.
That repricing came despite continued growth. Reuters reported that Anthropic told prospective investors that third-quarter revenue rose 45% sequentially to $16.8 billion and that the company was adjusted-profitable for the quarter. The problem was not a collapse in demand. It was that the sequential growth rate had slowed after revenue more than doubled from the first to the second quarter. In a market priced for exceptional growth, a lower growth rate can change the valuation debate very quickly.
The message is not that AI has stopped growing. It is that the market is asking a more demanding question: how much of that growth belongs to any one frontier-model provider after the hyperscalers put their own latest models into the field?
Competition is arriving through distribution, not only model quality
Meta’s recent release is Muse Wonder and Muse Wonder Full, not a product called “AI2C Muse.” Meta says the models are built for personal and professional agentic tasks and are available through the Meta Model API and Muse Code, with broader consumer use planned through Muse subscriptions.
Google’s Gemini 4 Argon is a different kind of competitive pressure. Google says Argon is designed for long-horizon coding, enterprise knowledge work and cyber defense. Its rollout is initially limited to trusted cyber defenders and early testers, with paid API customers and Google AI Ultra subscribers expected later.
Neither launch proves that Meta or Google will take a fixed amount of revenue from Anthropic or OpenAI. Model releases do not convert automatically into usage, pricing power or durable profit. But they widen the credible set of suppliers for consumers, enterprises and developers. That changes the economics even before market shares can be measured precisely.
The strategic advantage of Meta and Google is not only another capable model. It is distribution. Meta can place agents inside large consumer surfaces. Google can combine models with search, cloud, productivity software and infrastructure. Their ability to subsidize experimentation from existing cash flows also changes what a standalone lab must spend to protect its position.
This is why a lower prospective Anthropic IPO valuation can coexist with a healthy AI market. The market may be reallocating the expected profit pool across more credible competitors, more distribution channels and more ways to monetize models.
The category is expanding even as share is fragmenting
There is meaningful evidence that the user base is broadening. Sensor Tower projects time spent in generative-AI apps will rise from 17.2 billion hours in the first half of 2025 to 36 billion hours in the first half of 2026. It also reported that ChatGPT’s cross-web-and-mobile “True Audience” share fell below 50% in March as Google Gemini and Claude gained traction. At the same time, ChatGPT reached one billion monthly active users in May.
That is the core point. A falling share for the leader does not mean fewer users for the category. It can mean the category is reaching more use cases, more geographies and more distribution points than one company can own.
The early pattern looks less like a single product replacing another and more like a market separating into different jobs. Consumers may use one assistant inside a social platform, another for search and productivity, and a third for coding or enterprise work. Developers can choose models by capability, price, latency, context length, security controls and cloud relationship. Enterprises can use more than one model at the same time.
This is constructive for AI adoption. It is less automatically constructive for a valuation that assumes a single model developer will capture an exceptional proportion of industry profits.
Anthropic’s valuation debate is about the share of the profit pool
It is tempting to describe the gap between $2 trillion and $1.5 trillion as value that “went to Meta.” That makes a strong headline, but it is not a defensible valuation bridge.
Meta and Alphabet are diversified public companies. Their equity values reflect advertising, cloud, search, capital expenditure, regulation and the broader market, not only their AI models. Anthropic is a private frontier-model company with a different cost structure, funding model and prospective IPO profile. No transaction has moved a defined $500 billion from one balance sheet to another.
The more accurate argument is subtler. Each new credible model lowers the probability that Anthropic or OpenAI will command an outsized share of future AI revenue and operating profit. If investors lower that probability, they may lower the multiple they are willing to pay even if total AI revenue continues to grow.
That is not an AI demand slowdown. It is a move from a scarcity premium toward a competition-adjusted premium.
Bigger demand does not remove the coming capital test
The reassuring part of the story is adoption. The uncomfortable part is capital intensity.
Reuters reported that Anthropic committed to spend at least $518 billion over a decade on cloud services, computing and infrastructure. It also reported a $42 billion 2025 loss in the company’s IPO prospectus, even as the business pursued rapid revenue growth. The exact economics of every lab will differ, but the pattern is clear: the race is expensive before it is reliably profitable.
Gartner estimates that AI providers will spend nearly $1 trillion on AI infrastructure in 2026 and another $6 trillion between 2027 and 2030. Its forecast says more than $3.5 trillion of spending could be exposed to impairment by 2030 if investment continues on the current path and revenue does not support acceptable returns. Gartner’s forecast is not an observed outcome, but it captures the central economic risk: infrastructure deployment can outrun the returns generated by applications and services.
That risk should not be confused with a forecast that Anthropic, OpenAI or another named leader will fail. It does, however, make industry consolidation plausible. Some players may be recapitalized, acquired, merged, pushed into smaller niches or forced to cut investment. Formal bankruptcy is one possible outcome in a capital cycle; it is not the only one, and there is no evidence to support a precise claim that exactly one or two mega AI companies must fail.
The distinction matters for investors. A healthy technology adoption cycle can still produce poor outcomes for companies that overbuild, misprice compute or lack a durable distribution advantage. The internet did not slow because many early internet businesses failed. Demand grew while ownership and profit pools consolidated.
What would change this view
Three developments would challenge the “bigger cake, smaller slice” thesis.
First, a reacceleration in Anthropic’s revenue growth and a clear improvement in the capital required to serve each incremental customer would support a higher standalone profit share. Second, broad releases of Muse Wonder and Gemini 4 Argon would need to produce measurable user adoption and commercial traction, not only strong demonstrations or benchmark claims. Third, category-level engagement would need to keep expanding. If usage plateaus while providers continue to spend aggressively, the valuation concern becomes a genuine demand concern.
For now, the more balanced reading is that AI is entering its competitive phase. More capable models are reaching more people. The total market can grow rapidly while the expected payoff for each frontier lab becomes less certain. That is not a reason to dismiss the AI buildout. It is a reason to stop treating category growth and single-company valuation as the same bet.
