Anthropic's $2 Trillion IPO Question: Is AI Software, or the New Utility?

The number that matters more than $2 trillion: $65 billion

Anthropic filed a confidential draft S-1 with the US Securities and Exchange Commission on June 1. That is the extent of what the company has confirmed publicly. The number of shares and the offering price have not been set. Reuters reports that a public prospectus could arrive by late September, with institutional marketing starting as early as mid-October. Even the $2 trillion figure circulating in the market is not a locked-in valuation. It is an upper-end scenario that investment banks and market participants are debating.

Still, the number did not come out of nowhere. Anthropic's internal 2028 revenue forecast, confirmed by Reuters, runs between $190 billion and $200 billion. A $2 trillion valuation works out to roughly 10 to 10.5 times that 2028 projected revenue. Measured against the company's run rate of $65 billion as of late July 2026, the same $2 trillion figure is closer to 30.8 times.

That gap is the whole story of this IPO. Investors are not buying today's Anthropic. They are pricing in, right now, how much of a company's core workflow Claude will control two or three years from now.

If Claude stays a chatbot, $2 trillion is hard to justify

Treat Anthropic as a general-purpose AI chatbot company and the $2 trillion number becomes difficult to defend. The case for it rests on Claude becoming the new gateway for enterprise spending.

If companies use Claude mainly to summarize documents and search information, it behaves like a fast-growing SaaS product. But if it becomes an agent layer that automates coding, customer support, research, legal review, security analysis, procurement, accounting and sales, the competitive landscape changes entirely. AI could then redirect spending that currently flows to legacy SaaS licenses, IT outsourcing, professional services, cloud usage fees, and even some labor costs.

So Claude's real KPI is not "how smart are the answers." What matters is how much of a customer's existing spending on people and legacy software converts into recurring, measurable AI spending.

AI's hardest problem isn't revenue. It's the cost of each additional dollar of revenue

Traditional software carries relatively low marginal cost once the product is built and customers scale up. Generative AI works differently. Training models requires massive GPU clusters and data centers, and every additional unit of usage brings inference costs along with it.

That is why the number the market will scrutinize most closely after the listing is not growth alone. What matters more is how much gross profit and cash flow each additional dollar of revenue leaves behind.

Anthropic's $65 billion run rate is strong evidence that real demand exists. But sustaining a $2 trillion valuation requires four things to hold at once: expanding contracts with large customers, a steady decline in the cost per unit of inference, continued gains in model efficiency, and rising gross margin and free cash flow.

The real inflection point for an AI company is not the moment growth eventually slows. It is whether the company can cut the growth rate of its compute costs even faster than its revenue growth decelerates.

What a $15 billion credit line reveals about the AI business model

Reuters reports that Anthropic is finalizing a $15 billion revolving credit facility ahead of the IPO. Reading this simply as "the company is borrowing because it's short on cash" misses the point of how capital-intensive frontier AI actually is.

A frontier AI company cannot add servers overnight just because demand spikes. It has to secure GPUs, data center capacity, power and network bandwidth well in advance, and large contracts require long-term funding commitments. The credit line functions as a liquidity option, a way to avoid losing customers to compute shortages when demand surges unexpectedly.

At the same time, the number is a warning. AI does not scale as lightly as consumer internet software once did. Competition over model performance ultimately becomes a competition over GPU procurement power, data center capacity, electricity access and capital markets access. That is exactly why Anthropic increasingly looks like an "AI utility" rather than a pure software company.

Five things a $2 trillion valuation demands proof of

First, revenue quality. Investors need to know whether usage spikes are temporary or reflect long-term contracts with high net revenue retention.

Second, customer concentration. If revenue is heavily concentrated among a handful of very large enterprises or cloud partners, negotiating leverage can weaken even as headline growth stays strong.

Third, gross margin. If revenue keeps growing but GPU costs surge in lockstep, the business cannot support the kind of multiple typically applied to traditional software.

Fourth, infrastructure dependency. Being overly tied to a specific cloud provider, chipmaker or data center operator exposes the business to supply disruptions and weaker pricing power, both of which can shake profitability.

Fifth, pricing power. Any performance edge over OpenAI, Google, Meta and xAI needs to translate into real pricing premiums and customer loyalty, not just technical benchmarks.

The litmus test for AI IPOs

Anthropic's IPO matters for more than the size of one company's listing. It is the first time public markets will get to rigorously test where the money in the AI investment cycle begins, and where it ends.

When a model company raises capital, that money flows into GPUs, high-bandwidth memory, networking gear, servers, data centers, power and cloud infrastructure. That spending, in turn, becomes revenue for Nvidia, memory makers, data center operators and power infrastructure companies. But for this cycle to keep going, it eventually has to close the loop: enterprise customers need to use AI to boost productivity and keep paying for it, again and again.

In that sense, Anthropic is the link connecting the AI infrastructure boom to its final demand and capital expenditure. If the IPO draws strong demand, it signals that markets are starting to assign real probability not just to "AI usage" but to "AI economics." If the offering price comes in low, or the multiple compresses quickly after listing, the question changes. It stops being about whether AI demand is strong, and becomes whether that demand is profitable enough to justify the enormous expectations already baked into the price.

Insight Times Editorial Desk