Anthropic's IPO Filing Answers a Lot. These Questions It Doesn't.

Anthropic's S-1 filing is, by most measures, the clearest window anyone has gotten into how a company that only sells AI actually makes money. More than 300 pages, filed this week, covering revenue, computing costs, competitive risk, and corporate structure in granular detail. Analysts and reporters spent days combing through it. And yet, for all that disclosure, the document leaves several of the most consequential questions about Anthropic's business unanswered.

The headline numbers are striking on their own. Anthropic reported $16.29 billion in revenue for the first half of 2026, against $2.38 billion in cost of revenue — a gross margin of roughly 85%. That's up enormously from the $1.16 billion the company generated in the same period a year earlier, according to Violeta Todorova, senior research analyst at Leverage Shares, who spoke with CNBC.

But growth has a shape, and Anthropic's shape is bending. Revenue grew about 144% quarter-over-quarter in Q2. In Q3, that growth rate dropped to roughly 47%. That's still fast by any normal company's standard, but it's a sharp deceleration, and the filing doesn't say whether management expected it, planned for it, or sees it continuing. "What the prospectus didn't tell us is management's expectations around the sequential slowdown in the growth rate," Michael Field, chief equity strategist at Morningstar, told CNBC. Todorova raised a related concern: as computing demand scales up, it's not clear how much it costs Anthropic to generate each incremental dollar of revenue.

That computing question runs through the whole filing. Anthropic says "substantially all" of its cost of revenue comes from compute and infrastructure, and warns that its margins will "fluctuate over time," especially in periods when it takes on large new chunks of compute capacity and the expenses that come with it. The company also reveals, in its own words, that it has historically operated in what it calls a "compute-constrained environment," where demand for processing power has exceeded what's available to it. At one point, Anthropic says it sought more compute from cloud providers than those partners were willing to supply.

That history of scarcity makes the forward-looking math murkier, not clearer. Nick Patience, AI lead at the Futurum Group, told CNBC that investors can see compute costs shrinking as a share of revenue, but the filing doesn't break out how much of Anthropic's future compute commitments are earmarked for training its next generation of models versus simply serving existing customers. "That split dictates future margins," Patience said. He also flagged two open questions: whether compute costs can keep falling faster than Anthropic's spending commitments grow, and whether the company can lock down enough capacity to meet customer demand going forward.

Then there's the matter of who actually controls the company once it's public. Anthropic is structured as a public benefit corporation, and its filing shows that Dario Amodei and six co-founders each hold more than 200 million shares — a stake that could be worth billions if Anthropic reaches the trillion-dollar-plus valuation it's expected to command. More notably, those seven founders hold a single Class F share carrying 50.1% of total voting power on matters put to shareholders. Anthropic itself describes this governance setup as having "no direct public company precedent."

Field at Morningstar points out that the filing also discloses plans for multiple additional classes of shares down the line. What's missing is any indication of how the market will react to that concentration of control. "What we don't know is whether investors will be fine with the lack of control implied, or will this impact the share price," he said. "Something we won't know until the shares start trading."

The number nobody can pin down yet

The biggest open question is the simplest one: what is Anthropic actually worth? The company isn't saying. The filing states only that its IPO valuation will be set through negotiations between Anthropic, selling stockholders, and underwriters' representatives. Analyst estimates currently range from $1 trillion to $2 trillion, a spread wide enough to make the eventual number a major market event either way. Land near the top of that range, and it could inject some confidence back into an AI sector that's lost some of its swagger in recent weeks. Land near the bottom, and the reaction could be rougher.

One figure buried in the filing is worth flagging on its own: Anthropic is sitting on more cash than almost any other company in the country, and no company has ever gone public holding anywhere near that much, according to CNBC's Leslie Picker.

The filing arrives alongside a flurry of other AI infrastructure news that underscores just how much capital is moving through this sector right now. Nvidia and AMD said they'll put a combined $4 billion into Meta, which just announced plans to raise $95 billion for its own AI buildout. SpaceX closed $40 billion in debt financing specifically to buy GPUs. Microsoft, meanwhile, is offering buyouts to employees as it tries to balance its AI spending against capital efficiency. Anthropic's IPO doesn't exist in isolation — it's the most detailed data point yet in an industry-wide bet that's getting more expensive by the week.

For now, the filing has given the market plenty to analyze and not quite enough to resolve. The real test comes when Anthropic's shares actually start trading and the market decides, in real time, whether the growth, the margins, and the governance structure add up to the number everyone's been guessing at.