Anthropic’s S-1, by the numbers: A thriving business—with a $441 billion asterisk

Given its central importance both as a potential investment and bellwether for the future of AI, Wall Street research crews and everyday investors alike are carefully parsing the over 400 page document. And one big concern dwarfs all others: Can Anthropic generate revenues so gigantic that they far outpace the huge spending on compute its leaders already acknowledge is coming? Evidence that’s happening would provide much comfort before folks and funds wrestle with the next and more vexing head-spinner: Will any reasonable projections for its growth in sales and operating leverage prove great enough to make Anthropic worth anything like $2 trillion?

So far, the S-1 shows surprisingly excellent bedrock performance, and boasts great financial strength. On the potential downside, it’s committed to big spending on AI capacity that it can’t cancel, meaning that the expenses are locked in while the sales are far less certain. 

Anthropic’s operating basics look strong 

Let’s start with the positive operating news. For the first six months of 2026, revenues exploded 13x from $1.3 to $17 billion over the same period last year. But a big surprise comes from disentangling the income statement. Under GAAP, Anthropic suffered a staggering loss of $80 billion in the first half of 2026. But the vast bulk of that paper deficit arose from changes in fair value of convertible notes. Back out that non-cash charge and several others including equity compensation, and Anthropic actually garnered adjusted operating income of $1.6 billion for the first six months of 2026, for a 9%+ margin. 

So far, the cash flow scenario also looks favorable. For first half 2026, Anthropic collected $3.93 billion in cash from operations. Its capex, the item that’s mushrooming at SpaceX, was so tiny this writer did a double take: Less than $500 million. As a result, Anthropic registered positive free cash flow of around $3.5 billion.

Operating leverage, the ability to expand margins by growing revenues faster than costs, is also pointing the right way. Specifically, top-line growth is outpacing the expense of the data center capacity that runs Claude, Fable, and Mythos. In Q1, sales exceeded AI infrastructure costs by 40%. In Q2, it was more than double, $11.6 billion in revenues versus $5.6 billion for compute. Customer stats are also impressive. The S-1 disclosed that Anthropic now harbors a base of 11,000 enterprises featuring over $100,000 a year in revenue, up from 1,500 last year.

The balance sheet’s super-solid: Anthropic’s holding $122 billion in cash and short-term investments, following a round that raised $65 billion in late May. The IPO’s expected to amass another $100 billion, hoisting its reserves by late this year to well over $200 billion.

The rub: Big guaranteed costs are coming, on promising but uncertain revenues

Anthropic will need that war chest, and more. The S-1 details seven contracts listed under “infrastructure service agreements” and “infrastructure equipment leases”, the roster of partners comprising the likes of Microsoft, Amazon, and SpaceX. The grand total for these deals comes to $630 billion. The key takeaway: 70% of those arrangements, by dollar value, are “non-cancellable.” Anthropic’s obliged to pay whether or not it uses the capacity it’s reserved.  

In fact, Anthropic’s widening its strategy for attaining new compute. Instead of leasing from the likes of Microsoft where their partner pays for and installs the GPUs and other gear that account for around 80% of a data center’s cost, Anthropic is now going “hybrid” by in some cases buying the chips and assuming the cost of equipping “dedicated” centers operated by others. That’s a highly capital-intensive approach that, judging by its tiny capital spending so far, is totally new.

Going forward, it’s clear that Anthropic’s cash outlays for capacity will explode. The hope is that we’ll see a continuation of the current trend where revenues jump much faster. But the approximately $441 billion in guaranteed commitments is something investors should pay attention to. Most of the contracts appear to run seven years or less. So that’s at least $60 billion annually that’s got to be covered by jackrabbit sales. Plus, Anthropic acknowledges that the capacity budget will grow from here, stating on page 197 its intention “to secure further compute beyond our existing commitments, [and] expand our own data center footprint to secure additional long-term compute.”

Other items may rankle potential fans. One of them is possible dilution in the name of doing good. The board can issue an additional 2% of the outstanding shares each time the stock appreciates 50% over a “benchmark,” presumably starting at the offering price, to the “Anthropic Public Benefit Fund,” an institution dedicated to philanthropy. So if the stock rockets, each shareholder owns 2% less of the business. And if the stock price makes a round trip, they’re worse off.

On balance, the S-1 showed that a basic business is thriving, and for now, Anthropic’s passing the main test by growing revenues much faster than the costs of compute. On the downside, it’s taking a huge, leap-of-faith wager that those sales will absolutely rampage from here. If they don’t, Anthropic will encounter the opposite problem, baked-in costs that ramp faster than revenues, a formula for disaster.

Investors also need to decide if even in the best case, Anthropic can race at the virtually never-before-seen pace that would make it worth $2 trillion out of the starting gate. As the S-1 shows, so far the vision’s inspiring, and the numbers are good. It’s the size they’ve got to grow to that looks scary.