Everything About AI
NextFin News - Anthropic's initial public offering is no ordinary technology debut. It is a stress test for the entire artificial-intelligence boom, and the company's own filings supply the weights. The Claude developer reported revenue of $16.3 billion in the first half of 2026, up 14-fold from $1.16 billion a year earlier, and posted its first operating profit, $974 million, in the second quarter. Over the same six months it booked an $80.7 billion net loss and disclosed roughly $637 billion of compute commitments to cloud providers and chipmakers. The question this IPO forces investors to answer is not whether demand for AI is real — the S-1 proves it is. It is whether the financing structure built on top of that demand can survive a slowdown, and who pays when the bill comes due.
The timing sharpens the stakes. Anthropic, PBC filed its Form S-1 with the Securities and Exchange Commission on October 26, planning to list Class A shares on the Nasdaq Global Select Market under the ticker ANT, with Goldman Sachs and Morgan Stanley leading a 22-bank syndicate. The filing lands in the middle of a live market conversation about whether the AI buildout is a durable industrial revolution or the most expensive financing loop in technology history. What the S-1 reveals is that it is both — and that the two are now inseparable.
The Growth Is Real, and It Is Unprecedented
Start with the demand side, because the numbers are not subtle. Revenue in the second quarter alone reached $11.56 billion — more than Anthropic earned in its entire history through the end of 2025, when cumulative revenue across 2023, 2024 and 2025 totaled about $5.0 billion. Full-year 2025 revenue was $4.59 billion, an 11.9-fold increase over 2024. In the first half of 2026 the company grew 1,300 percent. Three straight years of roughly ten-fold growth — 9.4x, then 11.9x, then 14.0x — is a growth signature with no clean precedent in public-market history.
The customer base is widening as fast as the revenue. More than 11,900 customers now spend at least $100,000 a year, up from just over 950 twelve months earlier — a twelve-fold increase in the paying base. Net revenue retention stood at approximately 900 percent as of September 30, 2026. For comparison, Snowflake's celebrated IPO-year retention was roughly 170 percent; a 900 percent figure means the average customer is spending nine times as much a year later, which is what land-and-expand looks like when the product is displacing entire engineering workflows rather than supplementing them.
Enterprise adoption is the engine. About 80 percent of revenue comes from enterprise clients, including eight of the Fortune 10. The number of enterprise customers spending more than $1 million a year doubled from 500 in February 2026 to more than 1,000 by early April. And concentration — the classic IPO red flag — has actually eased rather than worsened: no single customer accounted for more than 10 percent of revenue in the nine months to September 30, 2026, and the top five together made up about a quarter. That is a sharp change from 2025, when customers spending over $100,000 represented 76 percent of revenue.
There is still a knot in the data. Two customers accounted for about a fifth of first-half revenue, and industry reports have identified them as SpaceX and Microsoft. Neither is named in the filing, and the concentration has fallen since, but the dependency is directional: if either retools its AI stack around a rival model, the revenue curve kinks.
The Loss Is Mostly Accounting — But the Cash Bill Is Still Coming
The $80.7 billion first-half net loss is the number that stops readers, and it deserves a careful bridge. About $72.4 billion of it — roughly 90 percent — came from non-cash fair-value adjustments on convertible notes, derivative liabilities and SAFEs. No cash was paid. Those instruments convert into equity, so as the company's valuation rose, the liability marking them to market rose with it. In accounting terms the company "lost" money because it became more valuable.
Strip out the fair-value noise and a different picture emerges. The first-half operating loss was $693 million, and that includes the second quarter, when operating income reached $974 million — the first profitable operating quarter in the six shown. In 2025 the company reported an operating loss of $8.06 billion on revenue of $4.59 billion. So the trajectory is from deep operating losses toward breakeven, not away from it. Management has also claimed adjusted profitability in the third quarter, though those figures are unaudited.
The real cash question sits in a different line item: compute and infrastructure. In the first half of 2026 it totaled $8.95 billion, and in the second quarter it grew 300 percent year over year — far slower than the 1,368 percent revenue growth in the same period. That gap is operating leverage, and it is exactly what a growth multiple prices in. But it also points to the structural constraint: every dollar of that revenue required a dollar of upstream compute, and the compute is not optional.
The Keiretsu: Cloud Giants as Customers, Creditors and Owners
The most revealing part of the filing is not the revenue curve. It is the web of relationships that produced it. Amazon, Google and Microsoft are simultaneously Anthropic's largest investors, its biggest cloud suppliers, and, through their marketplaces, its distribution channel. Amazon holds an estimated mid-to-high-teens equity stake worth roughly $135 billion to $160 billion at the last private valuation, and Google owns about 14 percent, hard-capped. Amazon also extended $8 billion of convertible notes plus two $5 billion preferred-stock purchases. In the first quarter of 2026 alone, Amazon booked an estimated $16.8 billion in pre-tax gains from mark-ups on its Anthropic stake.
In return, Anthropic has committed at least $110 billion to Amazon Web Services through 2033 and about $114.5 billion to Google Cloud through 2036, with a further $44.9 billion to Microsoft. Add commitments to SpaceXAI, Broadcom, Nscale and Lambda, and the disclosed total reaches roughly $637 billion. A September draft of the filing put fixed or minimum guarantees at $265.6 billion as of June 30, 2026, with another $245 billion executed since. The company has signaled it could spend as much as $750 billion on computing in the coming years.
Venture capitalist Tomasz Tunguz captured the structure in a single word: these businesses are forming a "keiretsu," each entangled in the others' equity, debt and go-to-market. The arrangement solves a collective-action problem. The cloud providers need frontier-model demand to justify their own AI data-center buildout; Anthropic needs guaranteed compute at scale and on credit. The IPO is the mechanism that lets each side book the other's commitment as an asset — the hyperscalers book a guaranteed tenant for capacity they have not yet built, and Anthropic books the capacity it needs to keep revenue growing at 1,300 percent.
This is the circularity that defines the AI boom. The cloud providers fund the very customer whose spending is then used to justify the cloud providers' own capital expenditure. Anthropic's revenue validates Amazon's and Google's AI infrastructure thesis; Amazon's and Google's infrastructure commitments validate Anthropic's growth thesis. The loop works as long as enterprise AI spending keeps compounding. It is the most elegant financing structure in technology — until demand growth decelerates.
The $637 Billion Question: Who Bears the Demand Risk?
Here lies the structural tension, and it is a mismatch of contract forms. Anthropic's revenue is consumption-based: customers pay for what they use, with no long-term lock-in. But its compute bills are largely take-or-pay. If enterprise AI spending plateaus, Anthropic still owes the cloud providers whether its customers consume or not. The risk transfer runs one way — from the hyperscalers, who secure offtake for their data centers, onto Anthropic's balance sheet, and ultimately onto public shareholders.
The math of the valuation makes the margin for error thin. At the $965 billion post-money valuation set by the May 2026 Series H round — which raised $65 billion — the company trades at roughly 20 times its $47 billion annualized run rate. Some investor discussions have floated figures near $2 trillion, though that reflects market speculation rather than a confirmed offer. To support $2 trillion, Dr. Chan Ahn, a former Goldman Sachs and JPMorgan executive, estimates Anthropic would need about $725 billion in annual revenue by 2036, assuming a 10 percent cost of equity, a 25 percent free-cash-flow margin and a 25x terminal multiple. At a 13 percent discount rate, the requirement rises to roughly $950 billion. Backers expect annualized revenue between $100 billion and $120 billion by the end of 2026 — which means the company must roughly double from there, every year, for a decade, while holding margins that no software company at scale has ever sustained.
Skeptics are blunt. Independent research firm New Constructs called the potential offering the "most ridiculous IPO of 2026," valuing the company at $150 billion — nearly 93 percent below the speculative $2 trillion figure — and wrote, "We don't think Anthropic has a viable business." It said supporting a $2 trillion valuation would require profits equal to twice Nvidia's trailing four-quarter total. Michael Burry, the investor known for betting against the subprime bubble, noted that Anthropic's valuation equals the combined market value of 78 S&P 500 companies. New Constructs also flagged competition from open-source models as an existential threat to the pricing power the valuation assumes.
The Counter-Thesis: Efficiency Gains Can Outrun the Debt
The bear case is not uncontested, and the S-1 itself supplies the strongest rebuttal: unit economics are improving faster than prices are falling. With the September 2026 release of Claude Fable 5.1, the company said it cut the cost of typical workloads by approximately 25 percent compared with the prior version, with savings of up to 45 percent for complex coding and highly agentic tasks. Opus 5.5, released the same month, reduced customers' cost per task by approximately 40 percent compared with Opus 5. Sonnet 5.5, released days later, runs about 30 percent faster and costs up to 30 percent less per task. If efficiency gains outpace price declines, margins can expand even as customers pay less per query — the classic software deflation story that made the 2010s cloud boom so profitable.
The company also argues it has crossed the profitability threshold. Second-quarter operating income of $974 million was its first profitable operating quarter in the six shown, and compute expense growing at 300 percent while revenue grows at 1,368 percent is the definition of operating leverage. Krishna Rao, Anthropic's chief financial officer, said of the company's latest infrastructure expansion with Google and Broadcom: "We are making our most significant compute commitment to date to keep pace with our unprecedented growth." The company's own guidance points to annualized revenue between $100 billion and $120 billion by the end of 2026 — a run rate that, if achieved, would make the $965 billion private valuation look like the floor rather than the ceiling.
And the competitive position is stronger than the open-source critique allows. Claude's programming models are widely regarded as the leading tools for generating complex code, and the coding-agent workflow is where enterprise AI budgets are being spent first. Once an engineering organization embeds a model into its development loop, switching costs are high — which is what 900 percent retention measures.
Control Stays Private, Even After the IPO
One more disclosure matters for public buyers. The seven co-founders will hold 50.1 percent of voting power through a single Class F share held by a Founder LLC, with no economic rights but decisive control over ordinary shareholder votes. The Long-Term Benefit Trust — whose trustees include former Federal Reserve chair Ben Bernanke, experimental physicist Christopher Stubbs, Clinton Health Access Initiative CEO Neil Buddy Shah and national-security expert Richard Fontaine — elects and can remove a majority of the board through a separate Class T share. Public shareholders buying Class A stock get financial exposure to the business; governance is permanently in the hands of the founders and the trust. The filing warns as much: the trust, the Founder LLC and the co-founders "may have interests that differ from yours and may vote in a way with which you disagree and that may be adverse to your interests."
What to Watch Next
The IPO's pricing will deliver the first public verdict. The share count, price range and raise size remain unset in the filing, but the company has said it expects to match or exceed the size of SpaceX's record-setting offering, which bankers have described as the benchmark for 2026. The roadshow is expected to begin in mid-October, with pricing targeted for the weeks before the November midterms and a listing window that some reports have pushed from October to November to allow third-quarter results into the story.
Three signals will separate the structural shift from the cyclical excess. First, revenue growth: if quarterly growth falls below roughly 100 percent year over year while compute expense keeps compounding, the growth-at-any-cost model breaks. Second, net revenue retention: a material drop from the reported 900 percent would show customers are not expanding usage. Third, concentration: the top two customers accounted for about a fifth of first-half revenue, and if either retools around a rival model, the revenue curve kinks.
The verdict splits cleanly by time horizon. Short term, the IPO is a liquidity event driven by positioning and scarcity — a $2 trillion debate is a trading range, not a valuation. Medium term, the fundamentals are real: 1,300 percent revenue growth, 900 percent retention, and operating leverage that is just beginning to show. Long term, the structure is the risk: $637 billion of commitments against consumption-based revenue is a duration mismatch that only perpetual hypergrowth can close.
Scenarios frame the range. In the base case, revenue reaches $100 billion to $120 billion annualized by the end of 2026, the company lists near or above the $965 billion private valuation, and the stock trades on a growth premium while operating margins inflect positive. In the upside case, efficiency gains compound, enterprise adoption broadens beyond the Fortune 10, and the valuation rerates toward the speculative $2 trillion as AI becomes the default operating layer for business software. In the downside case, growth decelerates toward 100 percent, open-source models erode pricing power, and the take-or-pay compute commitments turn from an asset into an anchor — forcing a repricing toward the $150 billion that skeptics argue is the fundamentals-based floor.
Anthropic's IPO does not decide whether the AI boom is real. The S-1 settles that. What it decides is whether the boom is self-funding or self-financed — and public markets are about to find out which.
"We are making our most significant compute commitment to date to keep pace with our unprecedented growth." — Krishna Rao, Chief Financial Officer, Anthropic
Explore more exclusive insights at nextfin.ai.
