Everything About AI
NextFin News - Anthropic, the maker of the Claude AI models, publicly filed its IPO prospectus with the Securities and Exchange Commission on Monday, and the numbers inside redraw the map of what the artificial-intelligence boom costs. The company reported $16.29 billion in revenue for the six months ended June 30, up from $1.16 billion a year earlier and more than triple the $4.59 billion it generated in all of 2025, alongside an $80.7 billion net loss that is mostly an accounting artifact of its own soaring valuation rather than cash it spent. The filing sets up a listing on the Nasdaq under the ticker "ANT" that could raise $100 billion or more and value the five-year-old company at as much as $2 trillion — a test of whether public markets will fund an AI arms race whose economics depend on demand compounding faster than the bill for compute capacity.
The central tension is not growth. That is undeniable. It is whether the business can scale fast enough to service a compute backlog that the filing and related disclosures put at more than $400 billion in locked-in commitments, with total potential spending reaching $750 billion in the coming years. Anthropic is no longer a private company burning venture capital in the dark. It is about to become the first frontier AI lab priced by public markets, and its S-1 makes plain that the industry's defining question has shifted from whether these models can do useful work to whether revenue can catch the infrastructure.
The Numbers Inside the Filing
Anthropic's top line is the story bulls will lead with. Revenue in the first half of 2026 was $16.29 billion, roughly 14 times the $1.16 billion recorded in the same period of 2025. The second quarter alone produced $11.56 billion, up from $787 million a year earlier, indicating that growth accelerated through the first half rather than decelerating. Consumption-based services — API and enterprise usage billed by volume — accounted for $13.1 billion, or about 80% of first-half revenue, with subscriptions contributing the remaining $3.2 billion.
The loss is the number that will dominate headlines, and it requires decoding. The $80.7 billion net loss for the first half compares with a $13 billion loss in the same period a year earlier. But roughly $72.4 billion of it came from non-cash fair-value adjustments on convertible notes and other financing instruments — accounting charges triggered primarily by the increase in Anthropic's own valuation. The operating loss, a closer proxy for the actual business, narrowed to $693 million from $2.14 billion a year earlier. The company is far closer to operating profitability than the statutory loss suggests.
The spending behind that operating loss is where the structural story lives. Anthropic recorded nearly $9 billion in computing and infrastructure expenses in the first six months of 2026, up from $2.3 billion a year earlier and already ahead of the $7.3 billion it spent across all of 2025. The company has signaled it could spend up to $750 billion on additional computing capacity in the coming years, including deals still under negotiation. The prospectus lays out computing agreements totaling around $430 billion over the next seven to ten years, the majority of which cannot be canceled.
On the balance sheet, the company is not cash-poor. As of June 30, Anthropic had raised $141.1 billion in total and held $112.4 billion in cash and cash equivalents plus $19.3 billion in short-term investments. That liquidity is what makes the $750 billion commitment conceivable. It also means the company will need to return to capital markets repeatedly, which is precisely what this IPO is for.
Efficiency Is Improving. Absolute Dollars Are Still Compounding.
The single most important metric in the S-1 may be the one that does not appear as a headline figure: compute cost as a share of revenue. In the first quarter of 2025, Anthropic spent $2.50 on compute for every dollar of revenue. By the second quarter of 2026, that had fallen to 48 cents on the dollar. Revenue generated per dollar of compute expenditure rose to $2.08 in the second quarter of 2026 from $0.40 in the first quarter of 2025, according to analysis of the filing by Baird.
That is the efficiency thesis in one ratio, and it is real. Model improvements, pricing power in agentic workloads, and scale economies are pulling unit economics in the right direction. An investor buying ANT is buying the proposition that this curve keeps bending.
But efficiency gains and absolute capital intensity are not mutually exclusive. They are happening at the same time, and the second is the harder problem. A business that spends $9 billion on compute in six months while generating $16.3 billion in revenue covers its compute bill about 1.8 times over. To justify a $2 trillion valuation, that coverage ratio must not merely hold; it must expand while the denominator — the infrastructure base — grows by hundreds of billions of dollars. The S-1 discloses that roughly 20% of first-half revenue came from just two customers, and the top five customers accounted for about 25% of revenue in the nine months ended September 30. Concentration like that makes the efficiency ratio more fragile than it looks. Losing or renegotiating with one of those customers does not just dent revenue; it leaves committed compute capacity idle.
This is the mechanism that separates Anthropic from the classic software company. Traditional software scales with near-zero marginal cost. Frontier AI scales with a marginal cost that is contractual, upfront, and non-cancellable. The S-1 converts what had been a narrative about AI demand into a balance-sheet obligation: the company has effectively pre-committed to a capital expenditure path that assumes demand keeps compounding. If demand pauses, the cost does not.
Public Shares, Private Control
Anthropic is going public without giving up control, and it has engineered the filing to say so explicitly. The company is incorporated as a Delaware public benefit corporation, a structure it says reflects its commitment to AI safety. A Long-Term Benefit Trust will have the authority to appoint a majority of the board of directors — four of seven — through a special Class T common stock that carries no economic rights. The seven co-founders, including CEO Dario Amodei and President Daniela Amodei, will retain 50.1% of voting power on general shareholder matters after the offering.
Public investors will own a claim on cash flows but not on direction. That is consistent with the company's founding thesis: the technology it is building is consequential enough to require insulation from quarterly earnings pressure. It is also a hard sell to the public-market investors who will be asked to fund the buildout. The filing notes that Anthropic has contributed about 847 million non-voting shares, representing 5% of fully diluted shares as of June 30, to a subsidiary dedicated to its public-benefit purpose.
Who Funds the Buildout
The S-1 reads like a map of the AI capital web. Alphabet, the parent company of Google, has a financing agreement that could provide up to $20 billion in additional funding, reducible by as much as $10 billion if Google exercises an option to purchase nonvoting Anthropic stock in a private placement alongside the IPO at the offering price. The agreement carries a 14.99% ownership cap that Google can waive. No funds had been drawn as of June 30.
Amazon holds approximately $6.2 billion in outstanding convertible notes that could become roughly 1.84 billion nonvoting shares when the IPO completes, and has agreed to provide up to $15 billion in financing in three $5 billion tranches. SpaceX, through its own filing, disclosed a deal under which Anthropic will pay $1.25 billion a month for the output of its Colossus data centers through May 2029 — more than $40 billion if the contract runs its full term.
These are not passive investors. They are customers, lenders, and infrastructure providers rolled into one. The arrangement reduces Anthropic's near-term financing risk but deepens its dependence on a small group of counterparties whose interests do not always align with public shareholders. A hyperscaler that also sells cloud capacity has little incentive to see Anthropic's margins expand too far, too fast.
The Market's First Read
The filing landed in a market already nervous about AI capital intensity. On the day, AI-related names and neocloud providers came under pressure, with CoreWeave sliding as investors weighed what Anthropic's spending disclosures imply for the rest of the chain. Bitcoin slipped 0.3%, and AI tokens were little changed — Bittensor's TAO up 1.2%, Render up 2.2%. The risk is not same-day price action. A raise of this size will need buyers, and the same growth-oriented capital that holds crypto and AI tokens may need to rotate to make room for ANT.
Valuation expectations are already elevated. Anthropic was last valued at $965 billion in its $65 billion Series H round in May, having surpassed OpenAI's $852 billion March valuation. Polymarket traders put the most weight on a first-day closing value of $1.75 trillion to $2 trillion, and assign a 74.5% probability that Anthropic will be worth more than Bitcoin by the end of the year. The underwriters — Morgan Stanley, Goldman Sachs, and JPMorgan — are expected to target a listing in the week of November 16.
Cyclical or Structural: A Regime Shift, Not a Cycle
The right way to read the S-1 is as evidence of a structural regime change, not a cyclical spending wave. Three things separate this from a normal capital expenditure cycle. First, the commitments are contractual and non-cancellable — around $430 billion locked in over seven to ten years — so the spending cannot be pulled back if demand softens. Second, the financing is structurally embedded: convertible notes with hyperscalers, revenue-sharing data-center deals, and a governance trust designed for a multi-decade mission rather than a product cycle. Third, the technology itself runs on a capability treadmill; a frontier lab that stops spending does not hold its position, it falls behind.
There is a cyclical leg, and it matters for timing. AI equity valuations have run ahead of realized earnings across the sector, and a softer macro environment or a pause in enterprise AI adoption could compress multiples before fundamentals break. That leg can mean-revert. The structural leg — the permanent elevation of capital intensity at the frontier — cannot.
The Counter-Thesis, and What Would Break It
The strongest argument against this reading is that the efficiency trend already solves the problem. If compute cost per revenue dollar can fall from 250% of revenue to 48% in five quarters, why can't it fall to 20% or 10%? On that view, the $750 billion commitment is not a trap but a rational pre-positioning ahead of demand that efficiency gains will more than cover.
The answer is that the efficiency metric measures the installed base, while the commitment measures the future base. The 48% figure reflects revenue earned on chips already deployed and paid for. The $750 billion is spending on capacity that must be filled by revenue that does not yet exist. The two numbers live in different time periods, and the risk is the gap between them.
The falsifying signal is specific. If compute spending as a percentage of revenue rises for two consecutive quarters while revenue growth decelerates below 50% year over year, the structural-efficiency thesis is wrong and the commitments become a burden rather than a moat. Watch the quarterly filings for that crossover.
What Comes Next
Short term, the stock's path will be set by the offering mechanics — the price range, the share count, and whether the $100 billion raise clears — plus the earnings reports from Alphabet, Amazon, and Microsoft, which will show how much more the hyperscalers are willing to spend on the AI buildout. A strong quarter from the cloud providers would validate Anthropic's demand assumptions; a pullback would tighten the window.
Medium term, the key data points are the ones in the S-1 itself: the concentration of the top five customers, the consumption-versus-subscription mix, and whether the operating loss continues to narrow toward zero. The company has more than 11,900 customers each generating at least $100,000 over the trailing twelve months, including more than 1,900 generating at least $1 million — a base that must broaden if the concentration ratio is to come down.
Long term, the investment case rests on whether frontier AI becomes a general-purpose platform with compounding demand, or a capital-intensive utility with compressing returns. The S-1 does not answer that question. It simply puts the stakes in dollars: $16.3 billion in revenue today, $750 billion in commitments ahead, and a $2 trillion valuation that requires the first to outrun the second.
"So it's hard to put a number on it, but I don't think $2 trillion is crazy. If the S-1 shows margins after compute, moving in the right way, $2 trillion starts to look like the opening bid, not the ceiling," said Carmel of Manhattan Venture Partners, an early investor in Anthropic and SpaceX.
Anthropic's IPO is not a bet that AI will grow. It is a bet on the shape of that growth — and whether revenue can compound faster than the infrastructure bill that precedes it. The S-1 shows a company winning the growth race and still racing the clock.
Explore more exclusive insights at nextfin.ai.
