Anthropic Files S-1 for Nasdaq IPO Under Ticker ANT as AI Begins Building AI

Anthropic has formally filed its S-1 prospectus with the U.S. SEC, setting up a Nasdaq listing under the ticker ANT. The document's opening line sets the tone: "For the first time in human history, intelligence became something humanity could build with its own hands."

It is the first time the company behind Claude has laid its books, its R&D pipeline, and its ambitions open for the world to read. And the filing describes a different contest than the usual one. In September, more than 90% of the code contributions in Anthropic's own codebase came from Claude, and much of the work on the next generation of Claude is already being done by the current one. Alongside that, Anthropic just posted its first profitable quarter: Q2 revenue of $11.555 billion, 14.7x year over year, on a GAAP basis.

AI Is Building AI: 90% of Code Now Comes From Claude

Inside Anthropic, much of the work of building the next Claude has already been handed to the current one. Claude runs experiments for researchers, generates and processes training data, runs automated evaluations and security audits, and writes and reviews code. Researchers use the stronger Claude to find ways to improve compute efficiency faster; that efficiency funds a stronger model; the stronger model then does the next round of work. Anthropic calls this "closing the loop," and calls it one of the most important structural advantages in frontier AI.

The loop spins fastest in the engineering org. On open-ended engineering problems that used to require a senior engineer's judgment, Claude's success rate on internal benchmarks climbed by about 50 percentage points over nine months, reaching roughly 90% in September. The job of engineers has changed: they increasingly set direction and review results rather than typing code line by line. In Q3, per-engineer daily code contributions to the codebase ran at more than 45 times the 2024 daily average, and by September over 90% of contributions came from Claude.

From January to September, the value Anthropic derived from using Claude internally — measured at public list prices — rose by more than 250x. As early as April, that monthly figure surpassed the company's monthly payroll (excluding equity-based compensation) for the first time. Headcount grew more than fivefold between the end of 2024 and September, yet Q3 revenue alone was 86x what the company booked in Q4 2024. Revenue per employee over the trailing twelve months has topped $6.6 million, versus $1.4 million a year earlier. Anthropic says in the prospectus that it is approaching what it calls recursive self-improvement — AI making itself stronger, generation after generation.

First Profit: $11.55 Billion in a Single Quarter

When Claude first commercialized in Q1 2023, Anthropic's revenue was zero. Full-year revenue was $41 million in 2023, $386 million in 2024, and $4.585 billion in 2025. This year it accelerated hard: Q1 2026 brought $4.731 billion, and Q2 2026 reached $11.555 billion, up 14.7x year over year — a 1,368% jump. Over the same period, compute and infrastructure costs rose only 300%.

The unit economics flipped. In Q1 2025, Anthropic spent $2.51 on compute and infrastructure for every $1 of revenue. By Q2 2026 that figure had fallen to $0.48, less than a fifth. The result: Q2 delivered GAAP operating income of $974 million, after stock-based compensation and all other costs.

The headline net loss for the first half was $80.7 billion, but $72.4 billion of that was a fair-value revaluation of financing instruments — not cash burned. Over the same half, operating cash flow was a net positive $3.929 billion. As of the end of September, Anthropic's net dollar retention rate stood at roughly 900%: enterprise customers who were paying a year ago now spend about nine times as much on Claude.

Recursive Self-Improvement: The Loop That Spins Faster

Anthropic's sharpest proof point is how fast capability compounds. On the CMU-developed exploit benchmark ExploitBench, Claude's score rose from 40% on Opus 4.8 to 78% on Claude Mythos 5, and then to 91% on Opus 5.5 — in roughly four months.

Anthropic is cautious about how it ships that capability. When it trained Claude Mythos Preview, it judged the model's offensive security strong enough that releasing it at scale without new safeguards would carry major risk. It added the safeguards first, then released the model publicly as Claude Fable 5.

The Compute Contracts: SpaceX.AI, Google, Amazon, Broadcom, Microsoft

To train and run the stronger Claude, Anthropic signed and repeatedly revised a compute agreement between May and October with SpaceX.AI, a wholly owned subsidiary of Elon Musk's SpaceX, to lease NVIDIA GPUs for a total of up to $120.5 billion. Once deliveries complete in December, Anthropic will owe SpaceX.AI roughly $3.1 billion a month.

SpaceX.AI is only one line on the list. Google, Amazon, Broadcom, and Microsoft are also compute counterparties; together the five contracts add up to more than $550 billion.

The Endgame: A Data Center Full of Geniuses

Anthropic believes that, on this trajectory, "powerful AI" will be widely available within a few years. Per the prospectus, it expects systems that match or exceed top human experts across most professions and scientific fields, that can work autonomously for hours, days, or weeks, and that can directly operate lab equipment and robots. Millions of Claudes working in parallel — Anthropic's phrase is a "data center of geniuses."

Aimed at the hardest human problems, that capability could compress decades of scientific progress into a few years, accelerating drug discovery, disease detection, and personalized medicine. The seven co-founders close their letter this way: "Our goal is to help humanity safely navigate this transition, open up new frontiers of discovery, and begin an era of wonder. Everyone who holds our stock — institution or individual — becomes part of that work."

Humans built intelligence, and intelligence has begun to build the next intelligence. From tonight, that recursive path toward ASI has its own ticker: ANT.