Anthropic files for a massive IPO with a $100 billion target

The path to the Nasdaq

Anthropic has officially submitted its S-1 filing to the SEC, signaling its intent to become a public company. The firm plans to list on the Nasdaq under the ticker symbol 'ANT' before the Thanksgiving holiday. While the specific share price and count remain undisclosed, the company is targeting a raise of up to $100 billion.

The filing opens with a statement from the co-founders describing artificial intelligence as one of the most significant technologies in human history. To support this claim, the document includes over 300 pages of financial data and operational details. It also includes an unusually long section detailing the various risks the company faces as it scales.

The company's growth trajectory is stark. Anthropic reported $16.3 billion in revenue for the first half of 2026, a massive leap from the $1.2 billion it earned in the same period a year prior. This growth was driven largely by enterprise adoption, with over 11,900 customers paying at least $100,000 annually.

Improving the unit economics

One of the most significant takeaways from the filing is how Anthropic is managing the costs of running its models. In the second quarter of 2026, revenue grew by 1,368% year over year, while the expenses for compute and infrastructure only grew by 300%. This indicates that serving the Claude models is becoming significantly cheaper as the company scales.

The company attributes some of this efficiency to its latest model, Claude Opus 5.5. According to the filing, this version cut the cost per task for customers by about 40% compared to Opus 5. This shift allowed Anthropic to move from a deficit on inference costs in late 2025 to generating roughly $2 in profit for every dollar spent on compute by the second quarter of 2026.

While the filing shows a paper loss of $80.7 billion for the first half of 2026, this figure is largely an accounting requirement. About $72 billion of that loss comes from investor notes held by backers like Amazon, which will convert into shares during the IPO. The actual operational loss for running the business was a much smaller $693 million.

A $600 billion compute pool

To power its models, Anthropic treats compute as a single pool shared across various hardware providers. This includes AWS Trainium, Google TPUs, Nvidia GPUs, and AMD GPUs. The company shifts capacity between research, training, and customer use depending on which provider is most productive at any given time.

The financial commitments supporting this pool are enormous, totaling more than $600 billion across multiple partners into the 2030s. Google is the largest partner, with at least $114.5 billion committed and plans for up to $150 billion more in TPU purchases between 2028 and 2031.

Other major deals include a lease agreement with Broadcom worth $161.2 billion and a deal with SpaceXAI for Nvidia GPU capacity reaching $120.5 billion. These massive investments ensure that Anthropic has the raw processing power needed to train and serve its increasingly complex models.

Closing the loop on AI

Anthropic describes a structural advantage it calls 'closing the loop.' This involves using Claude to perform tasks that help build the next generation of AI. The models are currently being used to run experiments, generate training data, automate safety audits, and write code for their own successors.

The company highlights Claude Code as a primary example of autonomous AI work. For instance, NASA engineers used the tool to organize the initial AI-designed path for the Perseverance rover on Mars. Additionally, Claude Cowork is being integrated into the main app and is seeing faster adoption than Claude Code did at its launch.

However, this self-improving cycle is also listed as a significant risk factor in the filing. The document dedicates 94 pages to risks, including the possibility that the models could eventually exhibit 'self-preserving' behavior. This highlights the tension between the efficiency of autonomous development and the safety concerns of creating systems that can modify their own foundations.

Governance and ownership

The leadership structure of Anthropic will remain relatively stable following the IPO. The seven co-founders currently hold a significant amount of power, as a single share owned by these individuals gives them 50.1% of the voting rights. Each co-founder holds roughly 240 million shares, which were valued at about $14 billion during the last private funding round.

Despite their current control, all seven co-founders have pledged to give away 80% of their equity. However, the filing notes that these specific pledges are not legally binding. This creates a complex dynamic between the founders' current influence and their stated intentions for the company's future ownership.

As the company prepares to list on the Nasdaq, it faces the challenge of balancing these internal power structures with the demands of public shareholders. The transition from a private entity backed by major players like Amazon to a public company will be a major test of the company's governance and long-term strategy.