Anthropic is preparing its IPO with $16.3 billion in revenue over six months

Anthropic's stock prospectus reveals $16.3 billion in half-year revenue, an $80.7 billion net loss, and strong growth for Claude.

$16.3 billion in revenue in six months, $80.7 billion in net losses, and over $131 billion in cash and short-term investments. The prospectus filed by Anthropic with the Securities and Exchange Commission (SEC) gives a new financial dimension to the company behind Claude. In preparation for its initial public offering on the Nasdaq under the ticker symbol ANT, the American laboratory is making public its financial statements, the scale of its compute investments, the growth of its enterprise customers, and a capital structure designed to preserve founder control.

The transaction details remain to be finalized. Anthropic has not yet disclosed the number of shares offered or the price range for its IPO. The valuation assumptions mentioned in the press therefore do not constitute officially established terms in the Form S-1. 1,300% growth in the first six months of 2026

Revenue growth is one of the most striking metrics in the prospectus. Anthropic generated $16.286 billion in revenue in the first half of 2026, compared to $1.163 billion during the same period in 2025, representing an increase of approximately 1,300%.

Comparison with prior fiscal years illustrates the acceleration of its business. In 2023, Anthropic reported $41 million in revenue. This figure rose to $386 million in 2024, and then to $4.585 billion in 2025. The first six months of 2026 thus represent more than three and a half times the revenue of the entire previous year.

The second quarter of 2026 accounts for a particularly significant portion of this growth, with $11.555 billion in revenue, compared to $4.731 billion in the first quarter. Business thus more than doubled between the two periods.

The company attributes this trajectory to the succession of new generations of Claude models, increased usage, and the integration of its tools into enterprise applications. Its business model relies on both subscriptions and usage-based pricing for consumed intelligence, particularly through APIs. An $80.7 billion net loss that does not reflect operating losses

The most dramatic figure in the document is the net loss. Anthropic posts $80.748 billion in losses for the first half of 2026, compared to $12.969 billion a year earlier.

However, this data requires an essential accounting distinction. Of this total, $72.388 billion reflects fair value adjustments on convertible debt and other financial instruments. These adjustments are primarily linked to the increase in the company's valuation, which alters the carrying value of certain obligations.

The net loss should therefore not be equated to an equivalent cash outflow. Operationally, the financial statements present a very different picture: the operating loss for the first half of 2026 stands at $693 million, compared to $2.135 billion in the same period of 2025.

The quarterly breakdown is even more revealing. Following an operating loss of $1.667 billion in the first quarter, Anthropic recorded $974 million in positive operating income in the second quarter of 2026. This quarterly profit does not mean the company has become permanently profitable, but it demonstrates that revenue growth outpaced operating expenses during this period. Nearly nine billion dollars spent on compute in six months

Model training and execution remain the laboratory's largest expense item. In the first half of 2026, Anthropic spent $8.945 billion on compute and infrastructure, compared to $2.334 billion a year earlier.

These expenses cover several activities: training and improving models, the resources required to serve users, capacity leased from cloud providers, as well as associated storage, networking, and equipment.

As a percentage of revenue, they represent approximately 55% of semi-annual revenue, compared to 201% a year earlier. This shift reflects revenue growth that significantly outpaced compute costs over the period.

Anthropic utilizes multiple hardware architectures, notably AWS Trainium chips, Google TPUs, and NVIDIA and AMD GPUs. The company describes a strategy of managing these resources as a shared capacity, allocated among research, training, and commercial use according to operational needs.

The prospectus also notes long-term infrastructure and compute commitments. Several specific dollar amounts remain listed as unpopulated fields in this preliminary version, preventing the determination of a final contractual total from the filed document. More than 11,900 customers spend at least $100,000 per year

Anthropic's growth relies heavily on enterprise clients. As of September 30, 2026, the laboratory reports having more than 11,900 customers generating at least $100,000 in trailing twelve-month revenue each.

Among them, more than 1,900 customers reach $1 million annually, while approximately 250 customers exceed $10 million. These thresholds correspond to the revenue realized by Anthropic from each organization, not the total budget those clients allocate to artificial intelligence.

The prospectus reports a net revenue retention rate of approximately 900% as of September 30, 2026. This metric compares the revenue generated by the same cohort of customers year-over-year, accounting for expansion, contraction, and churn. A 900% rate means that revenue from the qualifying cohort is roughly nine times higher than during the baseline period. This is not a 900% customer retention rate, nor is it a growth rate that will necessarily be replicated in subsequent years.

Anthropic also highlights changes in the economic productivity of its workforce. Trailing twelve-month revenue per employee exceeds $6.6 million, compared to $1.4 million a year earlier. This ratio describes the revenue intensity per person employed, without directly measuring individual employee productivity.

The company specifically cites Claude Code and Claude Cowork as key adoption drivers within organizations. Its prospectus notes that adoption of Claude Cowork, initially released as a research preview, progressed faster than that of Claude