Anthropic IPO filing shows $16.3 billion in first-half revenue and an $80.7 billion loss

Anthropic wants to go public, and its paperwork is the clearest look yet at what it costs to stay at the front of the AI race.

The company updated its filing on Nov. 2 to list on Nasdaq as ANT. The updated filing sets no price and no date. One industry newsletter, Tech Insights, put the plan at mid-November and as much as $100 billion raised. That timing is not confirmed in the filing as described.

What the filing shows

Revenue for the first half of 2026 was $16.3 billion, up from $1.2 billion in the same period a year earlier. Anthropic also reports more than $130 billion in cash and investments.

The net loss for the half was $80.7 billion. Tech Insights' reading is that almost all of it is an accounting effect tied to loans that investors can swap for shares. That explanation is the newsletter's interpretation of the filing, not an Anthropic statement we can quote.

The filing also says Anthropic may soon reach a threshold of AI-accelerated research, according to a weekly industry roundup.

The junk-rating warning

The Financial Times reported that Blackstone, one of Anthropic's largest lenders, told its own investors that the major credit rating agencies will likely give Anthropic a "junk rating" after the listing. That is a single-source report relaying a lender's expectation. The agencies have not announced a rating.

The reported reason is a mismatch. Anthropic has signed contracts for nearly $500 billion of computing power running into the 2030s. Customers have committed to only about $11 billion in future revenue. Until its income proves it can pay those bills, the agencies are expected to keep its credit rating low.

Tech Insights author Johan argues the bet is rational if you believe every white-collar company will reorganize around agentic AI within 3 to 10 years. In his telling, most firms are still at the "prove it before we move it" stage, and Anthropic's revenue will jump once they move past it. If companies keep using AI as a smart writing assistant, he writes, the investments probably make no sense at all.

AI building AI

A second document landed three days after the filing update. On Nov. 5 the nonprofit Model Evaluation and Threat Research, known as METR, published an assessment of Anthropic's development pace.

METR estimated a 1.9x speedup in Anthropic's rate of progress since October 2025, with a 95% confidence interval of 1.5x to 2.8x. It said most of the acceleration comes from AI helping build AI, not from faster compute scaling. Staff using Anthropic's best internal model self-reported being about 12 times as productive.

METR spent 11 weeks inside the company, from Aug. 12 to Oct. 29, and described its access to internal data as "unprecedented."

The caveats are heavy. Anthropic redacted much of the evidence METR considered most informative, saying the cuts were "aimed at removing confidential information that could accelerate other AI developers." METR acknowledged internal disagreement and said it could be overstating results, since the gains may be limited to measurable, well-specified tasks.

Anthropic says its engineers now ship eight times as much code per quarter as they did from 2021 to 2025. It also says it is "not there yet" on recursive self-improvement, meaning an AI system designing its own successor.

CEO Dario Amodei wrote on Sept. 12 that the industry should "slow the pace at which we improve the capabilities of AI models" so safety work can keep up. METR's report does not establish that safeguards have fallen behind.

OpenAI's cyber numbers

Anthropic is not the only lab putting out hard numbers this week. OpenAI published a write-up saying an autonomous pipeline running an unreleased cyber model scanned the latest Linux kernel in mid-October. It reproduced 1,724 of 2,308 candidate bugs and built root exploits for 948 distinct local privilege escalation flaws, at about $1,279 in API credits per exploit.

The Register reported that 588 of those were previously unknown. About 10 days in, 61 had been patched upstream and 43 had staging patches, per OpenAI's figures. These are OpenAI's own numbers, not independent findings. The company also admitted its agents' self-reports were unreliable early on: 20 of 200 claimed exploits reached root through a different bug than the one assigned.

OpenAI's own advice is to treat an unpatched low-privilege account as nearly equivalent to root against a well-funded attacker.

Separately, OpenAI's GPT-6 Cyber is restricted to vetted Daybreak Red teams. It reportedly outperformed its predecessor at finding high-severity vulnerabilities. Anthropic released FlashProve, an open-source tool for verifying which model produced an answer.

What comes next

The open questions are specific. The filing sets no price and no listing date yet. The rating agencies, per Blackstone's reported message, are expected to move after the listing. Whether $16.3 billion in half-year revenue can service nearly $500 billion in compute contracts is what the market will price once ANT has a date.