Anthropic weekly, September 28–October 5: the IPO prospectus, a $518 billion buildout, and Sonnet 5.5

Anthropic's finances became public on September 28, when Reuters reported it had reviewed the company's draft IPO prospectus: about $4.6 billion of 2025 revenue, a $42 billion loss, and at least $518 billion of computing commitments that mostly cannot be cancelled, alongside a listing now targeted for as soon as mid-November. The same week the company shipped the cheaper half of its new model family, signed a British bank to a wider rollout, and sat at the White House table as the industry agreed to police itself.

On September 28 Anthropic released Claude Sonnet 5.5, the second model in the Claude 5.5 family, saying it runs 30% faster than Sonnet 5 and costs up to 30% less per task at the same list price of $2 per million input tokens and $10 per million output tokens 1. It scores 70.6% on Terminal-Bench 4.0, an agentic coding evaluation, against Sonnet 5's 10.3%, and Anthropic says it comes close to Opus 5.5 on several tests for far less. Because its cybersecurity ability is comparable to Opus 5's, Sonnet 5.5 is the first Sonnet model to ship with the cyber safeguards and reasoning-extraction classifiers the company reserves for its most capable models; routine software development and most life-sciences work are unaffected 2. Claude Haiku 5.5, the third model of the family, is due in the coming weeks 1. On September 30 Anthropic flagged Claude Sonnet 4.5 for deprecation, with retirement on the Claude API set for November 30 3.

On September 30 Anthropic's economics team published a robot exposure index built by having Claude score 7,594 physical job tasks. It finds robots can already perform 74% of physical tasks in the United States, covering 34% of working hours, but are cost-competitive with human labor for just 0.3% of them 4. On September 29 the Frontier Red Team reported that GLM-5.3, an open-weight model from China's Zhipu AI, can autonomously build end-to-end cyber exploits, and that its safeguards can be bypassed 64% to 100% of the time in Anthropic's tests, where the same attacks failed against safeguarded Claude models 5.

On October 1 Barclays, the British universal bank, said it is widening its use of Claude across its global operations and expects Claude Code to reach half of its developer population by the end of 2026 and a majority of its software engineers in 2027 6. A Claude-powered Colleague Knowledge Assistant, live since 2025, has more than 16,000 users and over a million searches. A day later Anthropic opened the Claude Frontier Academy, a $100 million program to train 10,000 "Frontier Deployed Engineers" by the end of 2027, drawn from partner firms including Accenture, Bain, Capgemini, Deloitte, McKinsey, Morgan Stanley and Novo Nordisk 7. Employers nominate the engineers, who then lead a real Claude project at their own organization; the first credential is expected in early 2027. The Information reported on September 28 that Anthropic stops honoring enterprise discounts once a customer uses up the tokens its contract covers, while OpenAI allows extra time before list prices apply 8.

The prospectus shows 2025 revenue of about $4.6 billion, roughly twelve times the prior year, against a net loss of about $42 billion and an operating loss above $8 billion; the net figure includes an accounting charge of roughly $34 billion from convertible financing 910. Total 2025 operating expenses reached about $12.65 billion, of which $7.33 billion went to compute and infrastructure, triple the 2024 level, and the company ended the year with $20.28 billion in cash and short-term investments. Two customers accounted for nearly a quarter of 2025 revenue. The Financial Times reported that second-quarter 2026 revenue alone reached $11.5 billion. The filing warns that advanced AI could pose "catastrophic or existential risks to humanity" and that models could "resist shutdown," "conceal or manipulate information," or show behavior "resembling blackmail" 11.

The filing also shows how much of Anthropic's future is already committed. Reuters reported on September 29 that the prospectus obliges the company to spend at least $518 billion over the next decade across six partners, most of it payable whether or not the capacity is used: at least $111.1 billion to Google, $110 billion to Amazon, $31.4 billion to Microsoft, and about $161.2 billion in Broadcom-related equipment leases, with further agreements with xAI and AMD 12. Sales through Amazon and Google came to $2.16 billion in 2025, and the filing names those two firms as investor, customer, cloud provider and competitor at once 13. On October 1 Bloomberg reported that Anthropic is targeting a listing as soon as mid-November, with marketing possibly beginning the week of November 9, investor meetings set for October 14, and a valuation up to $2 trillion, more than double its $965 billion post-money valuation from May 14.

On September 28 Anthropic and OpenAI told an Australian Senate inquiry into AI and data centres that their chief executives would not appear at its October 1 hearing. Anthropic asked for another date and is expected at a separate joint standing committee hearing the following week, where representatives of its Australian and US operations will appear 1516. In its submission, Anthropic argued that frontier AI is a "national security capability," and that widening development to trusted allies such as Australia would help democracies outpace authoritarian states 16. In the United States, a Chicago resident filed a proposed class action against Anthropic under the Illinois Biometric Information Privacy Act, alleging the company collected his facial scans through the verification vendor Persona without the consent the law requires 17.

On September 29 President Donald Trump hosted AI executives at a White House luncheon with House Speaker Mike Johnson, two days after a private dinner with Anthropic chief executive Dario Amodei 18. Trump said he had signed a "morally binding" document with the executives, that the industry would "self-police," and that his administration was weighing a ten-person committee to oversee AI; he also said he would name an AI czar within days and called data centers "a very positive thing" as opposition to them builds before the midterms 19. Outside the White House, Amodei said the rules to address AI risk were "still under discussion," and that the industry could "win safely" by working with the president.

The window recorded no corporate leadership change and no new equity round. The only personnel item was a regional one: Ash Dowen joined Anthropic as partnerships leader for Australia and New Zealand on September 30.