Anthropic's Draft Prospectus Shows a $42bn Loss, OpenAI Seeks $30bn and the FTC Opens an AI Safety Probe
Title: Anthropic's Draft Prospectus Shows a $42bn Loss, OpenAI Seeks $30bn and the FTC Opens an AI Safety Probe
Date: 5 October 2026
Type: Blog
Author: SAASiQ (contact@saasiq.ai)
Word count: 1518 words
Reading time: 6 min
Published: 05-10-2026
A draft of Anthropic's listing prospectus, seen by Reuters on 28 September, shows 2025 revenue of nearly $4.6bn and a net loss of $42bn. The same document shows Anthropic has committed at least $518bn to computing over the next decade. OpenAI is in early talks to raise at least $30bn at a valuation of about $1.4trn, Bloomberg reported on 29 September. The US Federal Trade Commission has opened an investigation into OpenAI, Anthropic and the evaluation group METR. AMD agreed to buy World Labs, Fei-Fei Li's company, for about $8.2bn in shares.
Anthropic's draft prospectus
Reuters reported on 28 September that it had seen a draft of the registration statement Anthropic submitted confidentially to the US Securities and Exchange Commission on 1 June. SiliconANGLE says the version seen was prepared in June. The document has not been made public, and the figures could change before a public filing.
Revenue in 2025 was nearly $4.6bn, twelve times the year before. The operating loss was more than $8bn, excluding writedowns of liabilities. The net loss was $42bn. Fortune reports that nearly $34bn of that is an accounting charge on financing instruments that can convert into shares.
Computing and infrastructure cost $7.33bn in 2025, three times the 2024 figure. That was more than half of total operating expense of $12.65bn. Two customers supplied nearly a quarter of 2025 revenue. The risk section notes that many large clients are not tied to long-term contracts.
Reuters counts at least $518bn of computing commitments over a decade, with six partners. About 80 per cent of that cannot be cancelled or is payable whether the capacity is used or not. Google accounts for at least $111.1bn between April 2026 and July 2033. Amazon accounts for $110bn between May 2026 and April 2036, and Microsoft for $31.4bn between November 2026 and May 2033. Broadcom equipment leases come to about $161.2bn.
SpaceX is down for up to $84.5bn through 2029, for capacity built on Nvidia chips. SpaceX's own listing papers had put the agreement at nearly $45bn, or $1.25bn a month for three years. So the figure in Anthropic's draft is close to double.
The listing timetable and the Broadcom financing
Bloomberg reported on 2 October that Anthropic's roadshow could start on 9 November, with trading before the US Thanksgiving holiday on 26 November. An investor day is planned at its San Francisco headquarters on 14 October. Morgan Stanley, Goldman Sachs and JPMorgan are leading the offering. Prospective investors see $1.8trn to $2trn as a fair value, according to Bloomberg. Reuters had said the listing could value Anthropic at more than $2trn. Its last private round, a $65bn Series H in May, valued it at $965bn.
The draft prospectus says Broadcom agreed to make up to $42bn of financing available for the infrastructure. On 2 October Bloomberg reported that Broadcom's banks have begun raising $60bn of debt to pay for AI chips for Anthropic and other customers. Bank of America, Citigroup and Morgan Stanley are due to send out letters for a $42bn senior secured tranche. Blackstone is leading an $18bn junior tranche, committing $9bn from its own funds and selling the rest to other lenders. Bloomberg describes the deal as a test of investor appetite at a time of public opposition to data centre construction.
OpenAI's $30bn round
OpenAI is seeking at least $30bn at a valuation of about $1.4trn before the new money, Bloomberg reported on 29 September. The talks are at an early stage and the terms could change. The round would come before a stock market listing. Sam Altman said earlier in September that OpenAI would not list in 2026.
OpenAI's previous round, $122bn at a valuation of $852bn, closed in March, according to Bloomberg and TechCrunch. Reuters reported on 18 September that OpenAI's own projections show negative free cash flow of $278bn between 2026 and 2030. The new round follows OpenAI's decision on 25 September to pause training of its latest models, which SAASiQ covered on 28 September.
The FTC investigation
The New York Post first reported the FTC investigation on 30 September, and Reuters, Axios and Semafor followed the same day. It covers OpenAI, Anthropic and METR, a non-profit in Berkeley that tests AI models. The FTC declined to name any other companies involved.
The legal basis is possible unfair or deceptive practices under the FTC Act, or a failure to keep data reasonably secure. According to Axios, the FTC chair, Andrew Ferguson, is preparing civil investigative demands, which work like subpoenas, to obtain documents and testimony from executives. These are expected within weeks. The questions cover product safeguards, known risks, what consumers are told, internal testing, incident response and whether the labs' public safety claims are accurate. Axios reports the inquiry began over the summer.
Washington, Canberra and Sacramento
On Sunday 4 October President Trump named Jay Clayton, the Director of National Intelligence, as the White House AI adviser, NPR reported. Clayton keeps his intelligence role. He will lead a new body called the Super Intelligence Force, set up by executive order. Its first report is due within 120 days. Clayton told the Wall Street Journal it will also cover how the government handles disclosure of security breaches. Ferguson of the FTC is one of three vice chairs, with Emil Michael, the defence department's chief technology officer, and Scott Kupor, director of the Office of Personnel Management. Clayton is a former chair of the SEC.
In Australia, Altman and Dario Amodei did not appear at the Senate inquiry hearing in Canberra on 1 October. Both companies said the notice was too short, as the invitations went out on 27 September, Reuters reported. OpenAI's chief strategy officer, Jason Kwon, is due before the parliament's Joint Select Committee on Artificial Intelligence in Sydney on 6 October. Anthropic told Bloomberg it would send executives from the US and Australia.
In California, Governor Gavin Newsom signed SB 947, the No Robo Bosses Act, on 30 September. Employers may not rely only on an automated decision system to dismiss or discipline a worker. A person has to check the decision against other information, such as manager evaluations, peer reviews and personnel files. The worker must be told in writing that AI was used, what data it used and who to contact. The law takes effect on 1 July 2027.
AMD buys World Labs
AMD agreed on 28 September to buy World Labs for about $8.2bn, paid entirely in shares, Bloomberg and CNBC reported. World Labs builds models that generate interactive 3D environments, used to train robots and other physical AI systems. Its first product is called Marble. Fei-Fei Li, its co-founder, becomes AMD's executive vice president and chief scientist, reporting to the chief executive, Lisa Su. Her co-founders Justin Johnson and Ben Mildenhall will continue to run World Labs.
It is AMD's second-largest acquisition, after Xilinx, which was worth about $50bn when it closed in 2022. The deal needs regulatory approval and is expected to close by the end of 2026.
Chips, buybacks and other deals
Oracle has agreed to lease about 100,000 AI chips to Tencent for five years, the Financial Times reported on 30 September. The chips sit in several Oracle data centres in south-east Asia. The deal is worth about $7bn, with about 30 per cent paid upfront, and is Tencent's largest overseas computing deal. It gives Tencent the use of Nvidia processors that cannot be sold into China. The chips stay outside China, which US export rules allow for overseas cloud leasing.
Nvidia added $150bn to its share buyback authorisation on 28 September, leaving $235bn available to use through its 2028 financial year. Nvidia called it the largest increase to a buyback authorisation in history. Its shares rose about 2 per cent that day.
Salesforce signed an agreement on 29 September to buy Listen Labs, a San Francisco company that uses AI to run customer research. No price was given. Business Insider had earlier reported talks at about $2bn. The deal is expected to close by 31 January 2027.
Among the week's funding rounds, ElevenLabs, the voice AI company, completed a $300m share sale by employees at a valuation of $22bn, tech.eu reported on 30 September. That is double the $11bn valuation of its February round. Instinct, which makes an AI agent, raised $1bn at a $10bn valuation from Sequoia, Benchmark and Coatue, TechCrunch reported on 28 September.
For buyers
In SAASiQ's view, the questions the FTC is asking also suit any organisation buying AI services, and the answers can be written into the contract. They cover how a supplier tests its models, how it handles incidents and whether its public safety claims are accurate.
Anthropic's draft shows that about 80 per cent of its computing commitments cannot be cancelled. Its risk section also says many large clients are not on long-term contracts. Multi-year AI contracts set prices and capacity for years ahead, and the supplier's own fixed costs sit behind those terms.
Kwon appears before the Sydney committee on 6 October, and Anthropic holds its investor day on 14 October. Bloomberg reports that its roadshow could begin on 9 November.
SAASiQ - Intelligent Solutions for SaaS ©
