Anthropic IPO: S-1 Finances, RSI and Existential Danger
From a US$2tn float and US$42bn Broadcom compute deal to Claude's recursive self-improvement, Anthropic's SEC filing reveals the sheer cost of frontier AI
TL;DR
- Anthropic plans to list on Nasdaq under the symbol ANT, with a potential valuation exceeding US$2tn.
- The company reported US$16.3bn in revenue for the first half of 2026, but also significant operating losses and a history of net losses since its founding.
- A substantial portion of revenue comes from a limited number of customers, posing a risk.
- Anthropic has entered into a significant equipment lease with Broadcom, allowing for up to US$42bn in convertible notes to cover costs.
- The firm acknowledges that its advanced AI systems could pose "catastrophic or existential risks to humanity" if misused or allowed significant autonomy.
- Anthropic uses recursive self-improvement (RSI) to train its AI models, which accelerates innovation but also carries risks of compounding flaws and outpacing safety measures.
- Corporate governance is structured with multiple classes of stock to prioritize responsible AI development.
- Anthropic faces potential intellectual property claims, including a settled class action lawsuit for US$1.5bn regarding the use of pirated books for training.