The prospectus released by Anthropic reveals a scenario where high-growth profitability coexists with massive rigid computing power contracts. Its core business model involves selling intelligent services by Token, with 80% of its revenue derived from usage-based billing. The scale of major clients has expanded rapidly, while customer concentration has declined. Currently, cloud provider channels contribute nearly half of the revenue, but channel fee rates have risen significantly. It is anticipated that once funds are abundant in the future, Anthropic will build its own computing power and official website channels. Its inference business boasts a gross profit margin as high as 85%, but due to substantial training investments and channel expenses, it has only just achieved phased profitability. The core logic is that revenue growth must outpace the rigid increase in training and inference investments. Compared to domestic large model vendors such as Zhipu and MiniMax, Anthropic significantly leads in terms of profitability and computing power burden, leveraging the intelligent premium of its SOTA performance. The computing power investment of domestic vendors as a proportion of revenue is still equivalent to Anthropic's level in the first quarter of 2025. Within ten months, Anthropic's computing power procurement commitments have surged to nearly $1 trillion, mostly in the form of rigid contracts. To maintain a healthy computing power expense ratio, revenue needs to grow by 4-5 times on the existing annualized basis. The peak of phased profitability may be diluted by the launch of new computing power. Investments from industry players such as Amazon, Google, and Broadcom are essentially deposits to secure orders, with every $1 invested yielding approximately $3 in computing power orders, forming a deeply intertwined community of shared interests. This also compels Anthropic to subsequently prioritize maintaining penetration rates and offering discounted sales to digest locked-in production capacity, squeezing the survival space for cost-effective models. Additionally, the prospectus also discloses details such as non-cash significant losses and a special equity structure with decentralized non-voting shares. Currently, Anthropic has completed the first half of the large model business, where scale helps absorb the profit and loss statement's cash-consuming demands. However, there remains uncertainty regarding whether it can achieve its revenue growth targets and digest trillion-dollar rigid liabilities in the future.
