Anthropic’s IPO Filing Signals Big Bets on AI Infrastructure Demand

Anthropic’s confidential IPO filing, announced this week, underscores a central tension in the AI infrastructure market: the enormous capital required to train and serve frontier models. Co-founder Daniela Amodei, speaking at the Bloomberg Tech conference, framed the move as a practical necessity. “It’s a really big upfront cost to train the models and to serve inference on them,” she said. “Over time, the core set of companies advancing the frontier will need access to capital, and the public market is well suited to that.”

The company’s growth has been staggering. Annualized revenue hit $47 billion in May, up from roughly $9 billion at the end of 2025. But that trajectory faces scrutiny as enterprises like Uber report uneven returns on AI spending, raising questions about whether corporate budgets will tighten. Amodei isn’t concerned. She argues businesses are still learning how to deploy AI effectively, pointing to coding, financial services, legal, and healthcare as early drivers of value. “As the business community gets more familiar with the tools, we’ll all learn together,” she said. “My hope is that over time, AI will be more incorporated into daily work, and more value will be realized.”

On infrastructure strategy, Amodei explained why Anthropic avoids building its own data centers, unlike rivals OpenAI and xAI. “We prefer to plan for the best outcome without overextending ourselves,” she said. “It’s hard to predict compute needs perfectly. We’d rather have a bit more demand than we can serve than the opposite.” That philosophy led to a surprising partnership with xAI last month, disclosed in a SpaceX filing, costing Anthropic $1.25 billion per month for compute capacity.