The hypertrophied growth in revenue and valuations of OpenAI and Anthropic is just the tip of the iceberg. Beneath the surface lies a colossal dependence on access to computing power, data center financing, and regulatory decisions. These companies are at a point where scaling their business directly runs into physical and political constraints.
OpenAI's CFO clearly outlined the correlation: from 2023 to 2025, the company's computing power grew from 0.2 GW to 1.9 GW, while annual recurring revenue rose from $2 billion to over $20 billion. Computing has been identified as the most scarce resource. Anthropic, in turn, raised $65 billion in Series H at a $965 billion valuation, explicitly stating that the funds would go toward expanding capacity for Claude. Meanwhile, the startup's annual revenue has already exceeded $47 billion.
Infrastructure Race: Trillions of Dollars at Stake
Scaling advanced models is becoming a capital-intensive task for the entire chain, from chip suppliers to data center operators. Goldman Sachs analysts predict that the four largest hyperscalers—Meta, Microsoft, Amazon, and Alphabet—will spend $5.3 trillion on capital expenditures from 2025 to 2030. The Bank for International Settlements goes even further, warning that the five largest cloud providers will spend over $1 trillion on AI infrastructure in 2025–2026.
The key problem: these commitments are already outpacing the companies' profits and free cash flow. The market is betting on future monetization, but if expectations are not met, a sharp reduction in funding could collapse the entire investment cycle. Private markets, infrastructure funds, and real estate are becoming new sources of capital, but this increases systemic risks.
The Regulatory Sword of Damocles
Beyond the financial burden, companies face sudden restrictions from authorities. Anthropic has already encountered a halt in access to its Fable 5 and Mythos 5 models due to a directive from the U.S. government under export control regulations. The company linked the complaints to a possible bypass of protective mechanisms, but details remain undisclosed.
The Donald Trump administration asked OpenAI not to release GPT-5.6 to the public immediately due to security concerns. Although the restrictions were later lifted, this episode demonstrates the fragility of relations between AI developers and the state. Anthropic's conflict with the Pentagon, which labeled the startup a "supply chain risk," only exacerbates the situation. CEO Dario Amodei stated that the company would prefer not to cooperate with the defense department rather than undermine democratic values.
IPO as a Mirror of Uncertainty
Despite all the risks, both companies have filed confidential IPO applications. OpenAI on June 8, and Anthropic a week earlier. The terms of the offerings have not been determined, but for investors, the key question will not be valuation but the economics of scaling: how much does it cost to sustain user growth, how quickly do data centers pay off, and how stable are funding channels.
My comment: The AI market is in a classic hypergrowth trap: to maintain momentum, trillions must be invested, but each new dollar of investment increases dependence on external factors. Regulatory risks are becoming an integral part of the infrastructure, and any political decision can instantly shift the balance of power. An IPO under such conditions is not so much a celebration as a stress test.