AI giants face trillion dollar infrastructure hurdles
New IPO filings reveal massive future spending commitments that challenge the profitability assumptions behind current valuations.
By Aiko Sato ·
Anthropic is preparing for a public listing with a valuation estimated between $1.3 trillion and $1.4 trillion, a figure that has drawn intense scrutiny from financial analysts.
The core of the debate centers on the company's disclosed financial obligations. Its recent securities filings reveal approximately $900.9 billion in purchase commitments and lease agreements, primarily for computing infrastructure. Of this total, roughly $451.6 billion is expected to be incurred before the end of 2030, while the remainder extends through 2045. About $70.7 billion is due within 12 months of September 30, 2026. These figures represent significant future liabilities, though they are distinct from immediate debt.
Investors are questioning how the company will generate sufficient revenue to cover these costs. Anthropic’s annualized revenue run rate reached approximately $81 billion in October, a substantial increase from previous estimates. However, revenue is not profit. The company must still cover expenses for research, personnel, electricity, and facilities. A trillion dollar valuation requires the assumption that the firm will eventually produce extraordinary profits, a belief that is not yet proven by current cash flow.
The economics of serving users present further challenges. A single high end graphics processing unit can cost around $20,000. If such a unit serves 15 to 20 light users paying $20 monthly, it takes 50 months to recover the hardware cost alone, excluding other infrastructure and operational expenses. Even at hypothetical wholesale prices, the break even point remains difficult to reach without high utilization rates. This suggests that subscription revenue alone may not guarantee profitability for every customer segment.
Nvidia, the dominant supplier of AI hardware, stands to benefit from this infrastructure race regardless of which software platform leads. The company has secured massive supply arrangements for large technology firms, creating a barrier to entry for smaller developers. In September 2026, Nvidia announced an agreement to acquire Hugging Face for approximately $12.93 billion. This move is strategically significant because Hugging Face is a primary distribution hub for open models. Critics argue that controlling both the hardware and the distribution channel could limit competition, although Nvidia promises the platform will remain open.
The rise of open weight models adds another layer of complexity. Projects from various global tech firms have demonstrated that advanced AI capabilities can be run on private infrastructure. This reduces dependence on centralized commercial APIs. For businesses and developers, the ability to deploy models locally changes the economic landscape. It offers an alternative to expensive monthly subscriptions and raises questions about the long term sustainability of premium pricing for proprietary systems.
Quantization technology further supports this shift by reducing the memory requirements for large models. A model with 200 billion parameters might require 400 GB of memory at high precision but only 100 GB at lower precision. This reduction makes it feasible to run complex models on cheaper hardware. As these tools become more accessible, the assumption that customers will remain dependent on centralized services for years may prove optimistic.
The situation mirrors earlier technological booms where the underlying innovation was real, but the financial valuations were exaggerated. While the technology is advancing rapidly, the path to profitability for the largest players remains uncertain. Investors must weigh the potential for massive growth against the heavy costs of maintaining the infrastructure required to deliver it. The coming years will determine whether the current valuations can be sustained by actual profits or if they will face a correction.
Key open questions remain regarding unit economics and market competition. Specifically, how much profit does each customer generate after serving costs? Will regulatory bodies intervene to ensure fair access in the hardware and distribution markets? And can proprietary models maintain their price advantage against freely available alternatives? These factors will define the next phase of the AI industry’s development.
