Anthropic Files S-1 Again, Taking on an IPO
Here are my thoughts after thoroughly digesting the filing:
There’s no need to dwell on the positives; what we should worry about are the risks lurking beneath the tip of the iceberg.
Beyond the multiples in revenue and customer growth, here are several deeper issues and potential risks in Anthropic’s S-1 filing that have drawn significant attention from capital markets and industry experts:
1. Massive “compute commitments” (up to $750B / $518B+)
The filing reveals that, to maintain its model training and inference capabilities, Anthropic has signed long-term cloud-computing contracts worth hundreds of billions of dollars, including:
* Major suppliers: Long-term contracts have been signed with Google (approximately $111 billion), AWS (approximately $110 billion), and Microsoft (approximately $31 billion).
* Non-cancellable terms: More than 80% of computing capacity contracts are “non-cancellable” or “payable regardless of usage” agreements. The company must pay these enormous capital expenditures regardless of how market demand changes in the future.
2. Massive Book Losses and “Non-Cash Accounting Adjustments”
* Net losses in the first half of 2026 reached as much as $80.7 billion, of which approximately $72 billion was attributable to non-cash accounting adjustments (primarily changes in the accounting valuation of financial instruments arising from past financing arrangements).
* In terms of actual operations, computing capacity and infrastructure operating expenses alone came to nearly $9 billion in the first half of 2026 (already exceeding the $7.33 billion for all of 2025), showing that the rate at which large language models are burning through cash is accelerating.
3. The Paradox of Cloud Giants Being “Both Partners and Competitors”
The Risk Factors section of the S-1 filing specifically lists:
* Anthropic is highly dependent on computing resources from cloud service providers such as AWS and Google Cloud.
* However, these cloud giants are also investing heavily in developing their own AI models, exposing Anthropic to potential supply chain risks, including the possibility that access to computing capacity could be restricted, repriced, or terminated.
4. Special Equity Governance Structure (PBC and LTC)
* Public Benefit Corporation (PBC): The filing explicitly warns investors that the company’s mission of “responsibly developing and safely deploying AI” may, in certain circumstances, lead it to make decisions that prioritize AI safety over shareholders’ short-term interests.
* Multiple-class share structure: The founding team and the “Long-Term Trust” committee hold shares with greater voting rights (such as Class F and Class E), while the voting rights of publicly issued Class A shares are limited. Public-market investors have very little actual control over the company’s direction.
5. Customer Concentration and Insurance Risks
* High customer concentration: As of 2026, approximately 20% of revenue came from the two largest customers. A large enterprise customer switching providers could significantly hurt revenue.
* Catastrophic AI Risk: The filing states that if a major model safety incident or extreme misconduct occurs, the coverage limits of existing commercial insurance may not be sufficient to cover related damages or legal liabilities.