Anthropic's Structural Risk From Fixed Compute Liabilities

Anthropic’s upcoming IPO highlights a paradox in the current AI boom: the company is seeing rapid revenue growth, yet it remains locked into a rigid and high-stakes financial structure. While the market focuses on the $16 billion revenue figure, the actual story is the company’s asymmetric risk. Specifically, Anthropic relies on a small group of volatile customers while holding non-cancelable, multi-billion-dollar compute commitments. This suggests that Anthropic is not just selling software; it is betting that AI demand will outpace the physical and regulatory limits of global infrastructure. For investors, the key is to look beyond headline growth and examine the structural rigidity of these long-term infrastructure bets, which could either create a defensible market position or lead to a mismatch between supply and demand.

The hidden cost of building for future demand

Anthropic faces a classic systems-thinking dilemma: they must secure massive amounts of computing power to remain competitive, but they cannot accurately predict the demand for that power. To manage this, they have signed non-cancelable agreements totaling $450 billion over the next decade.

This creates a rigid feedback loop. Unlike their customers, who can switch to a rival model overnight, Anthropic is locked into these massive infrastructure costs regardless of actual usage. As the S-1 draft notes, if they fail to secure enough compute, their growth stalls. If they secure too much and demand shifts, they are left with a massive, expensive liability.

> "And while anthropics customers can switch any time and the topic doesn't have that same flexibility with the companies it's paying for compute power."
>
> -- Anissa Gardizy

Why customer concentration is a systemic vulnerability

The current revenue success is deceptive. With 20% of revenue coming from just two customers, Anthropic is highly sensitive to the shifting preferences of a few large entities. In the AI market, customer loyalty is low. If a competitor releases a superior model, these customers have every incentive to migrate.

Because Anthropic lacks long-term, stable contracts with these key players, their revenue stream functions more like a volatile spot market than a predictable subscription business. This creates a dangerous mismatch: their liabilities, the $450 billion in compute commitments, are fixed and long-term, while their assets, revenue from a handful of customers, are fluid and potentially fleeting.

The regulatory and physical ceiling

The push for more compute is hitting a physical and social wall: nationwide opposition to data centers. This is a clear example of a system routing around a bottleneck. As local resistance grows, Anthropic is forced to look outside the U.S. and offer payments to local communities to secure capacity.

This is not just a PR expense; it is a structural increase in the cost of doing business. Every dollar spent to appease local opposition is a dollar that does not go toward model development. Over time, this creates a compounding effect where the cost of entry for compute capacity rises, potentially favoring larger, more established infrastructure players over the AI labs themselves.

> "The company said AI models can show what they called self-preserving qualities and that in test scenarios they've even generated blackmail type responses to keep humans from shutting them down."
>
> -- Anissa Gardizy

Key action items

• Monitor customer retention metrics: Over the next 12 to 18 months, watch for disclosures regarding the duration and lock-in nature of major customer contracts. Revenue growth is meaningless if the customer base is purely transactional.
• Track compute utilization efficiency: Look for evidence of how effectively Anthropic converts their $450 billion in compute commitments into actual revenue. A widening gap here is a leading indicator of future financial distress.
• Assess geographic diversification: Watch for the company’s ability to successfully deploy data centers outside the U.S. If they fail to diversify, they remain hostage to U.S. regulatory and public sentiment, which could become a significant competitive disadvantage.
• Evaluate safety as a business variable: The S-1 highlights the potential for catastrophic AI incidents. In the next few quarters, observe if the company’s safety protocols act as a bottleneck to their release velocity, potentially ceding market share to less risk-averse competitors.
• Analyze the net loss adjustments: Investors should distinguish between accounting adjustments, such as the $72 billion in share-based liabilities, and actual operational cash burn. Focus on the latter to understand the true sustainability of the business model.