Anthropic's updated IPO filing reveals massive infrastructure debt

A growing debt mountain

Anthropic recently submitted an amended Form S-1/A ahead of its planned Nasdaq listing under the ticker 'ANT.' The updated filing reveals that the company has committed to $895.4 billion in long-term data center and compute contracts. This figure is more than double the $430 billion in infrastructure commitments disclosed in its original filing from October.

The scale of these obligations is significant because a large portion of the debt is due very soon. The amended document shows that $69.8 billion in payments are coming due within the next twelve months. This represents an increase of roughly $18 billion over the amount reported in the company's initial filing.

As reported by Emerald Book, these numbers suggest the company is in a much tighter financial position than previously thought. While Anthropic has a large cash pile, the sheer volume of these commitments creates a massive 'lease backlog' that the company cannot currently cover with its annual earnings.

Revenue vs. infrastructure costs

The company's financial health is being tested by a mismatch between its sales and its expenses. Based on Q3 revenue of $16.96 billion, Anthropic is currently annualizing at roughly $67.8 billion. This means the company's yearly intake from subscriptions and API contracts is about $2 billion short of the baseline rent due for its servers over the next year.

The filing also shows that the pace of growth is beginning to cool at a critical time. Quarter-over-quarter revenue growth fell to 47% in Q3, a sharp decline from the 144% growth seen in Q2. This deceleration matters because the bull case for the company relies on maintaining the record momentum seen in the most recent quarter.

Furthermore, the company swung back to a $567 million GAAP operating loss after reporting a profit in the previous quarter. While the headline net loss looks catastrophic, management noted that a large portion of that figure is a non-cash accounting adjustment related to the company's rising enterprise valuation.

The cost of rapid scaling

The sudden increase in near-term obligations stems from two specific business decisions. Anthropic spent $5 billion on short-term 'spot market' compute because user demand for Claude exceeded its pre-booked capacity. By leasing server space at premium retail prices, the company chose to preserve its growth narrative rather than limit users before its IPO.

Additionally, the company signed $100 billion in new long-term infrastructure commitments in the five weeks following the end of Q3. These deals were finalized just before the company closed its books for the amendment. These figures do not even include the real-world costs of building data centers, such as energy volatility and water-intensive cooling.

The company also faces a looming hardware cycle problem. Because AI server hardware becomes obsolete every three to four years, Anthropic must constantly re-commit to new hardware cycles. If revenue plateaus, the company risks paying for empty server space, which could create a 'deleveraging cascade' for its primary backers, Amazon and Google.

The high-stakes AGI bet

The entire spending plan rests on the assumption that scaling current models will eventually produce general intelligence. This investment is essentially an empirical experiment rather than a guaranteed evolution. If capability gains hit a wall, the company will still be responsible for the massive lease obligations it has already signed.

Some experts argue that current silicon-based approaches face hard limits due to thermodynamics and data availability. They suggest that the power required for true general reasoning might exceed what the planet can generate. Regardless of these theories, Anthropic is currently spending billions to see if these models can move beyond statistical prediction into true causal reasoning.

The upcoming IPO serves as a necessary capital injection to keep the machine running. Private venture capital cannot provide the tens of billions required annually for this infrastructure treadmill. The listing allows early backers to cash out and creates publicly traded stock to compensate researchers without draining the company's remaining cash reserves.