Anthropic's prospectus reveals massive computing power commitments to tech giants

A rapid climb to the top

Anthropic has experienced a period of explosive growth that many observers describe as rare in human history. According to its prospectus, the company's revenue jumped from $41 million in 2023 to $45.85 billion in 2025. This trajectory suggests a move toward becoming a dominant force in the global AI market.

The company's momentum accelerated significantly in the first half of this year, with revenue reaching $162.86 billion. This represented a 13-fold year-on-year increase. Even as the growth rate began to stabilize in the third quarter, the company continued to show outstanding performance compared to previous periods.

User data further supports this rise in market recognition. Anthropic reported that the number of customers contributing over $10 million reached 250 by September 30. Additionally, the company now serves 40 million customers who contribute less than $100,000, showing a broad reach across different scales of business.

The cost of staying ahead

To maintain its technical lead, Anthropic has entered into massive computing power agreements with several industry leaders. The prospectus highlights a total of roughly $6.37 trillion in global computing commitments, with Anthropic securing over $500 billion from seven major entities. These deals ensure the company has the hardware necessary to train and run its models at scale.

Broadcom stands out as the primary source for these resources, with a five-year financial lease for equipment valued at approximately $161.2 billion paid every quarter. Other significant deals include a $114.5 billion service agreement with Google and a $110 billion commitment from Amazon. These arrangements are often structured as

Complex financing and control

Because Anthropic lacks the immediate cash to cover these massive costs, it uses sophisticated financial tools like Note Purchase Agreements and convertible bonds. For example, its $15 billion Amazon deal only unlocks funds as specific computing milestones are met. This allows the company to secure infrastructure without draining its current cash reserves of $112.4 billion.

To prevent these massive investors from seizing the company, Anthropic uses a multi-layered share structure. While Class A shares provide economic benefits, the founders hold Class F shares, which grant them 50.1% of the voting power. Other classes exist for specific purposes: Class N for conversion channels, Class E for employee rewards, and Class T for the power to appoint or remove board members.

These legal maneuvers allow the company to scale while keeping the original team in charge. However, some analysts suggest this creates a situation where the entire industry might become trapped in a cycle of self-funding and mutual dependency among the largest tech players.

Comparing valuations and rivals

The current market landscape for AI leaders is highly competitive, as seen in the recent public offering of SpaceX. Driven by the promising prospects of AI technology, SpaceX's valuation soared to $1.77 trillion US dollars when it went public in June of this year. This high bar sets a significant benchmark for Anthropic as it prepares for its own listing and seeks a favorable valuation from investors.

While Anthropic has surpassed its primary rival in several metrics, it faces a shrinking technological edge. Recent data suggests that the performance gap between top AI models from China and the United States has narrowed to about 3%. This trend suggests that the era of uncontested dominance by Western models may be coming to an end as international competitors catch up.

The rise of low-cost Chinese models

Chinese models are presenting a significant challenge to Anthropic's commercial strategy due to their high cost-effectiveness. On the AI model evaluation platform LiveBench, the DeepSeek V4.1 Flash model currently holds a top-ten global ranking. Its overall performance is already comparable to the leading models produced by Anthropic and OpenAI, making it a formidable competitor in the global market.

The pricing of these models is particularly aggressive compared to Anthropic's offerings. For instance, the DeepSeek Flash V4.1 1017 model offers input costs of $0.15 per million tokens and output costs of $0.60 per million tokens. These extremely low prices could impact Anthropic's ability to maintain its market share as more users look for affordable alternatives to high-end models.

Other Chinese models also demonstrate a significant price advantage over Anthropic's flagship products. The Kimi K3.1 Pro offers input and output prices that are only half as expensive as Anthropic's Opus 5.5. Anthropic has attempted to address these concerns by cutting prices on its smaller models, but these reductions are still very reserved. The company faces a situation where it cannot easily reach the extremely low price points of its Chinese rivals due to its own cost structures and business logic.