Amazon capital and the capacity struggle
Amazon agreed to invest up to $25 billion in Anthropic this month, following an $8 billion investment in recent years. This capital arrives as Anthropic manages infrastructure strain from growing enterprise and consumer demand. Anthropic committed to spending more than $100 billion on Amazon Web Services technologies over the next decade, including current and future generations of Trainium chips. The company operates at a $380 billion valuation. While this funding expands capacity, the company still manages reliability issues that surfaced after the Fable 5 launch in June. The Commerce Department gave Anthropic 90 minutes to pull Fable 5 access from every foreign national just three days after the model launched. The massive scale of Amazon’s $25 billion investment aims to solve the inevitable strain on infrastructure that occurs when enterprise demand for high-reasoning models exceeds the current capacity of centralized power grids and existing data centers. Anthropic secured up to 5 gigawatts of capacity for training and deploying its Claude AI models. This comes as the four largest hyperscalers collectively target $700 billion in capital expenditures this year. Of the 12 gigawatts of new capacity operators promised in the U.S. for 2026, only about 5 gigawatts are actually under active construction in the United States.
The API market and token costs
The market for frontier models faces a pricing collapse as Chinese open-weight models undercut the leaders. DeepSeek V4 Pro prices output at $0.87 per million tokens, which costs 1/57th of the price for Fable 5. The shift to usage-based pricing reflects the reality that Chinese open-weight models from providers like DeepSeek have closed the capability gap enough to force frontier labs to compete on price alone. Anthropic released the 5.5 series to maintain its position.
| Model | Input (per MTok) | Output (per MTok) |
|---|---|---|
| Haiku 5.5 (up to 100K) | $0.10 | $0.50 |
| Sonnet 5.5 | $2.00 | $10.00 |
| Opus 5.5 | $4.00 | $20.00 |
| Fable 5.1 | $10.00 | $50.00 |
The API pricing for the 5.5 generation varies by task requirements. Claude Opus 5.5 costs $4 per million input tokens and $20 per million output tokens. Users can also use the Opus 5.5 Fast Mode tier, which charges $8 for input and $40 for output per million tokens. The model uses adaptive thinking where users set an effort parameter from low to max. Claude Sonnet 5.5 sits at $2 for input and $10 for output. Claude Haiku 5.5 costs $0.10 for input and $0.50 for output on prompts up to 100,000 tokens. This new generation uses a tokenizer that produces 30% more tokens for the same text compared to the 4.6 series. All current models support a 1 million token context window.
Enterprise deployment and the OpenAI rivalry
Claude Enterprise targets organizations that need governance and compliance. The plan requires a minimum of 20 seats at $20 per month per seat, which users pay annually. Users also pay separate API rates for token usage. This differs from the Claude Team plan, which costs $25 per seat annually for groups of 5 to 150 members. Claude Enterprise provides SCIM, audit logs, custom data retention controls, and a Compliance API. It also provides single sign-on, domain capture, role-based access controls, and analytics APIs for tracking adoption. You already know that token costs dictate which model wins the workload. OpenAI competes with its GPT-5.6 family, where the Sol tier costs $5 for input and $30 for output per million tokens per user. This rivalry intensifies as Amazon’s $25 billion investment follows the $50 billion agreement made with OpenAI. Anthropic reported that its annual revenue run rate exceeded $30 billion in early April 2026. Will the regulatory pressure on US-based frontier models eventually force a permanent decoupling of global AI infrastructure?
