Anthropic is renting GPUs from xAI's Colossus cluster for inference as token consumption grows exponentially, highlighting a token shortage that is driving up costs and pressuring AI companies' margins.
Elon won. He's the only one with capacity to sell tokens at scale, right when the token shortage officially started. Fun detail buried in the news: Anthropic is already renting Colossus 1 from xAI for inference. Their competitor literally runs on Musk's iron. The volume is exploding. Google: 9.7T → 3.2 quadrillion tokens/month (May 2024 → May 2026). Azure Foundry: up 7x YoY, north of 500T tokens through Foundry APIs. Token consumption is growing two orders of magnitude faster than prices are falling. **The subsidies are over.** OpenAI's leaked internals project a $14B loss in 2026 on $13B revenue. They burn $1.69 for every $1 they earn, and the entire delta is inference. Anthropic's gross margin sits at \~40%, OpenAI's at \~33%. Both are below SaaS-standard 70-80%. Cursor killed unlimited (June 2025). Claude Code capped Pro and Max (August 2025). GitHub Copilot flipped fully usage-based (June 2026). The pattern repeats because the math forces it. As OpenAI and Anthropic file for IPO, they will close this gap by raising API prices. Their margin is your cost now. **As a founder shipping AI products, how do you survive?** 1. Full cost visibility. You need cost per customer, per agent, per workflow, in real time, not at month-end. If you find out which customer is unprofitable when the AWS bill lands, you're already underwater. This is exactly what we built Credyt for: real-time metering on top of LLM API spend, with per-customer, per-agent, per-workflow attribution. Without that layer, the next three rules are guesses. 2. Cost-to-value math, per segment. Your margin lives in the gap between what a customer's usage costs you and what they pull out of it. If a segment isn't getting enough value, switch them to a cheaper model or self-hosted. If they're getting a lot, raise prices. Most founders skip the segmentation step and price one number for everyone, which is how power users eat the margin. 3. Prove the value out loud. Your customers won't accept a price increase unless they can see what they're getting. Numbers, stories, before/after. Show the workflow saved 14 hours of analyst time, the support deflection saved $X, the conversion lift moved $Y. Value lives in your customer's words and metrics, not yours. Token costs are going up. Costs will land on builders. The founders who survive this will be the ones who can answer "what does this customer cost me right now" without opening a spreadsheet.
Anthropic has agreed to pay SpaceX $1.25 billion monthly ($15 billion annually) through May 2029 for access to its Colossus AI data centers, revealing the massive demand for compute capacity and the intertwined nature of AI and infrastructure companies.
Anthropic will pay xAI $1.25 billion per month for compute capacity from the Colossus 1 data center, in a deal that could bring xAI over $40 billion in revenue.
Anthropic has surged to a $1.2 trillion valuation and 80x growth, surpassing OpenAI, while securing a major compute deal with SpaceX for the Colossus 1 data center.
An analysis of leaked financial data shows Anthropic posting its first operating profit while xAI burned over $4 billion, revealing divergent AI monetization strategies and implications for developers and enterprise adoption.
Vercel's AI Gateway data shows Anthropic leads in spending, Google in token volume, and agentic workloads carrying 59% of token volume. OpenAI's spend share tripled after recent model updates.