Who builds it, who pays for it, who earns?
Two published stories and a reading plan for the business behind AI, from chips and cloud capacity to models, agents and the companies selling them. Planned entries are research directions, not promised dates.
The stack
Hyperscalers: who rents the AI factory?
The cost of data centers, chips and cloud capacity, and who gets paid when a model is trained or served.
Starting source ↗CUDA: software as a hardware moat
Why developer tools around the GPU matter to chip sales and switching costs.
Starting source ↗Models: the product behind the API
Compare the ways model providers package and price access, without treating every benchmark win as a business.
Starting source ↗Weights: open, closed and licensed
Who can download, run, adapt and sell access to a model, and how its license changes the business.
Starting source ↗Agents: from answers to work
Where tools, human review, reliability and usage costs enter when a model takes actions.
Starting source ↗Harnesses: the system around the model
The economics of context, tools, checkpoints and evaluation across long-running tasks.
Starting source ↗Company studies
Anthropic's IPO prospectus asks who pays for intelligence
A reported look at the revenue, compute bill and market wager behind a possible listing.
OpenAI filed for an IPO too. Only OpenAI has not shown us the numbers.
A reported look at its confidential IPO filing, revenue run rates and compute commitments.
Coming soon
Thinking Machines Lab: customization and open weights
Explore its model and developer platform, and what open weights change for distribution.
Starting source ↗Kimi (Moonshot AI): model access as a business
Follow the developer model catalog and its approach to selling API access.
Starting source ↗DeepSeek: pricing and open-model competition
Use its published model and API pricing as a starting point for the cost-versus-capability story.
Starting source ↗Confluence Labs: learning when data is scarce
Study the economics of AI research that aims to learn from costly experiments and small datasets.
Starting source ↗