Foundation Model Labs

A foundation-model lab turns compute, data and research talent into a general-purpose model, and then sells access to that model through an API, subscriptions, enterprise contracts and its own products. A handful of labs sit at the frontier: OpenAI, Anthropic, Google DeepMind, xAI and Meta's superintelligence group in the US, with DeepSeek, Alibaba Qwen, Moonshot AI and others close behind in China and Mistral AI in Europe. They have grown revenue faster than any software companies in history: OpenAI from ~$3.7 billion in 2024 to ~$13 billion in 2025 (reported), and Anthropic from ~$4.6 billion for all of 2025 to ~$16.3 billion in the first half of 2026 alone (S-1, October 2026, reported). They have also raised and committed more capital than any private companies in history (OpenAI ~$122 billion at ~$852 billion in early 2026; Anthropic $65 billion at $965 billion in May 2026; compute commitments in the hundreds of billions of dollars each, reported), because their main input, computation, is bought in advance and their main product, a model, loses most of its value within a year or two as competitors and successors arrive. A lab's business model is therefore an answer to five questions: how fast is revenue growing and from which customers, what does each dollar of revenue cost to serve (inference), how much must be spent on training to stay at the frontier, how is that spending financed and committed, and what stops customers switching to a cheaper or open model? This section maps the labs, explains their revenue models, cost structure, training economics and release cadence, funding and compute deals, governance, distribution relationships, commoditisation pressure and legal overhangs, then sets out margins, key numbers, cases, pitfalls and how practitioners think. A worked example models a stylized lab's revenue against its compute burn under four scenarios, and a checklist sets out diligence questions.