70% GM and 15%-20% FCF margin Damn Anthropic is printing cash now
Deep|LLM: 26H1 Update (Part 2): Frontier Labs, Chinese Model Vendors & the Compute Bottleneck
Anthropic–OpenAI duopoly: Model-layer revenue is concentrating fast, with Anthropic ARR moving from $1B at end-2024 to above $47B by May 2026, versus OpenAI at roughly $33B on public / The Information basis. Anthropic leads in coding agents and enterprise contracts, while OpenAI re-accelerated in late May on GPT-5.5, Codex, API, and ChatGPT distribution.
Agent workflows unlock budgets: Token budgets and agentic workflows are becoming the new entry point for enterprise productivity spend. Anthropic has 300,000 enterprise customers, 1,000+ paying over $1M annually, and NDR above 500%, while OpenAI’s Codex WAU rose from 1M+ on February 5, 2026 to 5M by May 31.
Chinese model vendors accelerate: ByteDance’s Seedance 2.0 reached global SOTA in video generation and is running at RMB 15–20bn annualized revenue, while DeepSeek V4 reset cache-hit pricing and GLM-5.2 pushed open-source coding models closer to closed-source leaders. These models are commoditizing mid-tier tasks, but still trail Anthropic and OpenAI on long-horizon stability, multi-tool coordination, and enterprise product ecosystems.
Compute becomes the bottleneck: Data, algorithms, and model iteration are all converging back into physical compute demand. H100 one-year lease rates have rebounded 40% since 2025-10, Nebius raised H100 PAYG pricing from $2.95/hour to $3.85/hour from 2026-06-01, and Anthropic secured incremental capacity from SpaceX Colossus 1 with 220k+ NVIDIA GPUs and over 300MW.
Detailed Report https://fundaai.substack.com/p/deepllm-26h1-update-part-2-frontier
