The frontier AI labs are no longer competing primarily on model capability, they're racing to own the operating layer where intelligence gets deployed, governed, and monetized across enterprises and devices. This shift explains why OpenAI is building agentic coding tools that control your desktop, why Anthropic is expanding in London while negotiating Pentagon access, and why Google is gluing its AI directly into Chrome and Photos. The capability gap between Claude Opus 4.7 and the leaked Mythos matters less than who controls the infrastructure where these models actually work.
Factory's $1.5 billion valuation after three years, Upscale AI's reported $2 billion raise just seven months after launch, and Physical Intelligence's π0.7 robot brain all point to venture capital treating AI infrastructure as the new platform layer. But concentration is accelerating: first-quarter venture funding flowed overwhelmingly to large, well-funded U.S. companies, even as global deal count fell. The economics are brutal. Data center delays now threaten Microsoft and OpenAI projects. Meta raised Quest headset prices by $50 to $100 citing RAM shortages. When infrastructure becomes the bottleneck, whoever controls it owns the next decade of software. Anthropic's expansion to London and discussions to provide Mythos to the U.S. government reflect a calculation that regulatory access and geographic diversification matter more than staying small and pure.
The real tension surfaces in how these labs are positioning themselves against traditional software. Anthropic's CPO leaving Figma's board to build competing design tools, Runway's CEO betting AI can make fifty films instead of one blockbuster, and Canva's AI assistant calling external tools all signal the same threat to existing SaaS incumbents. Enterprise customers are beginning to see AI not as a feature but as a replacement for entire categories of software. InsightFinder's $15 million raise to diagnose where AI agents fail, and the emergence of tools like Antioch's robotics simulation platform, reveal that the real margin isn't in the model, it's in the operational layer that makes models reliable enough to replace humans at scale. Google blocking 8.3 billion ads while suspending fewer advertisers shows how platform power compounds when you control both the model and the distribution channel.
Sloane Duvall