The money is moving toward control layers and operational infrastructure while consumer-facing AI products keep failing. OpenAI shutting down Sora marks the third major consumer play the company has abandoned in months, yet Kleiner Perkins just raised $3.5 billion specifically to back AI companies at scale, and Databricks is acquiring startups to build security products around AI agents. The pattern is clear: venture capital and established tech companies are betting on the unglamorous work of managing, monitoring, and integrating AI into existing workflows rather than on novel consumer experiences. This reflects a maturation of the market from novelty to operational necessity.
JetBrains is building a control plane for AI coding agents, Mozilla developers are creating Stack Overflow equivalents for agent failures, and Anthropic is pushing Claude Code toward autonomous execution while maintaining what it calls safeguards. Arm is manufacturing its first in-house chip with Meta and OpenAI as early customers, while Agile Robots is embedding Google DeepMind's foundation models into its hardware. The infrastructure layer is consolidating around a few platforms and models. Simultaneously, the Pentagon's attempt to designate Anthropic a supply chain risk is meeting judicial skepticism, with a federal judge calling the Defense Department's motivations troubling. That legal pressure sits alongside a US-China Economic and Security Review Commission warning that China's open-source AI strategy is building competitive advantage precisely where US export controls cannot reach it.
The real competition is no longer about which company builds the best chatbot. It is about who controls the deployment stack, who owns the data flowing through production systems, and who can integrate AI into enterprise software without breaking existing operations. Zoom is betting its edge lies in capturing interactions across video and meetings. Spotify is adding tools to prevent AI-generated content from being attributed to real artists. Local-first products like Talat are finding traction by keeping data off the cloud. These moves suggest that builders have learned the hard lesson: consumer enthusiasm for AI features does not translate to sustainable business models, but operational necessity does. The winners will be companies that solve the coordination, observability, and integration problems that come after the model is trained.
Sloane Duvall