The court's dismissal of antitrust claims against Google's AI search reveals a widening gap between regulatory appetite and legal authority. Judges are willing to acknowledge that AI search disrupts traditional information markets, but they're declining to treat that disruption as anticompetitive conduct. Meanwhile, the actual power shifts happening in AI are moving faster than litigation can track. OpenAI is embedding itself into commerce through virtual try-ons in ChatGPT and shopping integrations. Shopify is building Canvas to let merchants construct stores through conversational AI. ServiceNow launched Flow as a standalone AI service desk that works in Teams, Slack, and email without requiring enterprise infrastructure. These aren't regulatory questions; they're market capture by integration. The companies winning aren't necessarily the ones with the biggest models but the ones threading AI into the friction points where users already work.
The infrastructure race is bifurcating between closed and open approaches, with real money following both. Google is restricting Gemini 4 Argon to a "trusted" set of cyber defenders through its Fairwind Program, treating frontier models as controlled access. SpaceX and Google are betting on space data centers, though Google estimates Starship needs 1,800 launches before that pays off. Meanwhile, Satlyt is raising $8 million to build open software that runs on multiple companies' satellites, explicitly positioning itself against SpaceX's closed model. On the chip supply side, Micron is warning that memory shortages will tighten through 2028, and Amazon is trying to offload $8 billion in Nvidia chips to investors to ease its balance sheet. The capital intensity is real, and so is the pressure to monetize it quickly. Microsoft and Google backing Apache Ossie signals that even the largest players see vendor lock-in as a problem worth solving collectively, but that's a signal of how entrenched lock-in has become.
The real tension is between agent-first infrastructure and the old software paradigm. Photon held a funeral for mobile apps and raised $4.5 million to build agents over messaging platforms. Brian Chesky is arguing that AI agents need their own operating system. Airbnb, ServiceNow, and Shopify are all building agent-native products. But this transition requires semantic clarity at scale, which most enterprises still lack. The dream of enterprise semantics remains unsolved: companies spend days arguing about what a "customer segment" means before they can build a report. Google and Microsoft backing Ossie, Omnissa shipping Elara to break data silos, and the broader push for interoperability all point to the same bottleneck. The companies that solve semantic consistency across their data will control how agents behave inside their platforms. That's where the real competitive moat forms. It's not about having the smartest model; it's about having the cleanest data and the tightest integration with where work actually happens.
Sloane Duvall