The AI industry is fracturing along two axes simultaneously: price competition is collapsing margins while regulatory friction and political backlash are fragmenting where and how models can operate. Neither trend favors the incumbents who built their trillion-dollar valuations on scarcity and control.
OpenAI and Anthropic are cutting prices on cheaper models as Chinese rivals gain ground, a race to the bottom that mirrors what happened to cloud infrastructure. Nvidia's NeMo Switchyard and Kog's GPU optimization work both target the same problem: inference costs are the new battleground, and routing requests to the cheapest capable model is becoming table stakes. Google's Gemini 3.7 Flash pricing undercuts its own 3.6 Flash from three weeks prior, suggesting internal pressure to hold market share in production deployments. Meta's release of Glimmer as open-weight while locking Muse Spark behind APIs reveals the real strategy: give away commodity models to build platform lock-in, not to prove philosophical commitment to openness. Mark Zuckerberg's letter about AI being "for everyone" arrived the same week his company lost control of Manus to Chinese regulators and abandoned the acquisition, a reminder that geopolitical friction now determines who can own what.
The political resistance to data center buildout is moving faster than any company's pricing strategy can outrun. Iowa City postponed a council meeting due to death threats over a Salix data center proposal. Americans oppose data centers in their areas by large margins, while lawmakers argue the US cannot afford to slow buildout. Apple partnered with Alibaba for China-specific AI because US models face regulatory exclusion there. Microsoft lost its AI ethics lead after one year; OpenAI's robotics chief quit over DoD contracts; researchers are departing over safety concerns. The pattern is clear: regulatory capture is being replaced by regulatory rejection, and the companies still betting on permissive deployment timelines are making bets on a political environment that no longer exists.
Sloane Duvall