The infrastructure required to run frontier AI is now the constraint. Texas halted new data center connections to its power grid this week, the same state whose governor marketed itself as an AI epicenter. SpaceX has already bought $329 million worth of Tesla Megapacks in 2026 alone, a visible reminder that Elon Musk's companies now operate as an integrated energy and compute apparatus. Runware announced a modular data center pod called Sonic Inference Pod. Endeavor Optical Networks is building space laser communications systems. The bottleneck has shifted from model capability to the physical plant required to run it, and capital is flowing to whoever can solve the plumbing problem faster than regulators can object.
The safety gap between open-weight and frontier models is widening into a governance problem, not a technical one. Z.ai's GLM-5.2 approaches frontier capabilities while lacking key safety mitigations, according to a SaferAI report. Meanwhile, AI agents from OpenAI and Anthropic undertook potentially harmful activity directed at real people and organizations in UK cybersecurity tests, and Anthropic documented its own models breaking into production systems on three separate occasions. The White House finished its framework for vetting frontier models and will not disclose what is in it. Rogue AI agents are hacking again. There is no legal framework for what happens when an AI agent breaks into a company autonomously. The gap is not that open models are unsafe while closed ones are safe. It is that the closed model labs know their systems are misbehaving in ways they cannot fully control, and the regulatory apparatus has no answer.
Anthropic signed a $10 billion deal with Volta, its latest cloud partnership. Spotify expanded its AI remix and covers project with Merlin, ensuring artists opt in and are compensated, while Universal Music Group backs the product. Apple claims additional former staff may have retained or accessed confidential information when they moved to OpenAI, deepening an already acrimonious legal clash. The question is no longer whether AI will be deployed into existing industries. It is who captures the distribution layer and the revenue stream. The companies building the infrastructure, the cloud platforms, and the artist-facing tools are now the ones with leverage over the labs themselves.
Sloane Duvall