The gap between AI's infrastructure ambitions and the actual constraints of building them is widening fast. OpenAI proposed robot taxes and public wealth funds while Iran published satellite coordinates of its $30 billion data center. Samsung forecasted record profits on the AI boom even as the US tech sector shed 15,000 jobs in March. The story underneath these headlines is not about AI capability anymore. It's about who controls the physical layer, the chips, the power, the real estate, the supply chains, and whether the companies building models can actually defend what they've built.
Robotics moved past the capability question. GEN-1 hit 99% reliability on tasks from box folding to vacuum repair, handling disruptions it wasn't trained for. Meanwhile, researchers cut AI energy use by 100x while improving accuracy by combining neural networks with symbolic reasoning. These are engineering wins that matter. But they sit alongside a harder reality: a new chip survives 700 degrees Celsius, which is interesting for future systems, but today's data centers are vulnerable to missiles. North Korea compromised npm. Anthropic's own security tool got a CVE. The physical infrastructure protecting AI's value is under active attack, and the companies running it are making strategic bets, Intel doubling down on advanced chip packaging, Xoople raising $130 million to map Earth for AI with L3Harris building the sensors, that assume they can build faster than adversaries can break.
The talent and capital structures are already reshaping around this. OpenAI alums launched Zero Shot to raise $100 million specifically to fund the infrastructure and tools layer. Rocket is selling AI-generated strategy reports at McKinsey prices to companies that can't hire consultants. Google launched an offline dictation app using Gemma to compete with Whisper Flow. These are not moonshots. They're market discipline at work: builders moving toward problems with actual friction and willingness to pay, away from the hype cycle. The question now is whether the companies that control the models, OpenAI, Google, Anthropic, can maintain margin when the value is migrating toward whoever owns the infrastructure that keeps the models running when the grid is contested.
Sloane Duvall