The data is in your home
The race to train robots is getting personal. One AI startup is now knocking on doors offering to clean apartments for free — the real payment being the data it harvests on the way, footage and movement it hopes will one day teach machines to do that work themselves. It's a striking illustration of where the industry now goes for training material: not the web, but the physical world.
That physical-world push is everywhere today. Shinkei, backed by Founders Fund, has built a refrigerator-sized robot named Poseidon designed to kill fish quickly and humanely. In Canada, farms are leaning hard into robotics to close a stubborn labor gap, drawn by both efficiency and necessity. OpenAI, meanwhile, is teaching software the same trick its rivals want for hardware: its Codex assistant can now watch you perform a task once and repeat it forever, blurring the line between demonstrating work and automating it away.
Powering the machines — and where to put them
All of this runs on data centers, and the strain is showing. Federal regulators have ordered America's grid operators to move faster in wiring up power-hungry AI facilities, a sign that electricity, not chips, may be the real bottleneck. Some are looking past Earth entirely: China has launched a state-backed "Space Computing Industry Innovation Center," a tech alliance aiming to build AI data centers in orbit.
The buildout is drawing fresh money and unlikely converts. Nebius is pledging £1.7 billion for UK AI infrastructure and partnering with Nvidia on a robotics lab. And Allbirds — yes, the wool-sneaker company — is reinventing itself around AI infrastructure, trading shoe boxes for server racks in one of the year's more improbable pivots.
Platforms, standards and a Hollywood retreat
The platform jockeying is heating up. Apple unveiled Core AI, a system for running generative models directly on your device rather than in the cloud. Tesla filed a trademark for "MEGAPOD," hinting at an AI computing platform tied to its Supercharger network. And in a rare show of cooperation, Google, Microsoft and Salesforce are rallying behind a common AI software standard — an attempt to write shared rules for the enterprise era.
Not every bet is paying off. Amazon MGM Studios has pulled the plug on its planned movie about OpenAI, a small but telling sign that Hollywood's fascination with the AI boom has limits.
The thread
Taken together, today's news traces AI's expanding footprint — into our homes, our farms, our power grids, and now space. The companies chasing it are no longer just software firms; they're cleaners, fishmongers, sneaker makers and carmakers, all converging on the same prize. The hard questions — power, labor, and who controls the standards — are arriving right alongside the ambition.