AI Chips and Infrastructure — Page 2

Meta is giving away 5,000 dongles and the firmware to build your own Muse hardware

The choice of the cheapest serious hobbyist board is the clearest statement of intent: Meta is not selling a device, it is finding out what form factor people want by watching them build one.

Nvidia's new entry-level AI desktop has half the memory of the old one and costs 25% more than it launched at

The cheaper machine is cheaper only against today's prices. Two of them cluster over a direct 200GbE link to run models up to 200 billion parameters.

Boston Dynamics taped its engineers' pinkies down for a day, then built Atlas a hand without one

The goal moves from grasping objects to working with them: in-hand reorientation, recovering a slipping grip, and holding a drill while pulling its trigger.

Google's four chips are in orbit, and they can only compute for fifteen minutes before they have to cool down

The satellite is the size of a fridge, carries four TPUs running a version of Gemma, and its heaviest components are the radiators that dump the heat.

Japan's largest power producer will put a 400-megawatt AI data centre next to its own gas plant, skipping the grid queue

The site sits on land beside JERA's Chiba thermal power station and takes its electricity behind the meter. The companies call it the largest single-site AI deployment planned in Japan.