Breaking physical bottlenecks in the era of Super Intelligence

Inference is on track to become the largest computing workload behind Super Intelligence, and power is now the limit on how fast data centers can add capacity for it. Much of a chip’s energy goes to moving model data rather than computing with it.

This venture is building an inference chip that spends far less energy moving data and is designed for high-volume manufacturing. Its first chips already match the accuracy of conventional digital hardware, and the next generation is designed to run a full frontier model on a single module. It sells first to the largest inference buyers.

Super Intelligence is moving out of the data center and into robots, drones, wearables, vehicles, and sensors. Today’s chips use too much energy to run capable models on these devices, so they settle for small models or depend on the cloud.

This venture is building a new computing architecture designed to run machine-learning models at a fraction of the energy of conventional chips. Device makers will license it and build it into their own chips, so capable models can run on the device itself.

Thousands of satellites now launch every year, and defense and in-space infrastructure programs are adding more. All of them run on solar power, and the high-performance cells the most demanding missions need are in short supply.

This venture is building a domestic manufacturer of space-grade solar cells around a much faster production process. The process runs on production-scale equipment and has produced independently certified cells comparable to space-qualified product. Space is the first market, and the same process can make other high-performance semiconductors that have been too expensive to produce in volume.

Quantum computers, cryogenic sensors, and large detector arrays all run at temperatures near absolute zero, and every signal in or out travels over wiring from room temperature to the cold stage. Each line brings heat and takes up space. Useful quantum computers will need hundreds of thousands to millions of qubits, and that wiring already limits how large these systems can grow.

This venture is building data links that carry those signals with far less heat and space, and early devices are already more energy efficient than the best components available today. The company will sell complete modules, from the cold stage to room-temperature electronics, to quantum and cryogenic system builders across the industry.