CoreWeave's $35B Bet: The Math Behind AI Infrastructure

An MIT researcher is using GPT-5.6 Sol with Codex to autonomously run quantum computing experiments, including analyzing results and calibrating qubits. The application demonstrates AI's capability to handle complex, iterative scientific workflows without human intervention. This represents a practical use case for large language models in experimental physics and quantum research.

Hannah Earley, 31, is leading Vaire Computing to commercialize reversible computing, a decades-old theoretical approach that recovers energy typically wasted as heat in chip calculations. The company achieved a key milestone last year by demonstrating a chip with a resonator that recovered more energy than it consumed, moving the concept from theory toward practical implementation. Reversible computing could significantly improve energy efficiency in data centers, laptops, and phones by retaining intermediate calculation data rather than erasing it, avoiding the energy loss that occurs during conventional chip operations.

Danijar Hafner, a 31-year-old AI researcher who worked at Google Brain and DeepMind, has launched a stealth-mode startup in San Francisco focused on developing robots that can navigate unfamiliar environments. Using model-based reinforcement learning and world models, Hafner's approach enables AI agents to plan ahead and handle scenarios they have not encountered during training, a capability critical for deploying robots in human spaces. His technique allows complex robotic tasks without extensive real-world trial-and-error training that has traditionally been required in robotics.

AI inference workloads are fundamentally reshaping data center architecture, shifting focus from raw compute power to integrated systems that optimize memory, storage, and networking together. Unlike training-centric deployments, inference demands continuous data retrieval and real-time response, making data movement the primary bottleneck. Organizations must rearchitect infrastructure around specific workload requirements rather than retrofitting AI into legacy systems, balancing performance, efficiency, cost, and scalability.