Extropic, a bold new startup, is developing a radically efficient computer chip that harnesses thermodynamic fluctuations to perform probabilistic computations, aiming to revolutionize AI hardware and drastically cut energy use. (Source: Image by RR)

Extropic Chips Aim to Handle Reasoning Models Like OpenAI o3 and Gemini Flash

Extropic is a startup aiming to revolutionize computing by creating a new type of chip that leverages thermodynamic fluctuations—typically considered a nuisance in electronics—to carry out probabilistic calculations with remarkable energy efficiency. Rather than relying on binary bits fixed as 1s or 0s, Extropic’s chips use probabilistic bits, or p-bits, which fluctuate between states based on controllable probabilities. This innovation could dramatically reduce energy consumption, especially in datacenters where AI workloads are becoming increasingly demanding.

The company, as noted in wired.com, claims its technology could be three to four orders of magnitude more efficient than current hardware, a bold promise in a field dominated by Nvidia’s high-performance GPUs. Extropic has demonstrated key functionality through oscilloscope readings that show controllable bit-state transitions, marking what it calls the birth of the first scalable and energy-efficient probabilistic computing platform. Unlike previous thermodynamic computing efforts that relied on superconducting materials, Extropic’s chips use fluctuations in standard silicon, making them easier to manufacture and deploy.

Extropic’s probabilistic hardware is particularly well-suited for Monte Carlo simulations—a type of computation that underpins financial modeling, biological systems analysis, and AI reasoning. The startup believes its chip architecture is ideal for powering advanced AI models like OpenAI’s o3 and Google’s Gemini 2.0, which rely heavily on probability and inference. This positions Extropic’s technology as a potential alternative for high-performance computing in an industry desperate for more power-efficient solutions.

While challenging established players like Nvidia may seem far-fetched, the current moment offers a unique opportunity. As global AI demand pushes energy consumption to new highs and governments invest heavily in AI infrastructure, Extropic’s radical approach may prove timely. With increasing concerns about the environmental costs of AI, reimagining computing architecture from the ground up might not be crazy—it could be essential.

read more at wired.com