Infinite is excited to announce our strategic investment in Extropic, the beginning of our partnership as they redefine the very nature of compute.

After almost a year of involvement with the Extropic team, it has become clear that their talent and insight are unparalleled, encompassing a unique set of competencies allowing them to reimagine compute. This comes at a major moment for Extropic as they recently unveiled the existence of their printed thermodynamic microchip¹ and circuit board. This fundamental milestone is only the beginning of Extropic's journey towards civilizational impact.

The Answer to the End of Moore’s Law - Thermodynamics

For over half a century, computing has advanced at an exponential rate, following the densification of transistors along the trajectory observed by Moore’s Law - the doubling of transistors on a microchip every two years (initially every 18 months) outlined in Gordon's aptly titled “Cramming More Components onto Integrated Circuits.”² The question now is: what happens when we can no longer cram more transistors into a chip? 

The golden age of transistor growth is now reaching the constraints of atomic scale, EUV printing limits, and Nyquist noise.³ As semiconductor fabs struggle to push nanometer accuracy and AI workloads drive energy demands to unsustainable levels, the industry has responded by continuously scaling costs and input investments—the solution has been simply more chips and more energy.

Enter Extropic, a company rewriting the fundamental rules of computation. Instead of fighting against the thermal noise that limits the continued processing gains of traditional semiconductors, Extropic harnesses it. 

Fundamentally Redesigned Compute

Extropic’s approach is radically different from both digital and quantum computing. Traditional silicon chips rely on precise, deterministic logic gates to perform binary calculations with transistor-based circuits. In contrast, quantum computing, which leverages a qubit's ability to exist in superposition, utilizes these multiple simultaneous states to accelerate computation. While compelling, its development is early and faces enormous unsolved challenges in computational stability and logistical scaling.

Thermodynamic computing, by contrast, enables probabilistic energy-based models (EBMs), to achieve rapid sampling of complex distributions—a capability that’s inherent in any stochastic process, from Monte Carlos simulation to AI inference. Extropic achieves this by evolving from the current digital compute approach that is based on binary bit states, either 0 or 1. Instead, Extropic's compute evolves this approach with the introduction of p-bits, probabilistic bits,⁴ which exist as a probability distribution between 0 and 1. In order to accomplish this, their hardware rewrites the base design of semiconductors with the incorporation of parameterized stochastic analog circuits.‍⁵ This results in a fundamentally new architectural approach to computing at the hardware level, that could propel AI and high-performance computing as computational power continues to scale exponentially. Through this innovation, Extropic’s accelerators promise to reach orders of magnitude improvement in energy efficiency and processing speed over today’s most advanced microchips.

This has a wide range of implications for both the entire technology sector and society at large. While the immediate acceleration of progress in generative AI and computationally-heavy modeling would benefit, it goes well beyond that enabling a more biologically-native approach to modeling our world. While digital compute forces algorithms to approximate highly complex physical processes with fundamentally binary logic, thermodynamic compute seeks to enable a hardware-native approach to simulating complex probability distributions. These probabilistic systems exist throughout our world, from weather patterns and chaotic systems to Brownian motion interactions at the molecular level. As quantum mechanics has revealed, our entire reality is fundamentally probabilistic in nature, and this new paradigm for compute adapts to that at the circuit level with "programmable randomness".

The Future: A Civilization Powered by Extropic Compute

Extropic’s vision goes far beyond bringing an increase in compute power to the microchip sector. While improving energy efficiency in AI and inference would alleviate the massive energy demands of data centers that companies like OpenAI, xAI, and Google now struggle to sustain, it would also extend much further. It could usher in new and previously impossible approaches to machine learning, where AI researchers could begin adapting AI models to embrace their inherent randomness. With the prevalence of probabilistic models across every major field from economics to geopolitics, it is only fitting that compute evolves to be inherently probabilistic in nature as well. Unconstrained, these models would exhibit greater natural-world intelligence, unveiling new insights into biology, chemistry and physics. Beyond that, we can imagine a world of decentralized AI, where a single chip could empower individual scientific and research contributors in these fields. Further decentralization would distribute AI beyond megacorps and server-farms, eventually leading to edge-devices that are imbued with AI intelligence from a singular thermodynamic chip. From intelligent autonomous vehicles to smartphone-native AI, these gains would start in academic and R&D sectors before eventually enabling entirely new and unimagined consumer applications. Empowered with a new scale of computational power, humanity will grow its awareness of reality in tandem. This expanded ability to harness the world of atoms could even result in breakthroughs in nuclear fusion, humanoid robotics, and the colonization of space. We are at the dawn of a new scientific revolution across sectors—from energy to biologically informed technology—and thermodynamic computing may very well be the engine of progress.

‍