Nuclear Computing: The groundwork for a Paradigm Shift

January 2025

by Miguel Santos [Mike] LUPARELLI MATHIEU

The world is entering into the era of Nuclear Computing.

The current technology stack, including algorithms and semiconductor manufacturing, relies on energy-intensive technologies and infrastructure. In the short to medium term, this is expected to be predominantly powered by Nuclear Energy.

Microsoft has signed a deal to reopen the Three Mile Island nuclear plant, while Amazon in partnership with X-Energy and Energy Northwest, OpenAI with Oklo, and Google through Kairos Power, among other key AI players, are investing in Small Modular Reactors (SMRs) to power their AI engines and data centres.

Nuclear Computing is expected to be operative not before 2030. More or less. And their prospects estimate that it will last for many years in the future. Otherwise, there won’t be enough incentives to push for it.

The same driving forces pushing on Nuclear Computing will also push on alternative and disruptive innovation to diminish the pressure on energy. There are some candidates that could make it. See the Annex for a brief overview of the main drivers of change.

Paradigm Shift

Key drivers influencing algorithms, semiconductor manufacturing, and connectivity are laying the groundwork for a transformative paradigm shift poised to redefine the foundations of computing.

It will probably impact in more than one aspect of the Computing Stack at the same time (e.g. algorithms, chips, and connectivity). Innovation within the algorithm landscape could be the most transformative in the short term.

Agentic AI is positioned to be leading the AI roadmap to General Purpose AI in the short term, but interests in Defence, Biotechnology, and Life Science will fund both the path to Nuclear Computing and its disruptive alternative.

Nuclear Computing is a signal that a change is near. Whether it is an incremental innovation or a disruptive one entailing some of the driving forces described below, something will happen and will have a huge [nuclear] impact.

Annex - Drivers of change

Algorithms. This driving force is moving on to finding more efficient algorithms that consumes less energy and less data to obtain similar results as existing AI engines. The following candidates are well positioned for becoming a disruptive driving force of innovation that might contribute to a paradigm shift.

Quantum Machine Learning leverage on qubits unit of information, Hilbert Space, superposition, and entanglement to reduce the computing time and data needed to optimize portfolios or classify events and signals, among others.

Liquid Foundations Models (LFMs) introduces dynamical system and signal processing mechanisms to outperform existing Generative AI models both in terms of size and energy consumptions (Liquid AI).

Spiking Neural Networks (SNNs, “simulate the neural and synaptic structures and functions of the brain to process information” IBM) introduce efficiency gains in AI engines.

Small Open Source LLMs have proved to obtain similar performance with smaller architectures suitable to operate in consumer-grade devices.

Semiconductors manufacturing. This driving force has technological and geopolitical implications. Though the semiconductor manufacturing dynamics is still looking for efficiency in size, processing power, and efficiency in terms of energy consumption, there is a feeling that existing transistor scaling is reaching its physical limits.

GPUs are still the most valuable assets for building the most powerful AI engines and manage the scale of large data. Microsoft, Meta, Google, and Amazon (Trainium2) are investing in their own AI chip manufacturer capacity. All of them are looking to not be so dependent on NVIDIA but also to reduce the cost of development of AI engines.

There are alternative designs looking for a disruptive change in this trend. Microsoft has a stake in the hardware used for Neuromorphic Computing, and in developing Quantum Computers. Google recently launched the Willow chip designed to deploy large scale quantum processors. Yet its production is focused on quality rather than quantity until the technology proves to be disruptive and scalable for consumer-grade.

Photonic Computing uses light instead of electricity for computing states to speedup process and consume less power while building AI engines.

A researcher from Stanford University (Felix Petersen) suggested a new strategy to teach AI by implementing logic gates networks that plans to operate in logic gate chips.

Some restrictions about semiconductors exports are challenging some countries and companies’ resiliency making geopolitics an aspect to be considered in semiconductors manufacturing and provisioning.

Japan, among many other countries in the world, mainly in Asia, has plans to become a world-leading semiconductor manufacturer by 2030, while others are closing deals with main semiconductor manufacturers to ensure the supply of chips for the next few years to come.

All these aspects of this driving force are pushing in three directions: a) efficiency in size, processing power, and energy consumption, b) supply resiliency, and c) a disruptive innovation.

Connectivity. The company Ericsson describes the next connectivity generation (6G) as “omnipresent wireless intelligence”. This entails and envisions the spread of computing stack (algorithms and semiconductors) both on-device and on-cloud.

People rely on mobile devices to connect with human and artificial intelligence. Edge Computing will consolidate more and more as the main entry point between the physical and the digital world.

Larger AI engines will still be consumed on-cloud while the most advanced innovations on computing stack will bring some of the existing AI functions and features to on-device.

Premium mobile devices underuse its computing and algorithm capacity. Upcoming advancement on on-device semiconductors and some of the innovations in algorithms will continue transforming and expanding the capabilities of the Edge Computing.

High speed connectivity will bring the kind of real time and on-device Collective Intelligence that could revolutionize how we interact with AI and more importantly, how we reinforce their learning.