Artificial intelligence has reached a tipping point. On one hand, generative models continue to push the boundaries of what machines can do. On the other, their environmental and energy costs are becoming unsustainable. In 2025, AI systems consumed water equivalent to the world’s annual bottled water consumption and generated a carbon footprint comparable to that of New York City—primarily driven by training and inference operations in massive data centers.

In this context, Cycle Momentum co-invested in Irréversible, a start-up tackling a fundamental challenge: how to deploy advanced artificial intelligence where energy is scarce, without relying on the cloud.

A Double Technological Wall: Energy Cost and Cloud Dependency

The problem identified by Irréversible stems from two converging dynamics that current technologies fail to address.

First, the cloud-centric AI model does not scale. As capabilities grow and use cases multiply, energy consumption, water usage, and associated emissions increase dramatically. This trajectory is simply not sustainable.

Second, the future of AI is increasingly shifting to the edge. Critical applications—autonomous drones, persistent sensors, industrial systems, sensitive infrastructure, and defense—require real-time decision-making directly where data is generated.
 These use cases cannot depend on continuous cloud connectivity due to latency, security, sovereignty, and privacy constraints.

At the core of the issue lies a fundamental architectural limitation. All modern digital processors are based on the von Neumann architecture, which separates memory and computation. This separation leads to massive energy losses caused by data movement—a physical bottleneck that cannot be solved through incremental engineering improvements.

The result is stark: entire classes of applications are currently impossible, despite a global market opportunity estimated at $10 billion by 2032 for ultra-low-power edge intelligence.

 

A Technological Breakthrough: Ultra-Low-Power Analog AI

To overcome this barrier, Irréversible is developing fully analog AI inference chips capable of consuming up to 1,000 times less energy than existing digital alternatives.

Their approach is radical: leveraging the physical properties of analog circuits to perform neural network computations directly, without moving data between memory and processing units. Computation happens within memory itself.

This architecture delivers what the team calls Minimum Viable Intelligence (MVI)—precisely the level of intelligence required for autonomous decision-making, without the overhead of general-purpose digital systems.

In practical terms, this enables:

  • autonomous drones capable of intelligent navigation during long missions without recharging,
  • industrial and environmental sensors that can operate for years without maintenance,
  • embedded monitoring and detection systems under extreme size, weight, and power constraints. 

Beyond individual devices, the environmental impact is systemic. By processing intelligence locally, Irréversible eliminates both the energy cost of continuous data transmission and the computational burden on data centers.

As AI inference shifts from centralized cloud infrastructure to distributed edge deployments—driven by sovereignty, security, and performance requirements as much as power constraints—this architecture offers a viable path toward sustainable, scalable AI.

 

Why the Origo Investment Is a Key Milestone

Irréversible’s pre-seed financing round, led by Quantacet, with participation from Frostbite Capital, Cycle Momentum (via Origo), Tofino Capital, UCeed, Colin Harris, BDC, and Capital Labs, represents a major milestone for the start-up.

For Cycle Momentum, the investment also sends a clear signal: energy-efficient computing is a climate issue in its own right. Ultra-low-power computing technologies are essential to reconciling digital transformation with planetary boundaries and technological sovereignty.

The funding is enabling Irréversible to:

  • accelerate the development of its hardware and software platforms,
  • complete multiple chip design and testing iterations,
  • expand its core engineering team,
  • deepen collaborations with partners building next-generation autonomous systems.

What’s Next: From the Lab to Real-World Deployment

Irréversible has already achieved significant technical milestones, completing multiple chip tape-outs to validate the core building blocks of its technology, which are currently undergoing characterization and functional testing.

In parallel, the team is developing the full software ecosystem required for industrial adoption:

  • tools to automatically adapt digital neural network models to analog hardwares,
  • simulation frameworks to manage analog circuit non-idealities,
  • standardized benchmarks to validate performance across a wide range of applications.

The immediate roadmap focuses on transitioning to application-ready systems, with new generations of AI chips and a customer evaluation platform developed in close collaboration with strategic partners in robotics, industrial IoT, and dual-use applications.

The objective is clear: to move analog AI from advanced research into real-world deployments, where power efficiency, autonomous operation, and computational sovereignty are critical.