London-based Vaire Computing’s Ice River chip cuts AI energy use by 30%

Vaire Computing has successfully tested its Ice River chip, which demonstrated a 30% reduction in energy consumption compared to standard processors by reusing energy rather than discarding it as heat. The proof-of-concept achievement addresses growing concerns about AI’s escalating energy demands, though questions remain about whether hyperscale operators will adopt such technology over their own internal solutions.

How it works: Ice River tackles processor inefficiency through two key innovations that fundamentally change how chips handle energy.

• The chip uses reversible logic gates that allow energy to flow in both directions, enabling calculations to reuse inputs for further operations rather than discarding them as waste heat.
• It employs adiabatic computing, where voltages rise and fall gradually like a pendulum rather than changing abruptly like a hammer, which generates less heat and allows energy recycling into subsequent operations.

In plain English: Traditional computer chips work like using a hammer—they hit hard once and waste the energy as heat. Ice River works more like a pendulum, gently swinging energy back and forth so it can be reused multiple times instead of thrown away.

What they’re saying: Vaire Computing executives emphasize the paradigm shift from wasteful to efficient energy use.

• “While traditional computer chips can only use their stored energy once via a typical logic gate, the Ice River chip uses a reversible logic gate, which allows energy to be utilized in both directions,” the company explained.
• Mike Frank, Senior Scientist at Vaire Computing, noted that current devices “use energy once and then throw it away.”
• “You can think of [the energy] as sloshing back and forth,” Frank said, describing the recycling process.

The big picture: The August 2025 test represents a significant milestone for energy-efficient computing as AI systems drive unprecedented power consumption across data centers.

• Kathryn Hulick, a journalist at ScienceNews, characterized the chip’s approach as “more like a pendulum than a hammer,” highlighting its gentle energy management compared to conventional processors.
• For cofounder Hannah Earley, witnessing the processor in action was transformative: “I’ve been sketching [the chip] on paper and [running it] in simulation.”

Strategic positioning: Vaire Computing has been building its team and partnerships to support long-term development of the technology.

• In 2024, the company recruited Andrew Sloss, Arm’s former unofficial tech futurologist, as VP of technology.
• Vaire also joined Silicon Catalyst UK, a semiconductor incubator, to accelerate its development efforts.

Why this matters: While the 30% energy reduction is promising, the technology faces the challenge of competing against hyperscale operators who are developing their own energy optimization solutions, potentially limiting adoption in the largest AI infrastructure deployments.