‘Processor that doesn’t use electricity’ resurfaces

Photonic computing does not mean computers run without power; it means some math is done with light instead of moving electrons through transistors. The latest version of that pitch resurfaced in an April 2026 YouTube episode centered on German startup Q.ANT and its light-based processor. (youtube.com, qant.com) The basic target is matrix multiplication, the repeated number crunching behind artificial intelligence models. In photonic systems, beams of light can carry and combine those values in hardware, while conventional electronics still handle control, memory, and data movement. (quantamagazine.org, news.mit.edu) That is why the phrase “processor that doesn’t use electricity” is shorthand, not a literal description. Even Q.ANT describes its product as a photonic co-processor inside existing computing infrastructure, and MIT researchers describe current photonic hardware as part of hybrid systems rather than stand-alone general-purpose computers. (qant.com, news.mit.edu) Q.ANT said on July 22, 2025 that its Native Processing Server was installed at the Leibniz Supercomputing Centre in Garching, Germany, calling it the first analog photonic co-processor integrated into an operational high-performance computing environment. The Leibniz Supercomputing Centre published the same deployment announcement that day. (qant.com, lrz.de) Q.ANT said again on March 17, 2026 that second-generation processors had been deployed at the same center for production evaluation. The company said the project is aimed at artificial intelligence and simulation workloads in high-performance computing, not at replacing every central processor or graphics processor in a data center. (qant.com, datacenterdynamics.com) The timing lines up with a wider scramble over artificial intelligence power demand. The International Energy Agency said in its 2025 Energy and AI report that global electricity use from data centers is projected to rise from 460 terawatt-hours in 2024 to more than 1,000 terawatt-hours in 2030 in its base case. (iea.org) Nvidia’s current answer is denser, liquid-cooled racks built around Blackwell chips. Nvidia says its GB200 NVL72 links 72 Blackwell graphics processors and 36 Grace central processors in one rack-scale system, and the company has promoted liquid cooling as a way to manage the heat and electricity demands of that hardware. (nvidia.com, blogs.nvidia.com) Photonic startups are pitching a different bottleneck: moving data and doing linear algebra with less heat. Lightmatter says its products use both photons and electrons, while Lightelligence says it is building optical computing systems for real-world deployments, showing that the field now spans several companies rather than one viral claim. (news.mit.edu, lightelligence.ai, lightmatter.co) The technical catch is precision and completeness. MIT said in December 2024 that many photonic devices still cannot perform every neural-network operation on-chip, which is why researchers have kept working on hybrid designs that reduce conversions back to electronics. (news.mit.edu, nature.com) Researchers are still reporting progress on that front. A Nature paper published in April 2026 described a photonic tensor processor for deep neural network inference packaged in a standard 19-inch rack unit with a high-speed electronic interface to PyTorch, underscoring that the near-term model is integration with existing software stacks, not a clean break from them. (nature.com) So the resurfaced claim is best read as a power-and-architecture argument, not a literal end to electricity in computing. The systems now being tested use light for selected workloads, then plug back into the same electronic servers, memory, and software that run today’s artificial intelligence infrastructure. (qant.com, news.mit.edu)