This NPU Is 5,000% Faster Than A GPU | Photonic Chips Explained

2026-06-22 · Source: Bug · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Emerging Technologies & Innovation, Cloud Computing & IT Infrastructure · Depth: Advanced, medium

Summary

Photonic MPUs are emerging as a highly energy-efficient alternative to GPUs and ASICs like Google's TPU for AI inference, particularly dense matrix multiplication. Unlike silicon chips that push electrons, photonic MPUs compute by routing and interfering light waves, generating virtually zero resistive heat and slashing calculation energy costs to roughly one femtojoule per operation. This makes the math hundreds of times more energy efficient than silicon. Qunnect recently launched the first commercial photonic MPU as a PCIe card, designed to offload specific math workloads from a CPU, working alongside a GPU. This optical chip delivers up to 30 times higher energy efficiency for certain math workloads, pulling only 150 watts. A significant challenge remains the energy drain from converting electrical signals to optical and back, which currently costs hundreds of times more power than the computation itself.

Key takeaway

For AI Architects evaluating inference hardware, consider photonic MPUs for dense matrix multiplication workloads. Your current silicon-based solutions face inherent energy limits due to resistive heat. Integrating optical cards, like Qunnect's PCIe offering, can drastically reduce power consumption for specific math tasks, achieving up to 30 times higher energy efficiency. However, be aware of the significant energy overhead from electrical-to-optical signal conversion, which remains a key challenge impacting overall system efficiency.

Key insights

Photonic MPUs offer extreme energy efficiency for AI inference by using light waves, overcoming silicon's resistive heat limitations.

Principles

• Light-based computation eliminates resistive heat.
• Wavelength division multiplexing scales bandwidth.
• Optical operations resolve complex math natively.

Method

Photonic MPUs route and interfere light waves, with weights programmed into optical components, converting input data into laser pulses for matrix calculation.

In practice

• Offload dense matrix math to optical cards.
• Integrate photonic MPUs in data centers.
• Target specific math workloads for efficiency.

Topics

• Photonic Computing
• AI Inference
• Optical MPUs
• Qunnect
• Wavelength Division Multiplexing
• Hardware Acceleration

Best for: CTO, VP of Engineering/Data, Director of AI/ML, AI Hardware Engineer, AI Scientist, AI Architect

Related on AIssential

• Q.ANT Hits Full Production Capacity for Photonic AI Processors — Photonic processors offer dramatic performance and energy efficiency gains over traditional silicon-based AI chips.
• Tensordyne Claims Massive Speed and Power Improvement Over Nvidia — Tensordyne's Napier chip uses logarithmic math for matrix multiplication, achieving significant AI inference speed and power efficiency gains.
• Microchip Breakthrough No One Expected — Purpose-built NPUs with data flow architectures offer superior energy efficiency for AI inference compared to general-purpose GPUs.
• This Will Power Everything — Optical interconnects are critical for scaling AI data centers beyond copper's physical and thermal limitations.
• Penn physicists use light-matter particles to boost AI chip speeds — Exciton-polaritons enable all-light signal switching in semiconductors, potentially enhancing AI chip speed and efficiency.
• Burning a Transformer into Silicon: The Case for GPU-Free AI Inference — Directly burning transformer logic into FPGAs offers ultra-efficient, GPU-free AI inference.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Bug.