Yes—but so far, the clearest evidence is a research demonstration, not a commercially available chip that can run a general-purpose LLM like a GPU. A 2025 paper reports prompted text generation with a transformer-based model on a photonic computing prototype. That shows optical hardware can execute an LLM-like workload; it does not show that standard LLM software can be installed on a photonic accelerator or that photonics can replace GPUs today.
What did the photonic LLM experiment actually do?
In a 2025 Nature Communications paper, Zhou and co-authors describe a single-layer photonic computing approach called SLiM and report text generation with a transformer-based model containing 0.345 billion parameters and 96 layers. The experiment used 356 token samples and produced text through four recursive generation steps. The paper reports a photonic loss of 3.04, compared with 2.96 for its digital result.
As an Amazon Associate I earn from qualifying purchases.
Those measurements describe this particular experiment. The loss values are not a comparison with a deployed commercial LLM, and the reported text generation is not a benchmark of service quality or production speed. The paper also reports a 10 GHz data rate, but that figure is not an end-to-end token-generation rate: it does not tell you how many tokens per second a complete user-facing system would deliver.
Do these 3 things before closing this tab:
1Repair Windows errors before they cause bigger problems2Scan for outdated or missing drivers - takes under a minute3Clear out junk files and repair common Windows errorsThe paper’s abstract also describes a 0.192-billion-parameter, 640-layer model for image generation. That is a separate result, not a larger language model.
#1 Best Overall
- Professional AI & Creator Workstation: AMD Radeon AI PRO R9700 GPU with 32GB GDDR6 is engineered for AI development, professional content creation, and compute-intensive workloads.
- Massive 32GB Memory Capacity: 32GB of GDDR6 memory on a 256-bit bus provides ample bandwidth for large AI models, 8K video editing, and complex 3D rendering.
- Advanced RDNA 4 with AI Accelerators: 64 Compute Units with 3rd Gen Ray Tracing and dedicated 2nd Gen AI Accelerators for groundbreaking AI performance and visual computing.
- Professional Blower Cooling: Efficient single blower design exhausts heat directly out of the chassis, ideal for multi-GPU workstation and server configurations.
- Enterprise-Grade Thermal Solution: Vapor chamber heatsink with industrial Honeywell PTM7950 thermal interface material ensures reliable cooling under sustained professional loads.
How can light compute part of an LLM?
Photonic chips use optical signals to perform selected neural-network computations. One important target is matrix-vector multiplication, a linear operation used extensively in neural networks. But an LLM is more than a collection of those operations: executing a complete workload also involves other computation, memory, data movement, control and software.
In the SLiM demonstration, the model and photonic operations were configured for the experiment. It is therefore more accurate to call this a photonic prototype executing a transformer-based text-generation workload than to call it a normal PC-compatible accelerator for installing standard LLM software.
Rank #2
- High-Performance AI Processing: The MX3 is designed to handle the most demanding AI computer vision workloads, delivering exceptional performance and efficiency.
- Flexible Integration: The MX3 can be easily integrated into your existing systems via its M.2 M-key form factor and support for Linux operating systems.
- Energy Efficient: The MX3 is designed to provide high performance while minimizing power consumption.
- Comprehensive Software Development Kit (SDK): The MX3 is supported by a comprehensive SDK that simplifies development and deployment.
- Hardware compatability: The MX3 is compatible with the PCI-SIG M.2 M-key 2280 Specification. It can be used with the Raspberry Pi 5 with a M-key 2280 HAT.
What limits photonic AI chips?
Analog errors can build up with computation depth
Optical neural networks are analog physical systems. As signals propagate and nonlinear computations are repeated, errors can accumulate, making deep networks difficult to execute reliably. The SLiM paper identifies this as a central obstacle and proposes a single-layer propagation design intended to make deeper computation more tolerant of error.
Demonstrated systems remain limited in scale and configurability
A 2026 scholarly commentary discusses end-to-end photonic inference demonstrations while noting that such systems remain far behind electronic accelerators in scale and configurability. In practical terms, a research result on a configured prototype does not establish that a photonic system can support the range of models, operations and deployment choices available with general-purpose electronic hardware.
A chip data rate is not application throughput
The 10 GHz figure reported for the SLiM experiment is an experimental data rate, not measured tokens per second for a production LLM service. End-to-end speed depends on the full system and workload, not just the rate of one component. The cited paper does not provide a controlled comparison with a production GPU deployment.
Commercial availability and framework compatibility are unproven here
The cited sources document research prototypes and evaluations. They do not establish that a photonic accelerator is generally available to buy, broadly compatible with current standard LLM frameworks, or ready to host arbitrary models. A successful research demonstration is evidence of technical feasibility for a specific workload, not proof of drop-in compatibility.
Rank #4
- ✅Powered by 26 Tera-Operations Per Second (TOPS) Hailo-8 AI Processor. 2.5W typical power consumption
- ✅Scalable, enabling simultaneous processing of multi-streams & multi-models
- ✅Enabling real-time, low latency and high-efficiency AI inferencing on the edge devices
- ✅Supports TensorFlow, TensorFlow Lite, ONNX, Keras, Pytorch frameworks
- ✅Supports Linux and Windows. Supports the temperature range of -40°C to 85°C
How to judge claims about photonic LLM performance
To compare a photonic system with an electronic accelerator, look for results using the same workload and accounting for the whole system. Useful comparison points include:
Recommended Free Tools
- Model and output quality: model size, task, prompt and quality measures.
- Software and operations: what models and operations the hardware supports, and how they are configured.
- End-to-end performance: latency and tokens per second for the complete workload.
- Total energy: energy for the whole system, including conversion, memory and control—not only the optical computation.
- Capacity and programmability: supported context and model sizes, and how readily the hardware can be reconfigured.
- Deployment status: whether results come from a laboratory prototype or a commercially deployed product.
Without those details, a chip-level operating rate cannot establish that a photonic system is faster, more efficient or cheaper than a GPU for running LLMs.
Best Value
Can optical computing run ChatGPT?
The SLiM result does not show ChatGPT running on a photonic chip. It demonstrates prompted text generation with a specific transformer-based research model. The cited evidence also does not establish that a photonic accelerator can run ChatGPT or other standard hosted LLM software.
Does this mean photonic chips will replace GPUs?
No replacement is demonstrated by these results. They show that photonic computing can execute a transformer-based text-generation workload, while the 2026 commentary describes a continuing gap from electronic accelerators in scale and configurability. The available evidence does not support an apples-to-apples claim about production throughput, total energy use or cost versus a GPU.
The practical verdict is that photonic AI chips are a promising area of specialized acceleration research, but the cited work is not evidence of a general-purpose, commercially ready GPU replacement for LLMs.
Free tools Windows power users keep installed
One-click scans. No signup required.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
