How photonic computing can move from promise to commercialization

Advances in silicon-based computing have driven innovation in recent decades. However, modern workloads – including massive projects like climate simulations and artificial intelligence (AI) training – are pushing high-performance computers to their breaking point.

Meanwhile, quantum computing is still in its early stages, years away from supporting commercial-scale applications. Into this gap steps a promising contender: photonic (or optical) computing.

Photonic computing uses light rather than electricity to perform computations. While the concept has been explored for decades, recent developments are bringing it closer to commercial viability.

In fact, Gartner just included photonic computing in its 2025 Hype Cycle for Data Center Infrastructure Technologies, a key indicator that the technology is gaining traction among industry leaders and investors.

Photonic computing leverages the speed and efficiency of light. Photons can transmit data faster than electrons and work in low-energy environments, which makes them perfect for processing intensive workloads like scientific computations, machine learning and optimization problems.

Furthermore, while electronic systems dissipate heat and require intensive cooling, optical systems operate with minimal thermal overhead.

The momentum behind photonic computing is growing, and today’s research is paving the way for a new world of faster and more efficient computing.

The photonic computing field features several architectures, each with its own benefits and trade-offs:

1. Free-space optics (FSO): The oldest form of optical computing, FSO systems process information by manipulating light in air or a vacuum using lenses, spatial light modulators and masks. FSO systems range in size from small boxes to occupying several racks and offer both flexibility and speed.

FSOs face some challenges before they can be turned into practical products. One key issue is making the systems more durable and reliable, requiring improved opto-mechanical engineering, like the integration of solid optical blocks, built-in spatial light modulators (SLMs) or photonic metamaterials.

Another problem is that current spacial light modulators, which manipulate and control light in these systems, are much slower than electronic devices. However, a new generation of faster, higher-resolution SLMs is being developed and should help overcome these limitations.

2. Photonic chips: Combining miniaturized optical components like lasers, beam splitters and interferometers into a compact form factor, photonic chips enable speed and easy integration into existing electronic architectures. There are many technologies in this field, but they all struggle with scaling to handle more complex tasks.

Current chip designs are mostly two-dimensional, which limits their utility. To increase their performance, some companies have tried linking chips together, but this often causes significant signal loss. Another challenge is that photonic chips can’t store memory optically, so they constantly have to switch between light and electricity, which reduces performance and accuracy. In large optical circuits, analogue signals can also become weak and noisy, making it hard to achieve precise results.

Because scaling is so difficult, some companies are shifting focus from building full optical AI chips to creating optical interconnects, which use light to move data between electronic components. This path forward involves creating new materials and devices that reduce signal loss and improve accuracy. For example, lithium niobate has shown promise in early experiments.

3. Optical fibre systems: These systems harness guided light within optical fibres, building on the mature infrastructure of fiber-optic communications to perform complex calculations, which are useful for solving difficult problems in optimization and artificial intelligence, amongst others.

One example is the coherent Ising machine (CIM), which sends pulses of light around a loop of optical fibre to compute. Unfortunately, it relies on electronics for key functions, resulting in optical-electrical-optical conversion that significantly reduces computational speed. Future designs will likely shift towards chip-based architectures for better integration and scalability. To fully realize the optical advantage, the electronic field-programmable gate array (FPGA) that controls the system must be replaced with a fully optical processor, thereby eliminating the 'conversion penalty.'

Another creative design uses special fibres with multiple cores to perform many calculations at the same time, like multitasking with light. Most multi-core fibre optic systems are still in the research phase.

Despite growing momentum, additional technical hurdles must be overcome before photonic computing can reach widespread adoption:

We are entering a critical window in the evolution of photonic computing. With global demand for faster, greener and more capable computing growing daily, photonic systems offer an alternative that complements, and in some cases outpaces, traditional silicon-based systems.

While fully optical free-space systems appear most feasible in the near term, hybrid systems that combine optical and electronic components may also play a significant role, especially if advances in OEO conversion reduce energy penalties.

In-memory computing – the simultaneous use of light for compute and storage – might hold promise. In the mid-term, architectures that combine spatial and temporal processing may unlock new levels of performance and efficiency.