15 photonics startups to watch, according to VCs

Photonics is the technology of generating, guiding, detecting, and manipulating light. In communications, it turns information into patterns carried by light, then converts those patterns back into electrical data at the destination. The article presents photonics as a response to AI’s growing physical demands.

A simple example is an optical connection between data-centre systems. A transmitter encodes data onto a light signal. The signal travels through an optical path, while a receiver detects it and decodes the information. Multiple light channels can carry separate streams of data through one connection.

The article does not provide technical measurements for any specific photonics system. It does show why the field is attracting attention: AI workloads require faster and more efficient computing, while chips, cables, and other components are struggling to keep pace. Photonics startups are therefore being watched as possible solutions to data-centre bottlenecks.

Venture capitalists are watching these 15 startups because photonics addresses a problem created by AI’s rapid expansion. The article says increasingly demanding AI workloads need faster and more efficient approaches to computing. That makes technologies for moving and processing data especially important.

The companies are not presented simply as ordinary software bets. They are developing hardware-related approaches in a market shaped by data-centre construction, chip limits, cable constraints, and rising processing speeds. Photonics may help connect computing components more effectively, making it strategically important as AI systems become more capable.

The article identifies the list as an investor survey and calls the startups ones to watch, but it does not provide enough readable detail to compare the individual companies. Its central signal is clear: investors believe photonics could become commercially significant because AI growth is exposing weaknesses in conventional data-centre infrastructure.

The article does not state a total amount of data generated or moved by modern AI systems. It describes the volumes as increasingly large and says that data centres are being built globally to support the AI boom. No figure is provided in the readable portion.

The scale also varies widely. A system training a model, serving answers to users, or moving data between storage and processors can create different traffic patterns. The exact amount depends on the model, workload, number of users, and data-centre design. Those details are absent here.

What the article does establish is the direction of change. More capable AI systems require greater data volumes and faster processing. As those demands rise, chips, cables, and related components struggle to keep up. The important takeaway is therefore a capacity problem, not a specific number: data movement is becoming a central constraint on AI infrastructure.

AI workloads are unusually demanding because more capable systems need both large data volumes and rapid processing. The article says data centres are being built globally, but the chips, cables, and other components inside them are struggling to keep up. Infrastructure growth alone does not remove limits inside each facility.

The bottleneck appears wherever information must move between computing components. Chips may process data quickly, while connections and supporting hardware cannot deliver that data at the same rate. When one part is slower, the whole system can spend more time waiting. The article does not identify one component as the sole cause.

This mismatch matters because AI performance depends on coordinated systems, not isolated chips. As workloads become more demanding, operators need faster and more efficient ways to connect and support computing hardware. That pressure helps explain why photonics has become a focus for investors and startups seeking alternatives to existing approaches.

When data moves too slowly between chips, processing capacity cannot be used fully. A processor may wait for inputs instead of continuously performing calculations. The result can be slower AI responses, longer training runs, and lower overall system efficiency. These effects follow from the article’s description of data movement as a physical challenge.

The mechanism is straightforward. AI work is divided across computing components, and those components must exchange information. If a connection cannot deliver data at the required rate, faster chips cannot compensate indefinitely. More hardware or longer operating times may then be needed to complete the same workload, although the article gives no numerical estimates.

Energy use and cost can rise when systems run longer or require additional infrastructure. However, the article does not provide measured figures for either impact. It does establish why the issue matters: AI workloads are becoming more demanding, and inefficient data movement threatens the speed and economics of expanding data-centre capacity.

Photonic technologies could improve data-centre connections by carrying information with light. This may support higher-speed links and better efficiency than some conventional electrical connections. The article frames that potential as important because AI systems require faster and more efficient approaches to computing.

The key mechanism is conversion. A transmitter places data onto a light signal, which travels through an optical connection. A receiver detects the signal and converts it for use by electronic computing hardware. Light-based links can also carry multiple channels, though the article does not give specifications for any particular startup or product.

Photonics is not described as a complete replacement for every electrical component. Instead, it is presented as a promising response to bottlenecks involving chips, cables, and other data-centre parts. Its importance is growing alongside global data-centre construction and increasingly capable AI systems. The 15-startup investor list signals commercial interest, but not guaranteed success.

Computers represent information as bits, usually encoded into changing electrical states inside electronic hardware. For transmission, those bits become controlled signals that travel through a connection. A receiver measures the signal, reconstructs the bit pattern, and passes the data to another component. Photonics performs the same broad task with light.

In a photonic link, a transmitter changes a light source or modulator according to the data pattern. The light travels through an optical path, and a detector converts the arriving pattern back into an electronic signal. Multiple channels can share a connection by using different light wavelengths, although the article does not explain this mechanism.

Light matters because networks need to move increasing data volumes quickly. The article says AI systems demand faster processing and that existing chips, cables, and components are struggling. Photonics is attracting attention because it may make these links faster and more efficient, though the article provides no performance figures.