The core challenge for scaling artificial intelligence may not be processing power, but the energy consumed moving data, according to Optalysys. Copper interconnects, traditionally used to link chips, are failing to meet the bandwidth and speed demands of expanding AI infrastructure, driving investment in photonics. But treating photonics solely as a connectivity solution “vastly underestimates the technology, and the opportunity,” the company asserts. Optalysys proposes a shift, exploring how photonics can move beyond data transport to participate directly in computation itself.
Data Movement Bottlenecks Constrain AI Infrastructure Scaling
Increasingly, the energy consumed shifting data eclipses the power needed for computation itself, a shift impacting the scalability of artificial intelligence systems. The US Department of Energy estimates data centers may account for approximately 9% of total US electricity consumption by 2030, a figure that emphasises the growing strain on power grids as AI workloads intensify.
Goldman Sachs projects US data center power demand will more than double, reaching 66 gigawatts in 2027 from 31 gigawatts in 2025, demonstrating the rapid escalation of energy requirements. PwC forecasts data center expenditure will rise from around $800 billion per year in 2026 to $1.8 trillion per year in 2050, highlighting the escalating financial burden of sustaining current data movement practices.
Current AI infrastructure relies heavily on traditional copper interconnects, but these are reaching their limits in terms of reach, speed and supply, while memory bandwidth struggles to keep pace with processor demands. As AI clusters expand, the overhead associated with moving data between compute elements is becoming as significant as the computational tasks themselves. Photonics offers a potential solution by enabling high-bandwidth, low-energy data transfer over longer distances, driving its increasing adoption in AI infrastructure.
Optalysys proposes a fundamental shift in how photonics is used, questioning whether optical infrastructure can move beyond simply carrying data between electronic processing points. The company asks if photonics has matured enough to participate directly in computation, a concept that could redefine the efficiency of AI systems.
When useful work occurs directly within the optical path, data movement transforms from a pure overhead cost into an integrated part of the process, reducing latency, improving energy efficiency and enabling more sustainable infrastructure scaling. “If the US is to maintain its leading role in that story, we must recognize that photonics should compute, not just connect,” the company asserts, emphasizing the strategic importance of this transition for US AI infrastructure leadership.
Optalysys’ Compute-in-Transit Enables Optical Data Processing
Modern AI infrastructure faces escalating constraints not solely from processing power, but from the energy demands of data movement between components. Optalysys is developing photonic Compute-in-Transit, a system designed to integrate computational capability directly within the optical data path, rather than relying on electronic processing after data transfer. This approach aims to use the strengths of both electronic and photonic domains, with electronics handling control and general-purpose processing while photonics executes high-throughput, structured operations aligned with signal propagation.
This contrasts with current systems where data is moved, then computed upon, and then moved again after processing, incurring significant energy expenditure in conversion and transfer. By adding compute capability within the optical path, Optalysys seeks to minimize these costs and improve overall system efficiency. This concept challenges the conventional separation of communication and computation, envisioning a future where these processes are more tightly integrated.
The architecture is not intended to replace electronic systems entirely, but rather to create a hybrid approach that optimizes performance across the entire data processing chain. According to Optalysys, the scalability, security and sovereignty of future AI systems will depend on architectures capable of more intelligent data handling. “Today, communication and computation are often treated as separate processes. We move the data, then compute on it; we transport, then transform,” the company explains.
Compute-in-Transit aims to break down this division, enabling a more smooth and efficient flow of information. Optalysys’ work positions photonics as a foundational compute architecture, rather than merely an enabling technology, the company says. The company is actively seeking collaborations to demonstrate the efficiency gains of Compute-in-Transit in real-world applications, and believes this approach represents a critical step towards sustainable AI infrastructure. The future of AI, they suggest, lies in systems that not only process data effectively, but also move and process it intelligently.
See today’s quantum computing news on Quantum Zeitgeist for the latest breakthroughs in qubits, hardware, algorithms, and industry deals.
