MediaTek and Microsoft Research are advancing a new class of data center interconnects, introducing an active optical cable (AOC) built on microLED technology to address mounting performance and energy constraints in AI infrastructure.
The development signals a strategic shift in how hyperscale operators approach connectivity, particularly as AI clusters scale beyond conventional architectural limits. The microLED-based AOC targets efficiency, reliability, and scalability three pressure points that increasingly define next-generation data center design.
Breaking the Copper vs Optical Trade-off
The new active microLED cable replaces traditional laser-based optical transmission with arrays of miniaturized microLED light sources, addressing long-standing trade-offs between reach, power consumption, and reliability. Conventional copper interconnects offer high efficiency but are limited in distance, while laser-based optical links provide longer reach at the expense of higher power consumption and reduced reliability.
However, this innovation reframes that trade-off. Instead of relying on high-speed laser channels, the architecture distributes data across hundreds of parallel low-speed microLED channels, fundamentally altering the efficiency curve of optical communication.
Power Efficiency Gains Through MOSAIC Integration
By leveraging Microsoft’s MOSAIC technology and MediaTek’s design expertise, the solution uses hundreds of parallel low-speed microLED channels instead of high-speed laser channels. This approach enables up to 50% lower power consumption compared to conventional VCSEL-based AOCs by eliminating the need for complex digital signal processing.
Moreover, the reduction in signal processing overhead directly translates into lower latency and improved thermal performance critical metrics for AI workloads operating at scale.
The design also improves reliability, with microLEDs offering greater durability and temperature stability, achieving performance comparable to copper links while extending transmission reach for large-scale AI data center clusters. Scalability is enabled by increased channel density or higher per-channel data rates, allowing aggregate bandwidths of 800 Gbits/s and beyond within standard QSFP and OSFP form factors.
Therefore, operators can scale bandwidth without proportionally increasing failure risk or thermal complexity, an increasingly important factor in dense AI deployments.
A key innovation is the integration of all electronic functions into a single monolithic CMOS chip, including logic, drivers, and amplifiers, reducing latency and power overhead. Additionally, advanced heterogeneous integration directly bonds microLED and photodetector arrays onto the chip, enabling compact designs with high channel density. Consequently, this level of integration simplifies system architecture while improving signal integrity across increasingly complex interconnect topologies.
Strategic Implications for AI Infrastructure
The collaboration continues to focus on further miniaturization and readiness for mass production, supporting the evolution of high-performance, energy-efficient AI data center infrastructure.
Meanwhile, as AI workloads drive exponential growth in east-west traffic within data centers, innovations like microLED-based AOCs could redefine the economics of scaling compute clusters. The ability to combine copper-like efficiency with optical reach positions this technology as a potential cornerstone for future AI infrastructure deployments.
