Microsoft has developed a new data transmission technology that replaces traditional copper wiring and laser-based fiber optics with light-based MicroLED systems, potentially reducing energy consumption by around 50 percent in AI data centers. The innovation, developed by Microsoft Research in Cambridge, U.K., addresses a growing problem: as AI workloads demand more computing power, the physical infrastructure that connects servers and graphics processing units (GPUs) is hitting hard limits. The company plans to commercialize the technology with industry partners starting in late 2027. What's Wrong With Today's Data Center Wiring? Data centers currently rely on two main approaches to move data between servers and GPUs. Copper cables are cheap and widely used, but they have a critical weakness: as data transmission speeds increase, the maximum effective distance shrinks to around two meters. This forces chips to sit closer together, creating what researchers call "unsustainable" power density within racks and making cooling far more complex. For longer distances, data centers turn to fiber optic cables powered by lasers. These can transmit data over much greater distances at high bandwidth, but they come with trade-offs: higher power consumption, larger components, and greater sensitivity to temperature and dust. As AI demand explodes, neither approach can keep pace with the infrastructure challenges ahead. "Increasing data rates force chips to be placed closer together, creating a rapid increase in a rack's power density, which is unsustainable," explained Doug Burger, Managing Director of Microsoft Research Core Labs. Doug Burger, Managing Director of Microsoft Research Core Labs, Microsoft How Does Microsoft's MicroLED System Work? Microsoft's solution uses commercially available MicroLED chips, similar to those found in consumer electronics, paired with imaging fiber, a type of cable typically used in medical endoscopy. The key innovation is the imaging fiber itself, which contains thousands of internal cores capable of carrying parallel streams of light. This allows the system to distribute data across many lower-speed channels rather than a few high-speed ones, a design philosophy researchers call "wide and slow" compared with the "narrow and fast" approach of traditional laser systems. The concept originated in 2020 when researchers explored whether commercial LED chips could be adapted for high-speed data transmission. After years of development, the team created a working prototype in the Cambridge lab and miniaturized it into a compact transceiver device compatible with existing server hardware. "Imaging fiber looks like a standard fiber, but inside it has thousands of cores. That was the missing piece. We finally had a way to carry thousands of parallel channels in one cable," explained Paolo Costa, Partner Research Manager at Microsoft. Paolo Costa, Partner Research Manager, Microsoft What Are the Real-World Benefits? Microsoft's internal testing suggests the MicroLED system could use around 50 percent less energy than current laser-based optical systems. Beyond energy savings, the system offers other advantages. LEDs are less sensitive to environmental factors like temperature and dust that can degrade laser-based components, potentially improving reliability. The system is designed to operate over distances of tens of meters, positioning it between copper and traditional fiber optics in terms of range while aiming to combine the advantages of both. The technology is being developed alongside other networking innovations within Microsoft's infrastructure. Hollow Core Fiber (HCF), already deployed in some Azure regions, transmits light through air within a hollow core rather than glass, enabling faster data transmission and lower latency over longer distances. According to Microsoft, HCF can deliver up to 47 percent faster data transmission and around 33 percent lower latency compared with conventional fiber. Steps to Understanding Data Center Cooling Innovation - Energy Challenge: Traditional copper and laser-based systems consume significant power and create physical constraints that limit how densely servers can be packed into racks, making cooling increasingly difficult. - MicroLED Solution: Light-based transmission using thousands of parallel channels reduces energy consumption by approximately 50 percent compared to laser systems while improving reliability and environmental tolerance. - Complementary Technologies: Microsoft is deploying MicroLED for short-distance connections within data centers while using Hollow Core Fiber for longer-distance, lower-latency connections across greater distances. Why Should Tech Companies Care About This Now? The timing of Microsoft's announcement reflects an industry-wide crisis. AI workloads are growing exponentially, and data centers worldwide are struggling to find enough electricity to power them. Energy has become the fundamental constraint limiting how quickly AI infrastructure can scale. Companies like NVIDIA, ASUS, and others are simultaneously investing in liquid cooling systems, modular data center designs, and power management software to address the same bottleneck. Microsoft's decision to make the MicroLED technology widely available signals the company's belief that industry-wide efficiency gains benefit everyone. As Frank Rey, General Manager of Azure Hyperscale Networking, noted, the combination of MicroLED and Hollow Core Fiber creates a comprehensive networking strategy: MicroLED handles the pure efficiency of LED over laser for power usage within data centers, while HCF extends reach and reduces the need for additional buildings, generators, and energy infrastructure. "With MicroLED, you have the pure efficiency of LED over a laser. That has a pure bottom-line impact to power usage at any given datacenter," stated Frank Rey, General Manager of Azure Hyperscale Networking. Frank Rey, General Manager of Azure Hyperscale Networking, Microsoft The broader context matters here. Other organizations are pursuing parallel solutions. Preferred Networks, Internet Initiative Japan, and Japan Advanced Institute of Science and Technology have deployed a direct liquid-cooled, high-density AI server system in a modular data center designed to achieve a Power Usage Effectiveness (PUE) rating of just 1.1 units in free cooling mode and 1.2 units annually, compared with typical data center PUE values of 1.5 or higher. ASUS has announced optimized liquid-cooling solutions achieving a PUE of 1.18 in its deployment for Taiwan's National Center for High-performance Computing. These developments suggest the industry is converging on a multi-pronged approach: improving networking efficiency through technologies like MicroLED, deploying advanced cooling systems to handle higher power densities, and designing modular, purpose-built data centers optimized for AI workloads. Microsoft's MicroLED breakthrough represents one critical piece of this larger puzzle, addressing the specific challenge of how data moves between components rather than just how heat is removed. The commercialization timeline matters too. With deployment expected from late 2027, the technology arrives at a moment when the AI industry will likely be even more desperate for efficiency gains. By then, the cumulative power demands of AI data centers globally could be substantially higher, making a 50 percent reduction in networking energy consumption a significant competitive advantage for early adopters.
