Microsoft has announced plans to commercialise a fundamentally different approach to moving data between servers and GPUs in datacentres. The system, called MOSAIC, replaces the expensive lasers at the heart of modern optical cables with inexpensive MicroLEDs, the same technology found in smartwatch displays. The company expects to bring the technology to market with industry partners by late 2027.
MOSAIC replaces lasers with inexpensive MicroLEDs and uses a fundamentally different architecture to transmit data inside datacentres, cutting energy consumption by roughly 50% compared to mainstream laser-based optical cables, according to Paolo Costa, the Microsoft partner research manager leading the project. The innovation addresses a genuine constraint facing cloud operators and artificial intelligence infrastructure builders: the power and reliability trade-offs inherent in today's cabling choices.
Datacentres currently face a hard choice. Conventional datacenter optical cables use lasers to fire data through a small number of high-speed channels, while MOSAIC inverts that by sending data across hundreds of parallel low-speed channels using directly modulated MicroLEDs. This architectural difference yields substantial practical benefits. MicroLEDs are smaller, cheaper, and more temperature-stable than lasers, which are vulnerable to heat fluctuations and dust.
The MOSAIC paper, which won Best Paper at ACM SIGCOMM 2025, reported up to 68% power savings and failure rates up to 100 times lower than conventional optical links. Microsoft's conservative operational estimates suggest the technology will deliver roughly 50% energy savings in real-world deployment. The system can reach up to 50 metres, well beyond the roughly 2-metre limit of copper cabling used for high-bandwidth GPU connections within a single rack.
The proof of concept miniaturised the lab prototype into a standard transceiver form factor compatible with existing datacenter equipment, requiring no changes to servers or switches. This compatibility matters enormously for adoption: operators need not redesign physical infrastructure or risk downtime during deployment.
Microsoft is complementing MOSAIC with parallel investment in Hollow Core Fiber, which carries light through an air-filled core instead of solid glass, delivering up to 47% faster data transmission and approximately 33% lower latency than conventional single-mode fiber. Microsoft acquired HCF developer Lumenisity in 2022 and has since signed manufacturing partnerships with Corning and Heraeus to scale production. The technology is already live in some Azure regions.
Frank Rey, Microsoft's general manager of Azure Hyperscale Networking, said the two technologies are complementary: MOSAIC serves connections inside datacentres, while HCF covers longer distances between datacentres and Azure regions. This layered approach reflects realistic engineering constraints. MOSAIC handles the dense, high-frequency switching demands inside a datacentre. HCF provides the long-haul efficiency needed to connect facilities across regions without the latency that hampers AI inference workloads.
The timing reflects infrastructure pressure. Cloud operators now face a fundamental constraint: as GPU counts in AI clusters expand into the hundreds of thousands, the network and memory infrastructure cannot keep pace with compute advances. With the rapid growth of AI and cloud demand, existing networking technologies are increasingly constrained by physical limits in distance, power consumption, density and reliability.
For operators evaluating these technologies, the practical question is clear: can MOSAIC and HCF actually deliver the promised economics at scale? The path to commercialisation still carries real uncertainty. Manufacturing processes for both MicroLEDs and hollow-core fibre remain capital-intensive. Supply chains for these components are immature outside research environments. And the 2027 timeline, while specific, carries the usual product development contingencies.
That said, Microsoft's parallel deployment of HCF technology in live Azure infrastructure, combined with peer-reviewed publication of MOSAIC research at a top-tier conference, suggests the engineering fundamentals are sound. The company is not making these announcements in isolation; they reflect internal confidence about near-term deployment within Microsoft's own global datacentre operations.
For the broader industry, the competitive pressure is evident. If Microsoft can mature and scale these technologies successfully, other hyperscalers operating at comparable scale will feel compelled to follow. The energy and reliability gains alone justify investment at that scale. The question is whether manufacturing capacity and industry standardisation can keep pace with demand.