AI data centers are facing mounting performance bottlenecks as memory and network limits reduce GPU utilization and overall efficiency.
Traditional copper and optical links force a trade-off: copper is reliable and power-efficient but limited to short distances under 2 meters, while optical fiber can span tens of meters but consumes more power and fails up to 100 times more often.
To address this challenge, Microsoft has introduced MOSAIC, a breakthrough optical interconnect that combines low power and cost, high reliability, and long reach of up to 50 meters.
MOSAIC uses a wide-and-slow architecture powered by microLED arrays, shifting away from today’s narrow-and-fast designs.
By transmitting data across hundreds of parallel low-speed channels, MOSAIC eliminates the inefficiencies of high-speed optics, reduces failure rates by two orders of magnitude, and cuts cable power consumption by up to 68%.
The technology is also fully compatible with existing pluggable transceiver form factors and protocols like Ethernet, PCIe, and CXL, making it a drop-in replacement for today’s copper and fiber connections.
Microsoft’s research shows MOSAIC can save more than 100 MW of power annually across global deployments, enough to power over 300,000 homes.
This innovation opens the door to new AI infrastructure designs, including multi-rack GPU clusters, disaggregated compute and memory systems, and network topologies that were previously impractical.
By delivering scalable, energy-efficient connectivity at long distances, MOSAIC could redefine how data centers are built, unlocking higher GPU utilization and enabling the next wave of AI performance.
You may also want to check out some of our other recent updates.
Subscribe to Vavoza Insider to access the latest business and marketing insights, news, and trends daily with unmatched speed and conciseness! 🗞️
