Microsoft open-sources MLVC codec targeting 88% bandwidth reduction on NPUs

The shift toward NPU-based encoding marks a pivot from fixed-function hardware to flexible, software-defined neural compression. For streaming operators, this suggests a future where device-side NPUs handle complex decoding tasks that previously required high-power GPUs or specialized ASICs, potentially lowering distribution costs while maintaining high perceptual quality. As Microsoft open-sources these tools, the industry moves closer to a unified neural coding standard that could bypass the fragmented patent pools of traditional codecs. Watch for the expansion of MLVC to 1080p resolutions and its adoption by third-party hardware vendors to validate its cross-platform stability.

Microsoft released the MLVC source code, model weights, and hardware conversion scripts on July 24, 2026, following research into the Deep Contextual Video Compression (DCVC) family. The project addresses a critical deployment barrier: ensuring that encoders and decoders maintain identical entropy maps across different silicon architectures. According to Microsoft, MLVC uses scale indices via a hyperprior to ensure cross-platform consistency, a significant hurdle for previous neural codecs. In subjective P.910 testing, Microsoft reported that MLVC matched the quality of 1 Mbps H.264 video at just 122 kbps for 360p30 conferencing workloads.

This development coincides with a broader industry shift toward AI-optimized hardware in consumer electronics. Per Giznova in February 2026, Smart TV NPU benchmarks now prioritize real-time 8K upscaling at 60–120Hz, emphasizing sustained throughput over peak performance. Furthermore, industry data from Netint in March 2026 indicates that high-volume VOD providers are 2.3 times more likely to invest in AI-based encoding to manage the dual pressures of rising quality expectations and operating expenditure constraints.

Parallel to these encoding advancements, Akamai’s July 2026 release of edge-native serverless functions targets the operational complexity of global delivery. By executing WebAssembly-compiled code closer to the user, Akamai enables real-time session revocation and token validation. This approach aims to eliminate entitlement drift, where mismatched policies across a multi-CDN footprint lead to either piracy via under-enforcement or subscriber churn due to false access denials.