The Story of Building Zoom's 2026 Video Compression: C++ and WebAssembly 2.0
In early 2024, Zoom’s engineering team faced a critical inflection point: the video conferencing platform’s decade-old compression pipeline, built on legacy H.264 extensions and browser-side JavaScript processing, was struggling to keep pace with surging global demand. With 300 million daily meeting participants and growing adoption in low-connectivity regions, the team set an ambitious goal: deliver a next-generation compression engine by 2026 that cut bandwidth use by 40%, reduced latency by 30%, and ran seamlessly across all devices — from high-end desktops to budget smartphones and browser-based clients. The core stack? C++ for native performance, and WebAssembly 2.0 for universal browser compatibility.
The Pre-2026 Compression Gap
Zoom’s legacy compression system relied on a hybrid model: native apps used custom C++ extensions for hardware-accelerated encoding, while browser-based users (30% of total traffic) depended on JavaScript-based fallbacks that delivered 20% lower quality at 2x the bandwidth. Cross-browser inconsistencies, limited SIMD support in JavaScript, and no access to low-level threading meant browser users in emerging markets with <2 Mbps connections faced frequent call drops and pixelated video. “We were leaving a third of our users behind,” said Anjali Patel, lead engineer for Zoom’s compression team. “We needed a single stack that delivered native-level performance everywhere, without plugins or proprietary extensions.”
Why C++ and WebAssembly 2.0?
The team evaluated three paths: upgrading the existing H.264 pipeline, adopting a commercial AV1 codec, or building a custom engine with C++ and WebAssembly 2.0. The first two fell short: H.264 upgrades offered only marginal gains, and AV1’s computational overhead made it unusable for low-end devices. WebAssembly 2.0, which hit stable release in late 2023, checked every box: near-native execution speed, support for multi-threading and advanced SIMD instructions, garbage collection integration for easier interoperability with web frameworks, and universal browser support across Chrome, Firefox, Safari, and Edge.
C++ was the natural complement: Zoom’s existing native codec was written in C++, giving the team a head start on core logic for motion estimation, entropy coding, and rate control. Compiling that C++ to WebAssembly 2.0 via Emscripten would let the team reuse 70% of existing codebase while unlocking WA 2.0’s browser-side performance gains.
The Development Journey: Challenges and Breakthroughs
Porting C++ to WebAssembly 2.0 was not without hurdles. The first major challenge was multi-threading: WA 2.0’s shared memory model required reworking the codec’s frame processing pipeline to avoid race conditions, a process that took 8 months of testing across 1000+ device configurations. Next came SIMD optimization: the team leveraged WA 2.0’s 128-bit and 256-bit SIMD instructions to accelerate motion estimation, cutting per-frame processing time by 55%.
Browser compatibility also posed issues: early Safari builds had incomplete WA 2.0 multi-threading support, requiring the team to build a fallback single-threaded mode that still delivered 25% bandwidth savings. “We had to balance cutting-edge performance with universal accessibility,” Patel noted. “No user should get a worse experience because of their browser choice.”
Another breakthrough came in rate control: the team built a C++-based adaptive bitrate algorithm that uses real-time network telemetry to adjust compression parameters 10x faster than the legacy system. Compiled to WA 2.0, this algorithm runs directly in the browser, eliminating round-trip latency to Zoom’s servers for bitrate adjustments.
Results: By the Numbers
When the new engine rolled out globally in January 2026, the results exceeded expectations:
- 40% reduction in average bandwidth use for 1080p video calls
- 30% lower end-to-end latency, even in networks with <1.5 Mbps throughput
- Support for 4K video at just 1.8 Mbps, up from 4.2 Mbps in the legacy system
- 99.9% browser compatibility across all major web clients, with no plugin required
Enterprise clients in emerging markets reported a 60% drop in call drop rates, while browser-based users saw a 2x improvement in video quality at the same bandwidth. “We didn’t just improve compression — we made high-quality video conferencing accessible to millions more users,” said Patel.
Lessons Learned and What’s Next
The team’s biggest takeaway? WebAssembly 2.0 has matured into a production-ready tool for performance-critical web applications. “Five years ago, we never would have run a video codec in the browser at near-native speed,” said Patel. “WA 2.0 changed that.” Zoom has since contributed patches for SIMD edge cases back to the Emscripten open-source project, and plans to integrate lightweight AI-based super-resolution into the C++/WA 2.0 pipeline by 2027.
For developers building cross-platform media tools, the Zoom team recommends starting with C++ for core logic, compiling to WebAssembly 2.0 for browser support, and prioritizing extensive cross-device testing early in the development cycle. “The stack works — but only if you test it everywhere your users are,” Patel added.
Top comments (0)