Deploying AI Inference Functions on Tencent Cloud Serverless with Rust and WebAssembly

This article summarizes Michael Yuan’s talk at ServerlessDays China 2021, where he explains how WebAssembly combined with Tencent Cloud Serverless enables developers to deploy TensorFlow AI models with only a few lines of code.

Yuan argues that Serverless is the future of cloud computing because it removes the need to manage servers and offers higher development efficiency and business flexibility. He notes that most Serverless functions are written in Java or Python, which are heavy and have long cold‑start times.

The talk introduces the Jamstack revolution (JavaScript, API, Markup) and explains how decoupled front‑ends rely on lightweight Serverless functions for backend logic. However, current Serverless functions suffer from slow cold starts and limited performance.

To address these issues, Yuan presents a solution based on Rust + WebAssembly. Rust provides memory safety and high performance, while WebAssembly offers a lightweight, portable runtime. Together they can run AI inference at 10‑20 frames per second on Tencent Cloud Serverless without GPU acceleration, a 100× speedup compared with Python.

The open‑source project WasmEdge is introduced as a CNCF‑sandboxed WebAssembly runtime optimized for edge and serverless scenarios. WasmEdge supports AOT compilation, TensorFlow extensions, storage, and blockchain extensions, and has been evaluated in IEEE Software as the fastest WebAssembly runtime outside browsers.

Yuan compares WebAssembly with Docker and traditional containers, showing that WebAssembly has much faster cold‑start times, lower overhead, and a capability‑based security model. He also highlights WebAssembly’s growing role in blockchain smart contracts and formal verification.

Finally, the talk discusses the potential of WebAssembly to replace Docker in the future, its lightweight nature, portability, and strong security guarantees, positioning it as a key technology for cloud‑native, edge, and AI workloads.