ASGI in 2026: Uvicorn vs Hypercorn and the State of Python Async Servers

ASGI has won. The async server gateway interface, first proposed in 2018, is now the standard for Python web serving. FastAPI, Starlette, Django with async views, and Quart all speak ASGI. The question in 2026 isn’t whether to use ASGI — it’s which server to run it on.

Uvicorn: The Default Choice

Uvicorn is the most popular ASGI server by a wide margin. It’s fast, it’s simple, and it’s the server that FastAPI recommends in its documentation. Uvicorn uses uvloop under the hood, a drop-in replacement for asyncio’s event loop based on libuv. On Linux, uvloop provides a 2 to 4 times performance improvement for I/O-bound workloads.

Uvicorn’s worker management is straightforward. The —workers flag spawns multiple worker processes, each with its own event loop. For production, pair Uvicorn with Gunicorn as a process manager. Gunicorn handles worker lifecycle, graceful restarts, and signal handling. Uvicorn handles the ASGI protocol and request processing.

Hypercorn: The Full-Featured Alternative

Hypercorn supports HTTP/1, HTTP/2, and HTTP/3. It supports WebSockets, Server-Sent Events, and the full ASGI lifespan protocol. If you need HTTP/2 server push or HTTP/3’s QUIC transport, Hypercorn is the only option in the Python ecosystem.

The tradeoff is complexity. Hypercorn has more configuration options, more edge cases, and a smaller community than Uvicorn. Most applications don’t need HTTP/2 server push. For the 99 percent of Python web applications running behind a reverse proxy like Nginx or a CDN, HTTP/2 and HTTP/3 are handled at the edge, not the application server.

Gunicorn With Uvicorn Workers

The most common production setup in 2026: Gunicorn as the process manager with Uvicorn workers. Gunicorn provides battle-tested worker management, graceful restarts, and integration with process supervisors like systemd and supervisord. Uvicorn workers handle the ASGI protocol.

The configuration is well-documented. Set the worker class to uvicorn.workers.UvicornWorker, configure the number of workers based on CPU cores (typically 2 to 4 per core for I/O-bound async applications), and let Gunicorn handle the rest.

Serverless and Edge

The serverless trend complicates the ASGI server question. Platforms like Vercel, AWS Lambda, and Cloudflare Workers don’t run a persistent ASGI server. They invoke your application as a function. Mangum and similar adapters bridge ASGI applications to AWS Lambda’s invocation model. The ASGI server runs per-request, not as a persistent process.

The Recommendation

For new projects: Uvicorn behind Gunicorn, behind Nginx or a CDN. It’s the well-trodden path with the most documentation, the largest community, and the fewest surprises. For projects that specifically need HTTP/3: Hypercorn. For serverless deployments: Uvicorn with Mangum or the platform’s native adapter.