Definitely something to watch out for.
From what I can tell, its quite a patchy (lol) landscape now.
Popular servers like Apache httpd and Tomcat do not adopt it yet.
AWS ALB does not support HTTP 3 afaict, but CloudFront does.
GCP ALB supports QUIC with a checkbox in the configuration, and it tells you the traffic to the backend is HTTP 1.1.
Actually it's fine if somewhere between the client and server side is already HTTP/3. That is where HTTP/3 is going to shine in comparison to HTTP/2 (and also HTTP/1.1) having that HTTP/3 solved the Head-of-line blocking issue which is inherent to how multiplexing/pipelining in single TCP connection works. The Internet is a lot more unstable with higher chances of loss packets compared to internally between the CDN and the LB say within the data centre or within the same infrastructure, hence HOL issue in both HTTP/2 and HTTP/1.1 is more likely to occur. Otherwise HTTP/2 is good enough and offering very good multiplexing performance under stable network. I'm not familiar with GCP, but with AWS, Cloudfront offers HTTP/3 to the Internet clients while communication between CF and AWS ALB can be via HTTP/2. Thus the performance is still there.
The toolset to debug HTTP/3 in my opinion is not as readily available as HTTP/1.1 and HTTP/2. It is also alot easier to debug HTTP/1.1 and HTTP/2 traffic by using Wireshark or just simply TCPDUMP while QUIC you will need specialised tools just to peek into the binary packets.
It may not be ideal from an end-to-end perspective, but actually in most infrastructures I have seen and designed, none of the traffic are really end-to-end. The server end normally ends at the CDN if you have a large scale design. Otherwise you cannot even perform any package inspection or the WAF is totally useless, if it is really end-to-end encryption between the client and the server nodes. In such design, it wouldn't matter a lot since the network between the CDN and the servers are normally internal, stable and high bandwidth, low latency networks and HTTP/2 would provide just as good performance under such ideal conditions.
Normally between the ALBs and the Apache server, just HTTP/1.1 with keepalive and pipelining would be very performant without opening new connections. HTTP/2 multiplexing would even be better. Apache HTTPd and Tomcat and Nginx support HTTP/2.
Reading
https://blog.cloudflare.com/http-3-vs-http-2/, you will notice HTTP/3 is not in all cases better than HTTP/2. Even if it is, it's marginal.
