Why Cloudflare Is the Perfect Infrastructure for Building AI Applications | Hacker News Reader

Why Cloudflare Is the Perfect Infrastructure for Building AI Applications

reconfigured.io

31 pointsmxschumacher10 comments

Wow! Cool to see the piece I wrote sparked interest. Happy to elaborate on different topics

Are there any self-hosted offerings similar to Durable Ojects? I like the idea of abstracting synchronization like this but enjoy being cloud agnostic.

Have a look at membrane - https://www.membrane.io

Maybe a Valkey cluster for simpler use cases? Although that's more like Workers KV than DO.

I wonder if temporal [1] could be used to create DO. Had anyone here tried it?

[1] - https://temporal.io/

Although I have not used it as DO, I have been using it for durable workflows for a few years. The cached results of the workflow activities could give the feeling but the latency to read and write would be a dealbreaker for many. There is a lot of overhead in writing the interface between application code and Temporal workflow (for example, you can have conflicting object IDs that can be configured).

Self hosting Temporal with a Postgres datastore is usually how I deploy it and it works great, but at that point I'd just write it to a Valkey store if I just wanted a easy DO interface.

I often come across Cloudflare.

The dev and ai tech seems interesting but also very opinionated, proprietary and, I assume, loved by free loading hobby devs but unknown and in-adoptable by scale ups and larger organisations.

Am I wrong? I think the fact that they’re introducing a container service reinforces my suspicion.

I usually work for small and medium sized organizations and Cloudflare is very often used for those. Although they started as a CDN and security product, their serverless stack is probably the best in the world (Workers, KV, Durable Objects, etc.) and R2 has some of the cheapest S3 storage for sharing files.

Their developer economics (ease of use) are amazing compared to the legacy clouds, especially AWS. What would take a dozen services and hours of config on AWS takes a few minutes on Cloudflare and probably one or two services. It's all highly abstracted and optimized for developers instead of managers. That's probably the same reason they're not as big at the large scale enterprises. AWS and Azure benefit from the kind of corporate billing and administration system that only large enterprises use and want.

They're basically the Vercel of the network infra space (or rather, Vercel is the Cloudflare of web hosting). Both take traditionally hard to manage networked microservices and package them into simple APIs that are highly abstracted, relatively cheap, and easy to scale. But they don't really make money from you until you're huge. So the small and medium customers are probably loss leaders for them, and by the time you get into enterprise territory, you're too locked in to switch.

They both typically have low startup costs and higher enterprise costs than the legacy clouds. But even then, some particular services like Workers and R2 are just awesome in their own right. IMHO Workers is the coolest thing to have happened to the web in the last like 20 years.

They will need to find their niche for sure. As a consultant with larger organizations, CloudFlare has never been mentioned nor recommended given most are already on one of the big cloud providers and those cloud providers already have Dev/AI offerings that are somewhat integrated. More crucially, these providers also offer support and sometimes investment to push deals through.

CF is definitely popular with bigger orgs too. Plus their sales team is pretty intent on pushing their paid offerings

Reply on news.ycombinator.com