Serverless Startups and AI in April
April 2025
Happy April everyone! We are in spring and there is still SOME snow (see the above picture). Google had Cloud Next and I was able to participate this year which was fun. Of course, I was showcasing Cloud Run with GPUs.
There were a number of announcements that you can see here but that’s not what you are here for. You are here to learn more about serverless computing and what happened in April.
Reading the news this month, I saw a number of articles on AI (of course), Wasm, and a few serverless startups. I also saw a few well established companies with stock symbols invest heavily in serverless. We even had a big announcement regarding serverless containers!
This excites me as it shows that serverless is getting its flowers ( receiving it’s deserved attention ). As I have stated before, I have been a major fan of serverless for years. The term serverless didn’t take off until circa 2015 with the launch of AWS Lambda. Seeing serverless being taken more seriously as a technology makes me happy.
But let’s jump right into the news.
Cloudflare meets… Containers?
I have spoken at length about serverless containers. It is undeniable that the container is the basic compute unit for the cloud. Yes, VMs still exist (and will continue to exist) and WASM is becoming a promising technology but containers are the core of cloud native technology.
Now again, when many people talk about serverless, they are thinking about functions. Obviously, serverless is more than functions. Well imagine my surprise when I learned that Cloudflare felt the same way.
Yes, containers are coming to Cloudflare. Cloudflare Workers, which has historically only supported a FaaS-style architecture, will be supporting containers in June of 2025.
Cloudflare has never really ventured much out of the FaaS world. It is amazing that they are embracing containers. They are recognizing the portability of containers. By supporting containers, they are entering the “cloud native” world and joining the “serverless container” world.
It will likely open a lot of opportunity too. We may see serverless GPUs evolve there, for example.
What the WASM?
WebAssembly (WASM) promises to be the next step in cloud binary packaging. Fermyon is positioned to be THE WASM company right now. I got to see them at our Google Cloud Next 2025 Day Zero event, Containers @ Next. They displayed something really cool on GKE.
They showcased the Spin Framework. This was created by Femryon but is now a part of the CNCF. It is a framework to deploy and manage WASM apps and can be run on Kubernetes.
One of the things that has been holding WASM back is that it lacked a reliable autoscaler. It is kind of like what we saw with Docker in late 2013. People saw the containers and thought that they were great but what happens when you need to run 100 of them?
Spin Framework on Kubernetes can help serverless WASM expand into new territories. Kubernetes already has great market share and the CNCF has started to explore WASM projects, it only makes sense to use Kubernetes as the platform for WASM.
Microsoft recently created a microVM for WASM workloads. This is a somewhat different approach from using Kubernetes but I think we need multiple ideas.
Serverless Joins the MCP
In case you haven’t been checking the tech news of the past few weeks, Model Context Protocol (MCP) has been the talk of the tech town. A quick summary, it is a protocol that allows developers to communicate with agents more easily. It was created by Anthropic back in November of 2024 but it has really taken off over the past month or so.
This graphic is a great demonstration of how this is supposed to work.
Before MCP, all of your agentic applications would have to make individual API calls to your LLM. MCP operates in a client-server model so your apps just need to connect to the MCP service and that server connects to the LLM.
Naturally, the server needs to live somewhere. It’s not a server in the traditional sense either. In other words, it doesn’t need to be an actual piece of hardware or even a VM.
Our friends at Cloudflare announced a serverless MCP server. I think as time progresses, we will hear more about serverless MCP and I will gladly report it.
As “Inference-as-a-service” startup!
Let’s do a speedrun of startups
Inferless, a serverless GPU company has launched. Read the interview with the CEO here.
NScale launches a serverless AI Platform for inferencing. It is based in Europe so this serverless AI concept is going global.
OVHCloud launched a serverless AI endpoint service.
Closing Thoughts
Serverless is taking over. I was pleased to read this post in InfoQ that talked about work my buddy did when he was at Capital One. They are ALL IN on Serverless. Banks tend to be very hesitant when it come to adopting the cloud and not only did Capital One go all in on the cloud, but they went all in on serverless.
I think we are going to see more and more serverless. I think MCP and even Google’s own A2A protocol will result in a rise of more serverless GenAI use cases.
SaaS companies are already evolving from monolithic to serverless architectures. As they begin to adopt agents, I fully expect for them to adopt serverless at a faster rate. We are hearing people clamour for more agentic AI and LLM use cases. DeepSeek runs amazing on serverless platforms as do other LLMs.
I fully anticipate more startups coming into existence dedicated to serverless AI. As more and more happens, I will gladly report it here!
—Photo courtesy … Me—