Last week we put Datum Compute, our microVM runtime, into production. Riding along with it was Galactic VPC, a "go anywhere" virtual backbone built on SRv6.
Over the last few months, with a big chunk of our team focused on launching a microVM service, folks often asked why a network cloud was spending so much time on compute. The short answer: a backbone is only as useful as what you can attach to it, and we needed something super cool to attach.
The vision
Two years ago we set out to help every builder and agent take advantage of the network tools that hyperscalers use internally. Large clouds and regulated enterprises run private backbones to cut costs, lower latency, improve uptime, and keep traffic off the public internet. Why shouldn’t everyone get those benefits?
We wanted everyone to be able to turn on a private “go anywhere” backbone in minutes and attach things to it: traffic protection, policy, interconnection, network functions, inference routing, etc. That's the vision behind Galactic VPC.
Why Galactic needed Compute, and vice versa
Alas, a private backbone on its own isn’t all that interesting. This is especially true if you’re a born in the cloud developer era. Why fly private when commercial is nice enough?
Private networks get interesting when you can see the “before and after” effect with real workloads: an app at the edge, a gateway, a function that inspects or routes traffic. Without those things nearby, you're asking people to adopt an amorphous concept, not a tool.
Remote computers without a network attached are useless. You can’t talk to them, and they can’t talk to everything else. In order to make the true value of Datum evident, we would need to bring along an entire suite of networking features.
While we always wanted to build a compute feature (to help users push intelligence to the edge) we moved it forward and made it our focus so that we could mature our core networking value alongside a primitive that every developer and agent wants: affordable, secure microVMs that start in milliseconds. And, of course, every workload comes up already attached to a private backbone.
Chicken, meet egg!
A whole-company focus
Once we made the decision to build Datum Compute, it became a company-wide focus. There was a ton of work to do around networking and orchestration (of course) but also a diverse set of platform features to effectively support both customers and our own operations.
For example, Compute runs on powerful bare metal servers across multiple Datum regions. While microVMs are incredibly efficient at using the underlying infrastructure, we still need to monitor usage, meter everything around the service, flow data into our billing system, and deal with projects that are either suspicious or past due.
While this is just one part of the puzzle, we used Compute as a perfect excuse to extend Milo to integrate with OpenMeter, built out a service catalog (to include pricing), improved our feature flagging, and designed a suspension process for risky projects.
Galactic VPC is why it’s so fun to share Compute with users. With your own private backbone you can:
- Bring diverse connections onto one network: Tailnets, Iroh tunnels, cloud VPCs.
- Manage and secure traffic with prebuilt gateways, ALBs, NLBs, GSLBs, and DNS.
- Deploy your own apps at the edge with Datum Compute.
- Connect private datacenters, neoclouds, and public clouds into one private network.
The payoff is the same one the big clouds get: lower egress costs, reduced latency, and a smaller security blast radius because traffic stays private. Peter covers more details about SRv6 in his recent Galactic VPC post.
Try it!
Datum Compute is live in every new project and free during beta. Spin up a workload and it lands on your own backbone automatically. Over the coming weeks we'll add more ways to work with Galactic VPC directly. We'd like to hear what you attach to it first.