From US-East to “deploy everywhere”
Five years ago, “multi-cloud” meant AWS, Azure, and GCP. Maybe a CDN or two. Peek over a developer’s shoulder today and that definition feels quaint.
Today’s applications run across dozens of providers. In addition to the traditional hyperscale clouds, builders and teams alike use Vercel for deployment velocity, Supabase for backends, and an exploding ecosystem of ‘Alt Cloud’ services—specialized providers for prototyping, code generation, AI inference, and more.
This behavior seems deeply embedded in the psyche of today’s builders: software can and should ship everywhere. Deployments used to mean some instances in us-east-1. Now it means functions at the edge, in specific countries, on specialized hardware—wherever your logic and data requirements dictate.
From my point of view, four things are pouring rocket fuel on this behavior:
Everyone is in innovation mode. Trying new services has never been easier!
The emergence of “local first”. WASM, hyper portable backends, local first movement in full swing, etc!
The rapid adoption and impact of AI IDEs. Dramatically more software being created, and running in parallel!
Agents are getting good. They’re hungry for bespoke, purpose-built infrastructure!
With these factors at play, it’s difficult to imagine that the availability zone of tomorrow is a data center or a single provider. It’s all over the place. Welcome to the “Everywhere AZ”.
And yet, the Splinternet Is real
While software is obliterating traditional boundaries and defying old assumptions, the internet itself is being carved up. This matters more than most people realize.
The internet was designed and scaled in the open. But with China’s Great Firewall, a new template was introduced that started to fragment the global internet around lines on maps. Russia passed laws in 2019 enabling complete internet isolation, and various countries across Asia and the Middle East have strict localization requirements. The EU, UK and US have introduced content regulations that are accelerating the fragmentation of what was once a borderless web. And with AI and intense personalization of services, it feels we’re on the cusp of a huge wave of rules, policies and regulations.
We’re transitioning from the “world wide web” to a collection of more regulated and controlled ecosystems, each with its own rules, platforms, and technical requirements. That means the flat, open internet upon which we built our mental models is becoming hierarchical, balkanized, and decidedly less flat, and it’s happening at the same time as an explosion in software and clouds.
So what’s a fast-scaling, AI-enabled team to do?
The missing network team
Traditionally, companies (especially regulated ones like financial services or telecoms) have solved these kinds of problems with private networks at the center. Essentially, don’t rely on the public internet, but instead create a controlled edge and strategically connect with the external providers, partners, customers, and sources that you validate and trust – or that you can prove meet the rules and regulations you’re subject to.
In addition to control, private networks have a lot of other benefits, especially around performance and cost. It’s amazing what a cross connect and consistent, large MTU sizes will get you!
But here’s the thing: modern companies often don’t have a private network. They don’t even have a network team. And honestly? They don’t want one.
That’s where we think providers like Datum can add real value.
Building for what’s next
We think the world will benefit if every agent (and by extension, builder, startup, and scaled company) has access to their own private network. We’re still early, but the crux of our approach boils down to arming each agent with access to the following:
Edge - Meet the internet and interact at scale with an intelligent Envoy proxy. Protection against Bots, threads, and DDoS included
Tunnels - Secure, flexible, peer-to-peer QUIC connections built with Iroh that land application traffic exactly where you want it.
Backbone - Deterministic, auditable internet “fast lanes” powered by a controllable SRv6 overlay with VPC style control and observability.
These concepts aren’t really new, but we think it’s critical that we modernize them (along with other concepts like meet me rooms and interconnection!) to meet the challenges and opportunities we face in a world disrupted by AI, and fragmented by regulation.
Instead of a finely crafted edge network that takes years to build out, we want to make virtual backbones available in minutes, not months.
Instead of connections built around an outdated VPN model, we want to see zero trust, secure tunnels built specifically for the AI-enabled web.
Instead of a physical private backbone built and maintained by hard to find network veterans, we want every agent to access that value on demand.
We’re in the early innings of executing on our vision, but it’s been a blast finding our way to the point where crazy ideas are becoming product reality. Clearly, we see a future where the internet is fragmented and parallel. We’re building accordingly.
Zac Smith is CEO of Datum. Previously, he co-founded Packet (acquired by Equinix) and has spent two decades building internet infrastructure for an increasingly distributed world.
__