Listen to the audio version of this article (generated by artificial intelligence).
AI agents have learned how to work.
They haven’t learned how to leave the office.
Right now, most of the inference demand driving the AI infrastructure boom is happening within company walls. The bank runs proxies on internal risk profiles. A logistics company uses agents to optimize its supply chain. The retailer deploys agents to manage purchases within systems it already controls.
This demand and spending are real. Trade in chips, memory, networking, storage and energy has already begun.
But it is still mostly contained.
VintServ has just begun working on the layer that could allow it to appear.
Why Vint Cerf’s DNSid Project is Important for AI Agents
Surf helped build the modern Internet. He is one of the architects of TCP/IP – the basic protocols that made the Internet possible.
After more than two decades at Google, he’s now turning to his next project: identity infrastructure for AI agents working on the open internet.
like TechCrunch reportedSurf is now offering advice Innovation laboratories On something called DNSid – which is basically a passport system for AI agents. The idea is to associate each proxy with a verified domain name and use cryptographic evidence to show where it came from, who authorized it, and who is responsible for what it does.
Surf helped solve the first formatting problem on the Internet. He is now working on the next stage.
Trust issue hinders the proxy web
The agent economy is still mostly confined within company walls.
Take a purchasing agent for a retailer, for example. They work within the retailer’s own systems – looking at their own inventory databases, working within their supplier relationships, and reporting to their IT team.
Most enterprise AI agents in production today work this way, within the confines of a single organization. That’s why current infrastructure demand, while already huge, may only represent a small fraction of what’s to come.
The largest use cases — and those that will generate the most demand for computing — involve agents crossing organizational boundaries.
- Purchasing agent who negotiates directly with the supplier’s agent.
- A financial agent deals in real time with a bank agent.
- A logistics agent coordinates across dozens of different conveyor systems simultaneously.
- A healthcare agent pulls verified records from multiple hospital networks to inform a treatment decision.
They all require agents to operate across the open Internet – interacting with unfamiliar systems, on behalf of humans who are not monitoring their every move.
But this world has a trust problem.
When an agent appears somewhere on the Internet today, there is no reliable way to verify who sent it, what it is authorized to do, or who is responsible if something goes wrong. Without a solution, the highest value proxy use cases simply cannot be safely deployed at scale.
That is the missing layer.
What happens when AI agents can operate over the open internet?
Before common Internet protocols, computer networks were islands. Each organization ran its own system. These systems were valuable, but they were not able to talk to each other easily.
Once a common standard allowed those networks to communicate, the Internet became a global market. Every person and organization suddenly needed to communicate. The demand for routers, cables, servers, and all the physical infrastructure underneath has become essentially limitless.
The proxy web is approaching a similar moment. Today’s enterprise agent deployments look a lot like those isolated networks of the 1980s: valuable, growing, but essentially contained. Once a common identity standard enables agents to operate across organizational boundaries, with accountability included, the addressable market for agent infrastructure will likely expand dramatically.
Every cross-enterprise workflow becomes a potential agent-to-agent interaction. Every government service, financial transaction, and logistics chain becomes a candidate for agentic automation – each requiring heuristics, memory, networking, and storage that currently fall outside the demand forecasts from which most investors operate.
The infrastructure thesis does not change. Size does it.
Investment Implications: The heuristics market is getting larger
The physical infrastructure – accelerators, high-bandwidth memory, optical networks, power, cooling, and storage – remains the core of commerce. This demand continues to grow, and the companies that supply it report it in their earnings quarter after quarter.
Surf’s work adds a new layer. If agents get a reliable way to identify themselves online, the demand story will move beyond an organization’s internal workflow and into the open Internet. This would create a larger market than most investors imagine.
Cloudflare (network) may be the cleanest way for the public market to run that identity and routing layer. It’s already present in developers’ Internet traffic flow, security, authentication, and infrastructure. Its tools are increasingly designed for a world where agents need to discover services, prove their identity, and transact across the web. If a proxy web bypasses an organization’s firewall, Cloudflare could become one of the layers of trust and routing underneath.
Moreover, the names of the broader infrastructure that provides the physical substrate upon which each agent workload runs are the same names that directly benefit when the agent economy expands. More agents working across more boundaries means more inference calls, more memory consumption, more network traffic, and more power consumption.
The market understands institutional agents.
You don’t fully price online agents – or their larger workload.
Bottom line: AI agent identity can unlock the proxy web
Agents actually work within companies.
The greatest opportunity begins when they can work between Companies.
This requires identity and trust – a way to know who sent the agent, what it is allowed to do, and who is responsible if something breaks.
Vint Cerf is working on that layer now.
If successful, the superinference cycle would expand to include the open Internet.
The physical infrastructure that supports this market is already being built. The companies supplying it have already reported the demand. and Names best placed for the next phase of this expansion – those that the market has not yet fully found – are exactly what we have been tracking.
This is trade. And it’s gotten bigger.
