- Equinix’s Arun Dev argues it’s time to upgrade an old mantra: Now, “the network is the agent”
- New Equinix Fabric Intelligence turns the network into a multi-agent system that can understand intent and take action
- As cool as it is, it could take another year or two for agentic AI to really catch on, Dev said
It’s one thing to say that networks will be critical foundational pillar underpinning AI and the agentic revolution. It’s quite another to assert that the network itself needs to become an intelligent agent.
But that’s exactly what Arun Dev, Equinix VP of Interconnection, said needs to happen and exactly what the colocation provider is trying to achieve.
Back in the 1980s, Sun Microsystems coined the phrase “the network is the computer.” Dev said that slogan needs an update.
“In the AI era, we believe the network is the agent,” he told Fierce. And that change means updating the network so it can become “an agent in its own right,” one capable of understanding intent, enforcing policies and adapting to growing connectivity needs.
Hence the launch of Equinix Fabric Intelligence. The company’s fabric offering itself has been around for more than a decade. But Dev said Equinix has been working since the launch of model context protocol (MCP) late last year to build in new agentic capabilities.
According to Dev, Fabric Intelligence is a multi-agent system that uses specialized agents to perform specific tasks. So, for instance, there are lightweight models being used for pattern recognition and large language models that provide the conversational interface.
The end result is a system that removes a lot of manual processes around provisioning and helps fill in enterprise skills gaps in deep networking expertise.
To be fair, Lumen Technologies has been banging the drum about the need for more adaptable and intelligent networks for a while now. It’s already simplified provisioning and is working with Blue Planet on agentic AI implementation across multiple domains.
“I think the network needs to have fundamentally net new capabilities in the world of AI,” Lumen CEO Kate Johnson told Fierce at MWC Barcelona last month. “I think there’s a need for programmability.”
But what Equinix is doing feels…a little different.
Putting the network to work
So just what does this intelligence look like in practice? Asked for an example, Dev pointed to the experience of one customer who kept seeing weekend spikes in utilization that maxed out their capacity.
Their old process called for manually alerting a tech, having them spin up more links, and then deciding whether to leave those links live for the next spike or turn them down to save money. But with Fabric Intelligence, that customer can now use an agent to automatically spin capacity up and down when spikes happen and eventually taper off.
Right now, Dev said a human is still in the loop. When the agent sees traffic spike, it asks whether the human would like to provision more capacity. It proactively asks two other questions: should it turn down bandwidth when utilization tapers and should it take the same action in the future if this happens again.
Asked about the difference between this system and older policy-driven actions, Dev pointed to the natural language interface. That, he said, makes it much easier for someone without deep networking expertise to get the job done. And given there’s currently a massive skills gap in this area, that matters a lot for enterprises.
But this raises the question: who bears the blame if a Fabric Intelligence agent makes a mistake?
“There’s a shared responsibility model here, similar to how cloud security works,” Dev said. “Enterprises own the policy decisions. We own the system that’s performing within those policies and their clear boundaries.”
On the horizon
Dev acknowledged that customers today fall on a spectrum, with some wanting AI to have read-only access and others confident enough to let it take action. He added he thinks the industry as a whole is still 12 to 24 months away from agentic AI becoming truly trusted and prevalent.
But once that threshold is crossed, things start to look a lot more interesting.
He speculated that in two to three years, agents from different vendors in the network will start to talk to one another to problem-solve. So, for example, if an application slows down, you might end up with an Equinix agent collaborating with Cisco and Palo Alto agents to solve the problem automatically.
That might sound far-fetched in a world where trust in AI is still hard to come by, but Dev likened the situation today to the dawn of the public cloud. When that first got off the ground, Dev said it was a struggle to convince enterprises to move their workloads over.
“So similar to public cloud adoption, this will take a bit of time to adopt he concluded.