Equinix was building data centres before AI became the industry’s favourite word. Now, the company is betting that its old strengths could become valuable again.
For more than two decades, Equinix has quietly provided the physical infrastructure behind the internet economy.
Its colocation facilities give thousands of businesses a secure place to install servers, networking equipment and storage. Customers pay for the space, electricity, cooling and connectivity they need without having to build and operate entire data centres themselves.
That model helped Equinix survive the dot-com crash, the cloud revolution and several technology cycles.
Now comes perhaps its biggest test yet: the AI infrastructure boom.
A Different Kind of AI Data Centre Play
The world’s biggest cloud companies and newer AI-focused providers are committing enormous sums to computing infrastructure.
Goldman Sachs Research estimates that hyperscaler spending on data centres could exceed $5 trillion by 2030.
Equinix is taking a different route.
Rather than concentrating exclusively on enormous facilities designed for a handful of hyperscale customers, the company operates a global network of smaller colocation sites that serve businesses with widely different computing requirements.
Its traditional estate includes 281 facilities across 77 metropolitan areas on six continents, according to Equinix.
That footprint could become particularly useful as AI workloads move beyond the giant facilities where models are trained.
The Next AI Battle May Be About Inference
Training an AI model is only part of the equation.
Once a model has learned from vast quantities of information, inference is what happens when it responds to new requests, makes predictions or performs tasks for users.
And as AI applications become more sophisticated, inference is expected to become a much larger part of overall computing demand.
That changes the infrastructure equation.
AI workloads do not necessarily need to run exclusively on the powerful GPUs traditionally associated with model training. Depending on the task, other processors—including CPUs—can play an important role.
For Equinix, that creates an opportunity.
Its facilities are often located close to major population centres and network exchanges, allowing computing resources to sit physically closer to businesses, devices and users.
When an AI application needs an immediate response, distance matters.
Nvidia Partnership Targets Open-Source AI
Equinix announced a partnership with Nvidia that will allow customers to access AI models through Together AI, an open-source AI cloud platform.
The planned service, called Equinix Inference Exchange, is scheduled to become available in the first quarter of 2027.
Together AI will handle customer billing and provide access to a range of open-source models.
Equinix describes the offering as an inference platform delivered as a service.
The idea is to give businesses greater flexibility to connect with different clouds and AI providers, select models and computing resources, and manage the cost of running those workloads.
Equinix executive Maryam Zand said the platform could also help customers optimise their spending as AI computing prices and requirements continue to change.
Why Location Could Become Equinix’s Superpower
Nvidia CEO Jensen Huang highlighted another potential advantage during an event connected to the announcement.
Equinix’s facilities are distributed across many locations, meaning customers can place computing resources closer to the places generating data while still connecting to distant infrastructure when necessary.
That matters as AI moves into applications that rely on enormous numbers of sensors, devices and real-time interactions.
Consultancy McKinsey expects inference to account for roughly half of AI computing by 2030 and potentially 30% to 40% of overall data-centre demand.
If that forecast materialises, proximity could become just as important as raw computing power.
But Equinix May Have Missed Part of the First AI Wave
There is a catch.
The AI infrastructure boom has been dominated by enormous facilities requiring power capacity measured in hundreds of megawatts—and increasingly, gigawatts.
Equinix’s hyperscale-focused xScale facilities have generally been much smaller than that.
Data-centre analyst Vlad Galabov argues that Equinix and other traditional colocation companies were slow to recognise just how quickly demand for massive AI facilities would emerge.
Newer competitors moved aggressively into that market, building large sites specifically for AI workloads.
That has forced established operators to rethink their strategies.
Equinix may have been early to data centres but late to the scale of the AI buildout.
Its Stock Is Telling a Different Story
Investors, however, have rewarded Equinix this year.
The company’s shares have risen about 33%, taking its market value to roughly $100 billion and making it the largest publicly traded data-centre REIT by market capitalisation.
Digital Realty, another major colocation operator, has a market value of about $68 billion.
But not everyone believes traditional data-centre companies are positioned to capture the AI boom.
Investor Jim Chanos has argued that companies such as Equinix and Digital Realty are capital-intensive businesses with relatively low returns on invested capital.
He also sees a fundamental difference between the legacy data centres operated by these companies and the enormous facilities being built specifically for AI.
Equinix Has One Thing AI Start-Ups May Not
The contrast with newer AI infrastructure providers is striking.
CoreWeave, one of the leading AI-focused cloud companies, recently reported quarterly revenue of $2.58 billion, more than double its year-earlier level.
But it also recorded a $626 million net loss.
Equinix generated $2.63 billion in quarterly revenue, up 16% year over year, while reporting $477 million in net income.
The difference illustrates two very different approaches to the AI opportunity.
One is built around aggressive expansion and concentrated AI demand.
The other is built around diversification.
Galabov describes Equinix as highly diversified, with customers using a broad range of computing technologies for purposes that extend well beyond AI.
That may limit its exposure to an AI bubble—but it also means the company may capture less of the upside if the current boom continues.
The Real Bet
Equinix does not need AI to replace its existing business.
It needs AI to become another major workload running through its network.
That is a much more measured bet.
The company is betting that the AI economy will eventually require more locations, more connectivity, more computing flexibility and faster access to users, not just enormous GPU campuses in remote areas with abundant power.
If inference becomes as important as analysts expect, Equinix’s decades-old strategy of putting computing infrastructure close to networks and customers could suddenly look less old-fashioned.
The company may not own the biggest AI data centres.
But if AI becomes increasingly distributed, Equinix could own something just as valuable: the places where those systems connect to the real world.