Long before the era of the hyperscalers and neoclouds, and decades ahead of the AI data center boom, there was Equinix.
Described as a colocation facility, Equinix provides data center space for over 10,500 customers, offering space, power, cooling and security for companies that need a place to house their servers, routers and storage systems.
That experience, dating back to the 1998 dot-com bonanza, has given Equinix an important and expanding role in the artificial intelligence data center market, where hyperscalers are expected to spend over $5 trillion by 2030, according to Goldman Sachs Research. On Wednesday, Equinix inked a deal with Nvidia that gives customers a flexible way to run their AI models on open-source cloud platform Together AI.
Equinix's stock price is up 33% this year, beating all of megacap tech and lifting the company's market cap to $100 billion. It's by far the most valuable data center real estate investment trust (REIT), ahead of Digital Realty, which has a market cap of $68 billion.
Hyperscalers like Amazon and Google, and neoclouds like CoreWeave are collectively spending hundreds of billions of dollars a year on infrastructure, with an outsized amount of that going to serve major AI labs like OpenAI and Anthropic. Equinix, by contrast, rents out space in its facilities to customers of all shapes and sizes that can run servers on a variety of compute platforms, from Nvidia to AMD.
At an event in San Francisco, California, on Wednesday tied to the Nvidia deal, the chipmaker's CEO, Jensen Huang, spoke by video with Equinix CEO Adaire Fox-Martin. Huang said that the location of Equinix's facilities allow you to be "close to where the action is, where all the sensors are." And because the architecture is distributed, "you could both simultaneously be close and be far away," he said.
Equinix is one of the oldest names in the data center business. Its 281 legacy colocation facilities span 77 metropolitan areas across six continents, according to Maryam Zand, a vice president at the company who runs its AI ecosystem strategy.
"The companies you already use are all running on us," she says.
Financing details for Wednesday's deal weren't disclosed. Zand said Together AI, which offers access to 200 open-source models, will be the seller of record, billing its end customers who use the new program. Called Equinix Inference Exchange, it will be available in the first quarter of 2027.
Focus on inference
While AI training teaches the model to learn from patterns in large amounts of data, inference refers to the decisions that are made based on new information. As AI evolves from simple call-and-answer chatbots to more complex agentic apps, inference has become more critical than training. That means workloads need to run on a wider variety of chips such as central processing units (CPUs), instead of relying solely on general-purpose workhorse graphics processing units (GPUs).
Equinix has mainly stuck to its old playbook of building smaller data centers in locations close to city centers, positioning them as network interconnection hubs. Digital Realty, by contrast, began to place greater emphasis on large-scale facilities for hyperscale customers.
For xScale, the side of Equinix's business that serves hyperscalers, most facilities are on the smaller side. They're under 100 megawatts, while some AI data centers are being measured in the gigawatts.
Vlad Galabov, a longtime data center analyst, said Equinix was "too slow" to react to the gigawatt-scale demand.
"There was new, hungrier guys who came in to start to build data centers for some of these burst projects that just quickly spun up," said Galabov, host of the AIDC Debate podcast. "Equinix, Digital Realty, all of the colocation companies are now seeing the need to plan more strategically."
Equinix's announcements Wednesday also included Equinix Fabric One, a new connectivity service meant to simplify networks that operate using multiple clouds and AI models.
Zand described the announced program with Nvidia and Together AI as an "inference platform as a service," that enables customers to connect to a variety of different clouds and providers, run inference on open source models, and "optimize their tokenomics" — or reduce costs.
She told CNBC that Equinix data centers are optimized for Nvidia's B300 Blackwell Ultra GPUs, but it also has some liquid-cooled facilities where customers can use the newer Vera Rubin chips.
Where Equinix has a particular advantage is its urban locations, which are important for inference and the high-speed communication required between servers and end users. By 2030, inference will make up half of all AI compute and 30% to 40% of total data center demand, according to McKinsey.
"The changes are very rapid in today's environment, in the whole AI ecosystem," Zand said. "Customers need to be able to move as the market moves."
Not all investors are excited about the story.
Short seller Jim Chanos told CNBC's "Closing Bell" in May that he was betting against the stock, as well as Digital Realty, saying "they're not great businesses" and are "not very profitable."
"They're very low return on capital businesses, very capital-intensive businesses, and they don't grow that fast," he said.
He added that, "the legacy data center companies are a distinct difference from the data centers that are being built for AI."
In the latest quarter, Equinix reported a 16% increase in revenue from a year earlier to $2.63 billion. CoreWeave, one of the leading neoclouds, saw revenue more than double to $2.58 billion. However, Equinix reported net income of $477 million in the period, while CoreWeave lost $626 million.
Galabov said Equinix is "super diversified" with a huge amount of general purpose compute and services across a very wide client base.
"They are not exposed to an AI bubble risk," Galabov said. "That means you're missing out on some of the boom. It's just inevitable. High risk, high return."
WATCH: Jim Chanos says Equinix and Digital realty are 'not great businesses'