Data centers are rapidly evolving into AI factories – News

AI isn’t just adding horsepower to data centers — it’s redefining what they are built to do.

The shift is structural, not incremental. Traditional racks and retrofit upgrades can’t keep pace with large-scale AI workloads, forcing a redesign of power, cooling and network architecture from the ground up. Companies such as GlobalAI are building around that reality, working closely with Nvidia to deliver single-tenant, sovereign environments that give enterprises advanced intelligence without surrendering control, according to Sami Issa (pictured), director and chief executive officer of GlobalAI Cloud.

“Once they have seen your secret sauce, it’s embedded within their weights and it’s near impossible to extract,” Issa said. “We envisioned that a couple of years ago. We started the company to cater to fill that gap of creating [this] infrastructure dedicated for our customers, and that’s what we do. It’s single tenet; we’re not a cloud. I keep telling everybody, we’re not a cloud.”

Issa spoke with theCUBE’s John Furrier for theCUBE + NYSE Wired: AI Factories – Data Centers of the Future interview series, during an exclusive broadcast on theCUBE, News Media’s livestreaming studio. They discussed how GlobalAI is helping transform data centers into sovereign, Nvidia-powered supercomputing infrastructure built for enterprise control and scale.

How data centers are becoming AI factories

What’s emerging now looks less like traditional colocation and more like vertically integrated compute plants. Nvidia’s systems are not incremental GPU add-ons; they are rack-scale supercomputers that demand redesign at the facility level. That shift changes who can participate and how quickly enterprises can deploy advanced AI workloads, Issa explained.

“Ninety-five percent of data centers in the world are running on eight to 12, maybe 40 kilowatt racks,” he said. “We are building 150, 250, going to a megawatt per rack. These are super computers. Each GB300 is an exaflop. A couple of years ago, you and I would dream of an exaflop machine. Now, we have many, many exaflops within our data centers because each rack is an exaflop.”

Those densities require a new kind of engineering discipline. Liquid cooling, power provisioning and facility design now operate at scales that resemble utility infrastructure. Nvidia’s rapid hardware cadence, including GB200 and GB300 systems with Vera Rubin on deck, compounds the complexity for operators trying to keep pace, Issa noted.

“We have the GP200s and GP300s running,” he said. “Vera Rubin is coming up in April, and we are running like mad to keep up. We are very, very close to Nvidia. They have been a fantastic partner to us. We can because we are technically strong; we have a very strong technical team. We can build these liquid-cooled data centers ourselves. We can bring up the bleeding-edge technology online as I’ve mentioned.”

Scarce expertise meets surging enterprise demand

The partnership with Nvidia goes beyond supply agreements. Bringing these machines online requires access to technical expertise and co-design support. The scarcity of qualified talent underscores how early the market still is, despite the surge in capital and demand, Issa mentioned.

“I was talking to an investor a couple weeks ago, and he said there are 300 individuals in the world who can actually build at scale GB300s today,” he said. “That is the best, the most active description of the market today. This is not 3,000 — 300 across planet Earth.”

Demand reflects that imbalance between supply and capability. Enterprises are signing multiyear contracts for dedicated capacity, often insisting on fully isolated facilities rather than shared environments. The appetite for AI compute is no longer speculative; it is contractually locked in, Issa emphasized.

“Demand is unlike anything I’ve seen in my entire career,” he said. “Demand is unbelievable. If I go to the market with 500 megawatts, it’ll be gone within a couple of days. If I go to the market with a gigawatt, it will be gone within a week.”

Here’s the complete video interview, part of News’s and theCUBE’s coverage of theCUBE + NYSE Wired: AI Factories – Data Centers of the Future interview series:

Image: News