Nvidia sells the GPUs that power the AI boom. Now it wants to sell the software that runs the buildings those GPUs live in. On September 30, Jacobs (NYSE: J) announced it was selected by Nvidia to deploy its Data Center Digital Twin at a large-scale U.S. AI research and development facility, under a three-year software-as-a-service agreement.
The platform is built on Nvidia Omniverse libraries, the company’s simulation and 3D framework, and it does far more than draw a pretty 3D picture of a building. It handles dynamic power-load balancing, energy forecasting, liquid-coolant leak detection monitoring, predictive maintenance, and operator training.
In plain terms: it’s a living, breathing software mirror of the data center, constantly updated, that can predict problems before they happen and help operators rehearse fixes before they touch a single real server.
What a Digital Twin Actually Is
The phrase gets thrown around a lot, so here’s the simple version. A digital twin is a real-time virtual copy of a physical thing: a building, a factory, a jet engine. Sensors feed it live data; software simulates what’s happening and what’s likely to happen next.
For an AI data center, this matters enormously. These facilities are among the most energy-hungry buildings on Earth. A single large AI training cluster can draw as much power as a small city. Keeping that power balanced, the cooling flowing, and the hardware healthy is a 24/7 job. A mistake can cost millions in downtime.
A digital twin lets operators see the whole system at once: which racks are heating up, where power is spiking, whether a coolant line is showing early signs of a leak. It forecasts demand so the facility can buy energy smarter. And it lets staff train on the virtual copy: practice a failure scenario in simulation rather than learning on the live, expensive real thing.
Jacobs’ EVP Amer Battikhi put it this way: the project reflects “the growing role of digital twins in helping operators manage critical infrastructure environments.” Corporate phrasing, sure — but the underlying point is sound. When the infrastructure is this complex and this expensive, flying blind isn’t an option.
Nvidia’s Second Act
Here’s the deeper story. Nvidia built its empire selling the picks and shovels of the AI gold rush: the chips. This deal is about selling the operating system for the mine.
Jacobs describes its twin as an “intelligent operating layer” for AI agents, and that phrasing is worth pausing on. It suggests a future where software agents don’t just answer tickets and summarize documents: they schedule power, reroute cooling, and orchestrate the physical plant. The data center becomes something an AI can operate, not just something humans monitor with dashboards.
That’s the quiet second act of the AI buildout. The first act was raw compute: buy more GPUs, build more halls. The second act is efficiency software that makes the same GPUs do more work per watt. Energy and cooling management is becoming the competitive moat of data centers, because power, not chips, is increasingly the scarce resource. Every AI lab on Earth is hunting for megawatts; the ones that squeeze more out of each megawatt win.
The market seems to like the trajectory. Nvidia shares traded near $233–235 on September 30, up roughly 3% in September, on track for a third straight monthly gain and within 2% of the May record close of $236.45. Investors are pricing in a company that’s expanding from hardware into infrastructure software, and infrastructure software has much nicer margins.
Why This Matters Beyond One Facility
This is Nvidia deploying the technology at its own R&D facility, eating its own cooking, as they say. If the twin proves out at a large-scale AI research site, it becomes a reference installation for every hyperscaler and enterprise building AI data centers next. And there will be many of those: the physical AI wave (humanoid robots like Digit, driverless freight) all runs on data centers that need managing.
There’s also an environmental angle worth celebrating. Smarter power-load balancing and energy forecasting mean less wasted electricity. Predictive maintenance means hardware lives longer instead of failing early. At the scale of AI data centers, even single-digit efficiency gains translate into enormous amounts of energy saved. That’s energy that never has to be generated at all.
The three-year SaaS structure matters too. This isn’t a one-time consulting gig; it’s software with a subscription, and subscriptions are how infrastructure companies compound. Nvidia is learning the enterprise software playbook, and it’s starting with its own house.
The Takeaway
Digital twins have been a promising idea for a decade. What’s new is the combination: AI-scale data centers creating the pain, Omniverse providing the simulation muscle, and AI agents arriving as the eventual operators. Nvidia hiring Jacobs to twin its own R&D facility is the signal that this stack is leaving the lab and entering the machine room. The AI buildout isn’t just about bigger chips anymore. It’s about smarter buildings.
