ZSquared

AI demand comes in two shapes, and both run on power. Training is how a model learns. It means large, concentrated jobs running on thousands of chips for weeks. Inference is the model doing its job: answering questions, recognizing images, running agents. Each request is small, but the volume grows with every user and every new application. Both need the same foundation: energized power, cooling, density, and connectivity. That's the capacity Z Squared is built to deliver, with AI and high-performance computing as our target workloads.

AI demand comes in two shapes. Training: huge, concentrated jobs to build a model. Inference: answering requests all day, growing with every user. Both need energized, cooled, connected capacity. Our target workloads are AI and high-performance computing. $ZSQR

"Energized" and "AI-ready" aren't the same thing. An energized site has power flowing. An AI-ready site adds everything modern AI workloads need on top of it: • Cooling, with a path from air to liquid as rack densities climb • Density: racks that draw far more power than typical enterprise servers • Fiber: fast, redundant connectivity • Physical security and monitoring • Operations staff and remote hands on site At Z Squared, power comes first. Everything else is engineered to the workload. #AIInfrastructure #DataCenters #LiquidCooling

What turns an energized site into AI-ready capacity? Cooling, with a path from air to liquid Density: high-power racks Fiber: fast, redundant connections Security and monitoring Operations people on site Power comes first. The rest is engineered to the workload.

Our model in one line: your compute, our everything else. Customers bring the GPUs, servers and workloads. We deliver the power, cooling, density, fiber, security, remote hands and operations underneath them. We don't build the compute layer. We build what it runs on. $ZSQR

Two ways to bring AI capacity online: Greenfield: find land, request a grid connection, wait. Queues for new data center campuses can run 3 to 5 years. Acquire and convert: buy a site that's already energized, then re-engineer it for the workload. We chose the second. $ZSQR

A closer look at our Union County Campus in El Dorado, Arkansas. It has about 8 MW of existing interruptible service from Entergy Arkansas today, and a development target of up to 150+ MW that isn't contracted or energized yet. This is where our conversion work starts. $ZSQR

Interruptible vs. firm power, in plain English: Firm: the utility commits to deliver it around the clock. Interruptible: the utility can curtail it when the grid is stressed. AI workloads generally need firm, redundant power. That's why we call our ~8 MW a starting point.

In 2018, Gartner predicted that by 2025, 75% of enterprise data would be created and processed outside a traditional data center or cloud, up from about 10%. Wherever the exact number landed, the direction holds: more data is created far from any data center.

Edge isn't only about speed. It's also about not moving data you don't need to move. A store with 100 cameras doesn't need to stream every frame to the cloud. Process the video on-site and send one message: "Aisle 14 is out of stock." A few bytes instead of a video stream.

Why does distance matter? Physics. Light in fiber travels about 200 km per millisecond. A round trip to a data center 1,500 km away costs at least 15 ms before any work gets done. Fine for email. A real problem for anything that has to react in real time.

"The edge" isn't one place. It's a range: Device: phone, car, camera On-site: a server closet in a store or plant Network: compute inside a telecom network Regional: smaller data centers near cities Core: hyperscale cloud campuses Closer is faster. Further is bigger.

The easiest way to picture it: The cloud is a central warehouse. Huge, efficient, far away. The edge is the corner store. Smaller, but right where you need it. Most modern systems use both. The corner store handles what's urgent. The warehouse handles the rest.