Power, Water, and Fiber: The Physical Trio

Three of the six variables are physical infrastructure the facility literally cannot run without: power to energize it, water to cool it, and fiber to connect it. Treat these as the load-bearing walls of any site evaluation.

Power comes first, always, for the reasons the previous module laid out. But learning the variable means learning its texture, not just its priority. When you assess power at a site you are asking several distinct questions. How much firm capacity can be delivered, and when? What does that power cost, both the energy itself and the transmission and delivery charges layered on top? How reliable is the local grid, measured by outage history and reserve margins? And how clean is it, which matters increasingly because many large buyers have public commitments to carbon-free energy and will weight a site partly on its emissions profile. A site can score well on availability but poorly on price, or well on both but poorly on reliability. Power is not one number; it is a small dashboard, and you read all of it.

Water is the variable beginners most often underestimate. Many large data centers reject heat using evaporative cooling, which consumes water, and a large campus can draw on the order of millions of gallons per day at peak. That draw must come from somewhere: a municipal supply that may already be stretched, groundwater with its own depletion risks, or surface water subject to drought and seasonal variation. Water introduces two kinds of risk. The first is physical scarcity, siting a thirsty facility in an arid or over-allocated basin invites future curtailment. The second is social, because water is emotionally and politically charged in a way electricity often is not. A community may tolerate a data center's power draw but revolt over its water use during a drought, which is where this variable collides with the community variable in Module 2's second lesson. Note also the direct trade-off with power: designs that use less water, such as air cooling or closed-loop systems, generally use more electricity to move and chill air. You can spend power to save water or spend water to save power, and the right balance depends entirely on which is scarcer at the specific site.

Fiber is the third physical requirement, and while it is the fastest and cheapest of the trio to remedy, it is not free to ignore. A data center is only useful if data can move in and out of it at high capacity and low latency. That requires physical fiber routes reaching the site, ideally from more than one direction so a single cut cannot isolate the campus, and ideally from more than one carrier so the operator is not captive to a single provider's pricing. Latency, the time for a signal to travel to major population centers and internet exchange points, matters enormously for some workloads and barely at all for others. A campus serving real-time, interactive applications needs to sit close, in network terms, to the users it serves. A campus running large batch training jobs can tolerate being farther out, which is one reason such workloads are migrating toward cheaper, more remote power even at the cost of higher latency. Fiber is thus both a hard requirement, you must have a route, and a workload-dependent optimization, how much latency you can accept shapes where you can go.

The deeper lesson is how the trio interacts. The most attractive power, often the cheapest and most available, is frequently found in remote regions, far from dense fiber and sometimes in water-stressed terrain. The best fiber and lowest latency cluster in and around major metros, where power is expensive and constrained and land is dear. Water abundance follows geography and climate that may align with neither. There is rarely a site that maximizes all three, so the developer is always making a physical trade: accepting weaker fiber to reach stronger power, or paying a power premium to stay near the network core, or engineering around water scarcity with a costlier cooling design. Faraday's read: the workload decides the trade. A latency-sensitive application forces you toward the metro and its power penalty; a power-hungry, latency-tolerant workload frees you to chase cheap energy in the hinterland. Know the workload before you weight the trio, because the same site is excellent for one and disqualifying for another. This is why there is no universal ranking of sites, only rankings relative to what you intend to run.