Two claims anchor most public debate about new data centers: they use “too much electricity” and “too much water.” Both claims contain real signal. But several of the specific numbers driving the panic don’t survive basic arithmetic – and that matters, because bad math makes it easy to dismiss a governance problem that is actually real.
The question worth asking isn’t “do data centers consume resources.” They do, at scale. The question is whether the specific figures being used to justify or oppose a project are accurate, and whether the policy built on top of them fairly assigns cost and risk.
Claim 1: “A 1 GW data center needs 6 GW, including cooling and everything else”
This claim misunderstands how facility overhead is measured. Power Usage Effectiveness (PUE) compares total facility power to IT equipment power:
- PUE = Total facility power / IT equipment power
- Google reports a 2025 fleet-wide average PUE of 1.09 – meaning 1.0 GW of IT load draws about 1.09 GW at the facility level.
- The Uptime Institute’s 2025 Global Data Center Survey put the industry average PUE at 1.54.
The math: even at the industry-average PUE of 1.54, a 1 GW IT campus draws roughly 1.54 GW total – not 6 GW. A “6x” multiplier isn’t a credible planning assumption for facility overhead under any commonly cited efficiency benchmark.
None of this makes the load small. A 1.1-1.5 GW continuous draw is still a massive, single-point commitment for a local grid – comparable to a mid-size power plant’s output, sited in one place, running near-continuously. The exaggeration is in the multiplier, not in the underlying concern about grid strain.
Claim 2: Water headlines (“20-70 million liters a day”)
Water Usage Effectiveness (WUE) measures liters of water per kWh of IT energy. Equinix’s own published range spans from near 0 (air-cooled) to about 2.5 L/kWh (evaporative cooling in hot, dry climates); the Environmental and Energy Study Institute cites an industry-average WUE of roughly 1.8-1.9 L/kWh.
Worked example, 1 GW IT load, running continuously:
- Daily IT energy = 1 GW × 24 h = 24,000,000 kWh
- At WUE = 0 (air-cooled): 0 liters/day of on-site cooling water
- At WUE = 1.85 (industry average): ~44.4 million liters/day
- At WUE = 2.5 (top of Equinix’s evaporative range): 60 million liters/day
So a figure like “20-70 million liters/day” is achievable, but only for evaporative or hybrid-cooled campuses – it is not a universal number, and it says nothing about whether that water is potable, reclaimed, or drawn from a stressed basin, which is the detail that actually determines local impact. Direct-to-chip and closed-loop liquid cooling push the real number toward the low end of this range, not toward zero – “reduced” isn’t “irrelevant” in a water-stressed region.
Order of Magnitude: “How much water do the Datacenters really consume?”
Lets drill down to numbers as an order of magnitude.
FAO reference data puts sugarcane’s seasonal water requirement (ETm) at 1,500-2,500 mm, depending on climate. Run the same kind of worked calculation:
- 100 acres = 40.5 hectares = 405,000 m²
- Season volume at 1,500 mm: 405,000 m² × 1.5 m = 607,500 m³ = 607.5 million liters
- Season volume at 2,500 mm: 405,000 m² × 2.5 m = 1,012,500 m³ = 1.01 billion liters
- Annualized to a daily rate: roughly 1.7-2.8 million liters/day
Compare that to the 44-60 million liters/day worked out above for a 1 GW evaporative-cooled campus. Matching a large campus’s water draw with sugarcane would take roughly 1,500 to 3,500 acres. To put in perspective, area under sugarcan plantation is in magnitude of 15 million acres. This is some point to ponder. Can we give away 3000 acres (or 0.02%) of sugarcane plantation to accomodate a Datacenter water needs?Does not seem too much water guzzler to me. What do you think?
What this means: the exaggeration is real, and so is the underlying issue
Correcting the arithmetic doesn’t make the concern disappear – it relocates it. A 1.1-1.5 GW continuous grid draw and a multi-million-liter daily water demand are genuinely large infrastructure commitments, even using the honest numbers. The distortion in public debate isn’t “data centers use resources,” it’s specific multipliers and comparisons that inflate or misdirect the real figures. Once the numbers are right, the actual fight is over governance: who pays for the grid and water infrastructure a campus requires, and who bears the risk if projections don’t hold.
The policy question that actually matters
Before approving a large campus, a region should be able to answer:
- Are transmission upgrades paid by the operator, or socialized into household rates?
- Are tax breaks, discounted tariffs, and land grants tied to measurable local benefits?
- Is water priced transparently, with scarcity and drought-stress signals built in?
- Are reclaimed and non-potable sources contractually prioritized over potable supply?
- Do approval terms require peak-load management and demand-response participation?
- Does site selection actively avoid already-stressed water basins and fragile grid feeders?
A framework for fair expansion
- Require public disclosure of PUE, WUE, and hourly load profiles at the campus level – not fleet-wide averages that hide the specific site’s impact.
- Tie incentives to local job quality, grid support services, and measurable water stewardship, not just capital investment totals.
- Phase capacity additions to completed infrastructure upgrades, rather than approving full load ahead of grid readiness.
- Use location-sensitive approvals that reward low-stress sites and restrict high-stress ones.
- Mandate independent annual audits of energy, water, and community commitments, with real penalties for missed terms.
Bottom line
Resource anxiety about data centers is reasonable, but several of the numbers driving that anxiety don’t hold up: the “6x power” multiplier, water figures quoted without cooling method or climate, and agricultural comparisons that are off by an order of magnitude. Get the arithmetic right, and the debate moves to where it belongs – not “is this scary,” but “is the cost fairly assigned.” That’s a policy design problem, and it’s solvable with disclosure, phased approvals, and audits, not with better slogans.
Data notes and sources used
- IEA, Electricity 2024: global data center electricity was about 460 TWh in 2022, projected to exceed 1,000 TWh by 2026. (iea.org/reports/electricity-2024)
- Google Data Centers, Power usage effectiveness: 2025 fleet-wide average PUE of 1.09. (datacenters.google/efficiency)
- Uptime Institute, 2025 Global Data Center Survey: industry-average PUE of 1.54 (n=681), down from 1.56 in 2024.
- Equinix, What Is Water Usage Effectiveness (WUE)?: WUE ranges from near 0 to about 2.5 L/kWh depending on cooling design and climate. (blog.equinix.com)
- Environmental and Energy Study Institute (EESI): industry-average WUE cited at roughly 1.9 L/kWh.
- FAO Land & Water, Sugarcane: seasonal water requirement (ETm) of 1,500-2,500 mm depending on climate. (fao.org/land-water)
If your organization is planning large digital infrastructure, we can help evaluate architecture, energy-water trade-offs, and rollout governance: https://kotibyte.com/services/
Discuss your use case with our team: https://kotibyte.com/contact/
Leave a Reply