Across the world, the race to build AI-ready data centres is being sold as progress: more compute, faster innovation, and the promise that new models will translate into productivity gains, medical breakthroughs, and smarter services. But behind the polished renderings and press releases, a different story is increasingly surfacing—one about energy strain, water stress, cooling emissions, and the uneven way environmental impacts are being measured, disclosed, and managed.
In recent months, scrutiny has intensified as regulators, local communities, and researchers ask a blunt question: when tech companies announce expansions, where is the full accounting of what those expansions do to the places they land? The concern is not simply that data centres consume resources—everyone in the industry acknowledges that. The concern is that the public narrative often stops at high-level commitments, while the granular, ongoing details that would allow outsiders to understand trade-offs are missing or delayed.
What’s emerging is a pattern: companies can be quick to describe capacity targets and power procurement plans, yet slower to provide consistent, comparable information about how electricity is sourced over time, how much water is actually used under real operating conditions, how cooling systems affect local air quality, and what monitoring exists for ecosystem impacts. The result is a growing sense of a “cover-up,” not necessarily in the sense of deliberate deception, but in the sense of incomplete transparency—an information gap that makes it hard for communities to evaluate whether promises match reality.
Energy: the invisible bottleneck that becomes a climate question
The first pressure point is energy. Data centres are often described as “power-hungry,” but the more consequential issue is what kind of power they draw. A facility’s climate footprint depends heavily on whether its electricity comes from renewables, low-carbon generation, or fossil-heavy grid mixes—and on whether the company’s procurement strategy actually changes the grid’s emissions profile or merely shifts costs.
In many regions, the grid is already stretched. When large loads arrive quickly, utilities may respond by adding generation, increasing imports, or running existing plants harder. Even if a company signs a contract for renewable energy, the timing and structure of that contract matters. Some arrangements are long-term and backed by new renewable builds; others rely on existing generation or financial instruments that do not guarantee additional clean power at the moment the data centre is consuming electricity.
Local officials and environmental groups have repeatedly asked for clarity on these distinctions. They want to know not just what percentage of power is “renewable,” but how that figure is calculated, whether it reflects hourly matching, and what happens during periods when renewables are scarce. They also want to understand whether the company’s growth plan assumes grid upgrades that may take years, and what interim emissions look like.
There is also a second layer: efficiency claims. Modern data centres can be highly optimized, and industry metrics such as PUE (power usage effectiveness) are meant to capture how efficiently energy is converted into computing rather than wasted as heat. Yet PUE is not a complete story. Two facilities with similar PUE can have very different emissions profiles depending on their energy mix, their load patterns, and their cooling approach. Moreover, PUE can be reported in ways that obscure operational variability. A facility might perform well under certain conditions and less well under peak demand or during unusual weather.
As AI workloads scale, the load profile changes too. Training runs can be bursty; inference can be steady but still grows rapidly. That means the environmental impact is not static. It evolves with utilization rates, scheduling strategies, and the degree to which companies can shift workloads to times when cleaner power is available. Without transparent reporting over time, outsiders cannot tell whether improvements are happening or whether the facility’s footprint is simply expanding alongside its compute.
Water: the resource that turns infrastructure into local politics
If energy is the climate question, water is the local one. Data centres require water for cooling, either directly through evaporative systems or indirectly through power generation and other industrial processes. The public debate often focuses on headline numbers—how much water a facility uses—but the deeper issue is location-specific impact.
Water availability varies dramatically by region. A facility that draws from a plentiful supply in one area may be unacceptable in a drought-prone basin elsewhere. Even when total withdrawals appear modest relative to regional consumption, the timing and source matter. Withdrawals during dry seasons can have outsized ecological effects. Water drawn from rivers can affect temperature and flow, which in turn influences aquatic life. Water drawn from groundwater can contribute to long-term depletion if recharge is limited.
Cooling technology complicates the picture. Some data centres use air cooling, others use evaporative cooling, and many use hybrid systems. Evaporative cooling can reduce electricity use compared with some alternatives, but it increases water consumption. Air cooling can reduce water use but may increase energy demand, especially in hot climates. Companies may emphasize whichever metric is most favorable, while critics argue that the trade-off should be evaluated holistically.
Another challenge is that water reporting is often fragmented. Environmental permits may require monitoring, but the frequency, methodology, and public accessibility of that monitoring can vary. Communities may receive updates only when something goes wrong or when permits are renewed. Researchers may struggle to obtain datasets that would allow them to compare predicted impacts with observed outcomes.
This is where the “cover-up” narrative gains traction. Not because every company hides information, but because the information that would allow independent verification is not always provided in a consistent, accessible way. When residents ask for baseline measurements—what the water system looked like before the facility expanded—they may find that the baseline is either unavailable or not comparable across sites.
Cooling and emissions: the heat problem that doesn’t stay inside the fence
Data centres generate heat, and managing that heat is not just an engineering task—it can become an environmental issue. Cooling systems can emit pollutants indirectly through electricity generation, and directly through exhaust streams, chemical treatments, or cooling tower drift. In some cases, facilities may also affect local microclimates by altering airflow patterns and increasing ambient temperatures around the site.
The industry has made progress in reducing emissions and improving cooling efficiency, but the public conversation often lags behind technical nuance. For example, companies may claim that their cooling systems are “closed loop” or “low emission,” yet the actual performance depends on operating conditions, maintenance practices, and the chemistry used in water treatment. Drift control measures can reduce particulate emissions, but they require monitoring and enforcement.
There is also the question of how heat is handled at scale. When multiple data centres cluster in a region, the cumulative thermal impact can become significant. Even if each facility meets its permit limits, the combined effect may change local conditions. Regulators and scientists have begun to ask whether permitting frameworks account for cumulative impacts, or whether they treat each facility as an isolated unit.
Cooling emissions management also intersects with energy sourcing. If a facility relies on more electricity to cool efficiently, its emissions depend on the grid. If it uses more water to reduce electricity demand, its emissions depend on water sourcing and treatment. The environmental footprint is therefore a system-level outcome, not a single metric.
Public reporting and accountability: the missing longitudinal view
Perhaps the most persistent complaint is about time. Many disclosures are made at the moment of approval or construction: environmental impact assessments, projected resource use, and commitments to mitigation measures. But the public rarely sees a clear, ongoing dashboard that tracks actual performance against those projections.
That longitudinal transparency is what communities need to trust the process. If a company says it will use a certain amount of water, then the public should be able to see monthly or seasonal withdrawal data, along with comparisons to baseline conditions and to permit limits. If a company claims renewable energy procurement, the public should be able to see how that procurement translates into actual emissions reductions over time, including during peak demand periods.
Accountability also includes what happens when things deviate. If a facility’s water use rises due to operational changes, does the company report it promptly? If cooling performance worsens during heatwaves, does it disclose the reasons and the mitigation steps? If emissions exceed expectations, are there corrective actions and public updates?
Without this, the debate becomes a contest of narratives. Companies point to compliance and engineering sophistication. Critics point to gaps in disclosure and the difficulty of verifying claims. Regulators may have access to internal data, but the public often does not. That asymmetry fuels suspicion.
A unique angle: the “infrastructure optimism” problem
One reason the transparency gap persists is that the incentives for companies and the incentives for communities do not align neatly.
For tech firms, the primary goal is speed: securing power, breaking ground, and meeting capacity timelines. Environmental reporting, by contrast, is slow and iterative. It requires baseline studies, monitoring programs, data validation, and sometimes third-party verification. Those steps can delay projects or complicate messaging. Even when companies intend to be responsible, the business pressure to deliver can lead to a communications strategy that emphasizes what is planned rather than what is proven.
Communities, meanwhile, experience impacts locally and immediately. They may notice increased traffic, construction noise, strain on water systems, and changes in local air quality. They want answers now, not after the next permit cycle. When they don’t get them, they interpret silence as concealment.
There is also a structural issue: data centres are often built through complex corporate arrangements. Ownership may shift. Operators may subcontract cooling and energy management. Power procurement may involve utilities and intermediaries. That complexity can make it harder to produce a single, coherent public account of environmental performance.
But complexity is not an excuse. If the public is expected to accept the environmental trade-offs of AI infrastructure, then the public deserves a clear, consistent reporting framework that can survive corporate restructuring and operational changes.
What “coming clean” could realistically look like
Calls for greater transparency can sound vague—“be more honest,” “publish more data.” But there are concrete steps that would meaningfully change the debate.
First, companies should publish standardized, site-level environmental dashboards that include energy sourcing details, water withdrawals, cooling approach metrics, and emissions estimates. These should be updated regularly, not just at