⚠️ Data Center Pollution: Texas Air Quality at Risk in 2026

Why these pollutants matter

Soot (fine particle pollution) and NOx aren't interchangeable. Fine particles can get deep into the lungs. NOx contributes to ground-level ozone, which is a major part of smog. Both matter for public health and local air quality. Carbon dioxide, meanwhile, is mostly about climate change.

That distinction matters because a project can talk heavily about carbon while local air pollution gets less attention. It might also buy clean electricity credits without changing the actual source of power serving the local grid during peak demand. A solid assessment has to separate greenhouse gas emissions, local pollutants, and the real electricity infrastructure running at the time and place where power is used.

Why Texas is a useful warning case

Texas is already a major market for data center development. HARC's report says the state has 26 gigawatts of existing and under-construction data center capacity. The source also cites a Wood Mackenzie report saying Texas is preparing for four times more proposed data center capacity running on on-site power than any other state.

Capacity is a measure of potential power demand, not the amount of electricity consumed every hour. Even so, the scale shows why Texas is a useful test case: a big proposed build-out can create pressure to add generation quickly, and those generation decisions can affect more than just the land where a data center sits.

The HARC analysis isn't proof that every planned project will be built - or that emissions in Texas will match the modeled results exactly. It is a way to see how outcomes can change depending on choices made by developers, utilities, and regulators.

The trade-off between speed and emissions

The business logic is simple: if a data center can't secure power, it can't operate at the scale it plans for. Waiting for grid upgrades can delay both construction and revenue, so a developer might choose generation it can deploy sooner.

But a faster solution can shift costs elsewhere. Gas generation can increase local emissions and make a facility's energy supply harder to change quickly. On-site wind and solar can lower some emissions, but output varies. In practice, a project still needs a plan for times when renewables don't match demand - whether that means grid supply, storage, or other backup options.

That's why blanket claims like "powered by renewables" need scrutiny. The key questions are concrete: What generation is being built? When will it operate? Does it supply the facility during the hours the facility actually uses power? What happens when it can't? And what emissions come from backup or supplemental generation?

These questions apply just as much to AI businesses as to data center operators. If a company's product roadmap assumes more compute, infrastructure choices start to function like part of the operating model. And the cost isn't only the price per kilowatt-hour. It can include connection delays, exposure to energy market changes, and the environmental consequences of the supply plan.

AI demand is part of a broader infrastructure decision

It's easy to frame the issue as an unavoidable side effect of AI growth. But that framing lets the people making infrastructure choices off the hook. Not every proposed data center will necessarily be built, and what gets built depends on forecasts, facility design, workload efficiency, and power procurement decisions.

The practical point isn't that every AI service should minimize compute at any cost. It's that compute demand should be tied to an actual product need. Training and inference workloads can behave differently, and operators can make choices about utilization, cooling, hardware efficiency, and where workloads run. Those decisions don't replace cleaner power - but they can affect how much new generation a project ultimately requires.

Separate capacity announcements from delivered power

Large announced capacity figures can dominate headlines, but they aren't the same as measured electricity use. Capacity is the maximum power a facility (or set of facilities) could require. Actual consumption depends on what gets built, how fully facilities operate, and how demand patterns play out over time.

If you want a more grounded assessment, it helps to keep four things separate:

• Proposed capacity: projects under consideration that may never be built.
• Under-construction capacity: projects further along, but not necessarily operating at full demand.
• Operating load: electricity actually consumed by facilities that are running.
• Emissions: pollution tied to the generation serving that load, including on-site sources where relevant.

Mixing these measures makes it harder to compare forecasts with what actually happens. It can also push businesses and policymakers toward decisions based on headline numbers rather than observed demand and documented supply.

Watch local pollution as well as carbon

Climate reporting tends to focus on carbon dioxide because it's a major driver of global warming and it can be measured across supply chains. But for communities near power plants - or near data center generation - local pollutants should be assessed separately. A lower-carbon claim doesn't automatically answer whether soot or NOx levels are rising nearby.

This isn't an argument against carbon accounting. It's a call to report the metrics that match the question being asked. If the question is climate impact, carbon emissions matter. If the question is local air quality, pollutant emissions, location, and operating patterns matter too.

The same idea applies to AI companies. A sustainability claim is more useful when it spells out what was measured, which electricity sources count, and whether the figures reflect actual operations or contractual arrangements. Without that detail, comparisons can look precise while pointing to very different infrastructure realities.

1. Validate demand before building for forecasts

A proposed facility should be checked against a credible demand case, not just an optimistic market projection. Developers and customers should distinguish contracted workloads in the near term from speculative future demand. This won't remove uncertainty, but it makes the uncertainty visible.

For AI businesses, the parallel question is whether additional compute supports a real use case or just adds infrastructure costs without a clear product payoff. Improving workload efficiency won't remove every power-system constraint, but avoiding unnecessary demand is usually cheaper than trying to supply it with new generation.

2. Make the power plan explicit

A project should spell out how it expects to meet demand: grid electricity, new renewable generation, gas generation, storage, or some mix. It should also be clear about timing. A future renewable project doesn't fix emissions from a facility that starts operating earlier on a fossil-heavy power supply.

The plan should also cover what happens when on-site renewables aren't available, and whether backup systems are expected to run regularly or only during emergencies. The more the plan depends on assumptions, the more those assumptions should be stated plainly.

3. Measure emissions that match the impact

Greenhouse gases and local air pollutants should be tracked separately. Where possible, report actual operating data alongside projections. Be explicit about the calculation boundary: which facilities and generation sources are included, what time period is covered, and whether numbers reflect physical electricity supply or contractual procurement.

This doesn't require every company to publish a detailed engineering study. It does require enough information for customers, regulators, and nearby communities to understand what a claim covers - and what it leaves out.

4. Treat community impact as part of project delivery

Air pollution isn't an abstract externality when it affects where people live. A responsible project should consider nearby communities when evaluating generation choices, operating patterns, and mitigation steps. That's more likely to help if community engagement happens before key decisions are locked in and if concerns can still change the plan.

Regulators and utilities matter here too. They can scrutinize demand assumptions, assess grid impacts, and make sure new load doesn't shift unaccounted costs or pollution onto existing customers and communities.

5. Review performance after launch

A power plan is essentially a forecast until the facility is operating. After launch, demand and emissions should be compared to the original projections. If the project uses more power than expected - or relies on fossil generation more often - the gap shouldn't just fade into an updated sustainability statement.

This feedback loop is what separates a plan that looks responsible on paper from one that stays accountable in practice. It also helps operators understand what assumptions held up, which didn't, and where investments in efficiency or cleaner supply actually made a measurable difference.