I’ve been following the infrastructure buildout behind generative AI pretty closely, and honestly, the latest research from the Houston Advanced Research Center has me genuinely concerned. We’re looking at a scenario where the physical backbone of the AI boom could reverse decades of hard-won environmental progress in Texas, and most developers building on top of these systems have no idea what’s happening beneath the hood.
The numbers are staggering. If all proposed data center projects in Texas connect to the grid using the current energy mix, fine particle pollution would more than triple to 2006 levels. NOx emissions would jump 136 percent, hitting pollution levels we saw back in 2004. We’re talking about erasing 20+ years of progress in a five-year window.
What makes this particularly frustrating is that this outcome isn’t inevitable. It’s a choice.
The Energy Problem Nobody Wants to Discuss
Data centers are ravenous for power. A single advanced AI training facility can draw tens of megawatts continuously, and Texas has 26 gigawatts of existing and under-construction capacity. The state is also seeing four times more proposed onsite power generation than any other state, which tells you something about the scale we’re dealing with.
Here’s where it gets interesting for developers: the energy infrastructure decisions being made right now will directly impact whether the platforms you build on can scale sustainably. If data center operators choose gas turbines for speed and certainty, you’re building on increasingly dirty infrastructure. If they pair grid power with renewable energy, the environmental footprint drops dramatically, and frankly, that becomes a competitive advantage.
The problem is economic incentive. Connecting to the grid takes years due to wait lists. Building onsite gas generators takes months. Guess which option most developers are choosing? The efficient path loses to the fast path, and we all suffer the consequences.
What This Means for the Industry
I think we’re at a critical inflection point for AI development. The narrative around generative AI has been purely about capability and cost per inference. What’s been missing is honest conversation about infrastructure externalities. You can’t separate the technology from its physical instantiation, and right now, that instantiation is dirty.
Developers need to start asking their infrastructure providers hard questions. What’s the carbon intensity of the data center hosting your models? Is renewable energy powering the compute or fossil fuels? These aren’t abstract environmental questions anymore, they’re becoming regulatory ones. California’s already moving on data center emissions, and other states will follow.
The renewable energy option exists right now. Solar and wind paired with data centers can reduce NOx emissions to 10 times lower levels. But it requires longer timelines and upfront capital. The race to build AI infrastructure is basically choosing the dirty option because it’s faster.
The Real Cost of Speed
What frustrates me most is that this is a self-imposed constraint. The technology to do this cleanly exists. Solar panels and wind turbines aren’t experimental. The bottleneck is purely about timeline preference and financing models that don’t account for externalized environmental costs.
If you’re building AI products or infrastructure, you should care about this because regulation is coming. Environmental lawsuits are coming. Tax incentives for clean energy are coming. The companies that move first on sustainable data center infrastructure will have a moat that purely capability-focused competitors won’t have.
There’s also the talent angle. I’ve talked to multiple senior engineers who are increasingly uncomfortable building on infrastructure they know is environmentally destructive. As the AI industry matures, this becomes a real competitive factor for hiring. Companies that can honestly say their infrastructure is powered by renewables will have an advantage.
The path forward requires pressure from multiple directions: developers asking questions, regulators implementing standards, investors demanding sustainable infrastructure, and operators investing in the longer timeline solutions. Right now none of these forces are strong enough. The HARC report suggests we might breathe dirtier air than our parents did within five years unless something changes.
That’s not the legacy I want the AI industry to build.