The environmental impact of AI systems is real, measurable, and evolving—and our understanding of it is incomplete in ways that matter.

Training large models consumes significant energy. Inference at scale requires ongoing power draw. The hardware depends on rare earth minerals extracted under conditions we often don't see. Data centers require cooling. These are facts, though the precise magnitudes shift as technology changes and as different actors measure (or decline to measure) different things.

We don't pretend to have perfect knowledge of our tools' environmental costs. The research itself is moving, sometimes contradictory, and shaped by who funds it and what they choose to examine. Some opacity is technical complexity; some is institutional indifference; some is active obfuscation by parties with interests in particular narratives.

What we know clearly: nearly everything about modern technological life—the devices we use, the networks that connect us, the climate control in our buildings—carries environmental costs that extend far beyond what's immediately visible. AI is part of that continuum, not separate from it. The question isn't whether our work has environmental impact, but how we choose to act. We act on our current knowledge and follow our curiosity to seek to learn more about what we do not yet know and might not yet be possible.