AI’s Gas Demand Is Rising — So Is the Case for Hardware Circularity

Data Centers Are Guzzling Up Gas in the US (Bloomberg Green, July 26). And that’s only half the story.

The chart making the rounds this week shows US gas consumption from data centers roughly doubling every few years through 2035, climbing past 6 billion cubic feet a day as every region of the US builds out to feed AI compute.

The headline writes itself: AI investment = higher near-term emissions.

That’s the first-order effect. The second-order effect, that few are tracking is playing out in the IT Asset Disposition market. And for those who understand how impactful second order effects are – it’s far more interesting.

The second order effects:

1️⃣ Hyperscalers and enterprises are racing to procure the latest GPUs and accelerated compute. That forces faster refresh cycles for everything around them (e.g., servers, storage, networking gear)

2️⃣ Faster refresh cycles push more legacy hardware onto the secondary market sooner, and in greater volume, than historical replacement schedules would predict.

3️⃣ That’s fueling a genuinely thriving secondary market for high-quality IT assets: with a clear preference emerging for refurbishment and reuse over recycling for parts, since well-maintained enterprise hardware still has years of useful life left.

4️⃣ At the same time, rising IT hardware demand, met by a rising supply of quality assets, is pushing up secondary market prices.

5️⃣ Higher prices for used enterprise hardware are now feeding back into corporate procurement decisions. Under fixed IT budgets, some organizations are slowing their own refresh cycles rather than paying for higher priced IT, or purchasing premium refurbished kit.

6️⃣ The result: IT assets stay in productive use for longer across the broader economy, reducing the pressure to manufacture more new IT in the first place.

The full picture: AI infrastructure buildout drives is certainly driving up near-term operational emissions, but also pushing up hardware prices – leading companies to extend the useful life of their current hardware, and improving circularity economics across the much larger installed base of enterprise IT that isn’t running frontier AI workloads at all.

Higher emissions in one part of the system. Lower embodied-carbon pressure in another. Today, both true at once.

The AI/emissions story isn’t wrong. It’s just incomplete. The full lifecycle view, including what happens to the hardware AI is displacing, is where the real ESG picture, and the real opportunity, sits.

THE UNTOLD HALF OF THE AI EMISSIONS STORY

Higher near-term emissions from AI buildout, offset by a growing case for reuse across the wider IT economy