Onpode
Cover art for Nvidia could solve the energy crisis its own chips created — municipalities are banning data centers

Nvidia could solve the energy crisis its own chips created — municipalities are banning data centers

July 18, 2026 · 9 min

Marcus Vale & Ben Okonkwo

Data centers consumed 1.2 billion gallons of water in Columbus in one year, and in July 2026, U.S. Energy Secretary Chris Wright authorized PJM Interconnection to cut data centers off the grid during a heat wave. Nvidia's closed-loop cooling addresses municipal water use but leaves upstream power-plant water demand — and a projected 240 TWh U.S. increase — untouched.

Across the United States, a growing wave of municipal and state legislation is targeting data center construction in response to concerns about the AI industry's intensive consumption of electricity and water.

0:009:18
Make your own on Onpode

Describe any topic. Hear it in minutes.

More Onpode episodes on Nvidia

About this episode

The obvious story is that municipalities are panicking about AI data centers — banning them in San Marcos, pausing them in New York, drafting moratoriums across 27 states. The less obvious story is what that panic is actually responding to. In July 2026, during a mid-Atlantic heat wave, PJM Interconnection hit 162,700 megawatts of peak demand. The U.S. Energy Secretary signed emergency orders authorizing grid operators to curtail data centers and push them to backup generators. This wasn't a warning. It was a live event. The episode works through what Nvidia's Rubin architecture — with its closed-loop liquid cooling — actually solves, and what it doesn't. It solves the thing that shows up in a public records request: Columbus's 1.2 billion gallon figure from Amazon, Meta, and Google in a single year. It doesn't touch the water consumed at the power plants feeding those facilities, and it doesn't change the load economics that are straining the grid in the first place. There's also the Jevons paradox hanging over the whole conversation: historically, efficiency gains in computing have made total deployment grow faster, not slower. The IEA's 240 TWh U.S. increase by 2030 already accounts for efficiency improvements. The number still doubles. What's most interesting here isn't whether the moratoriums succeed — it's what they're actually negotiating for, and whether Nvidia's efficiency story is the key that unlocks that negotiation or becomes irrelevant the moment the grid runs out of room.

Frequently asked

Why are cities and states banning data centers in 2026?

Cities and states are banning data centers in 2026 because of acute grid and water strain. San Marcos, Texas voted 4-3 to ban them over water consumption. New York Governor Kathy Hochul issued Executive Order No. 62 pausing new hyperscale data center construction. Twenty-seven states are considering similar legislation, driven by grid capacity limits, not just climate politics.

Did the U.S. power grid actually fail because of data centers?

During a July 2026 mid-Atlantic heat wave, PJM Interconnection hit 162,700 megawatts peak demand. U.S. Energy Secretary Chris Wright — a fossil-fuel industry veteran — signed Emergency Orders 202-26-33 and 202-26-32, authorizing PJM to cut data centers off the grid and shift them to backup generators. The grid did not fail, but the circuit breaker tripped in real time.

Does Nvidia's new chip cooling technology solve the data center energy problem?

Nvidia's Rubin architecture uses closed-loop liquid cooling that eliminates direct water use at a facility, addressing what shows up in municipal public records. However, it does not reduce the electricity load — or the water consumed at power plants generating that electricity. The IEA projects a 240 TWh U.S. increase by 2030 even after accounting for efficiency improvements.

What is New York's data center moratorium and what does it require?

New York Governor Kathy Hochul's Executive Order No. 62 imposes a pause on new hyperscale data center construction permits and directs the creation of a regulatory framework and a Community Investment Framework alongside that pause. The order functions as a structured pause with an off-ramp built in, not a permanent ban — distinguishing it from outright prohibitions like San Marcos, Texas enacted.

What is the Jevons paradox and why does it matter for AI data center energy use?

The Jevons paradox describes how efficiency improvements historically increase total resource consumption by making a technology cheaper to deploy at greater scale. Applied to AI data centers, every gain in compute-per-watt tends to accelerate total deployment faster than the efficiency saves energy. The IEA's 240 TWh U.S. projection to 2030 already incorporates efficiency improvements and still reflects a roughly 130% increase.

Grounded in 12 sources
AI Data Centers and the Water Use Feedback Loop · arxiv.org
US heatwave raises alarms over AI data centre energy ... · aljazeera.com
Nvidia says AI's water challenge is largely solved · axios.com
Silicon Valley Has Lost Its Biggest Advantage - The Atlantic · theatlantic.com
Data Center Developers Anxious After Hochul's Construction Pause - Business Insider · businessinsider.com
AI Policy Questions That Congressional Lawmakers Should Know And Be Prepared To Discuss - Forbes · forbes.com
New York Halts Large Data Centers Being Built For A Year - Forbes · forbes.com
Maps Show Data Centers in Areas Where Power Grids Struggling - Newsweek · newsweek.com
Map Shows Where Data Centers Are Being Banned - Newsweek · newsweek.com
New York State halts construction of all new data centers - TechCrunch · techcrunch.com
Amid Growing Data Center Backlash, Nvidia Could Have the Answer | The Motley Fool · fool.com
Nvidia's Cooling Fix Targets the Backlash Stalling 75 AI Projects Worth Billions · ainvest.com
Read transcript

Ben Okonkwo: Marcus, long week — did you catch that Columbus water figure that came out of the public records request?

Marcus Vale: 1.2 billion gallons. One year. One metro.

Ben Okonkwo: Amazon, Meta, Google — about 3% of Columbus public utility water consumption, and that's just what showed up in city records. I keep thinking about the water utility manager who had to read that number for the first time.

Marcus Vale: Which is why Nvidia's Rubin play is so smart — closed-loop liquid cooling, no direct water use at the facility. That's the answer to that number. And frankly, that's why I think Kathy Hochul's Executive Order No. 62 has a shelf life measured in months, not years.

Ben Okonkwo: Hm. So the moratorium dissolves the moment the IEA projects 945 TWh by 2030 and Nvidia ships better cooling — that's the bet you're making?

Marcus Vale: Basically, yes. Twenty-seven states are considering legislation, Sanders and AOC introduced a federal moratorium act — it feels like a wave, but efficiency curves have always outrun political panic. That's the pattern.

Ben Okonkwo: Interesting — because historically that pattern has a name, and it doesn't end the way you're describing.

Marcus Vale: Jevons. Yeah. That's where you're going.

Ben Okonkwo: Right — but Jevons is actually the smaller problem here. The grid already broke. Not 'will break.' Broke. July 2026, mid-Atlantic heat wave, PJM Interconnection hit 162,700 megawatts peak demand. Chris Wright — Energy Secretary, not some EPA bureaucrat — signed Emergency Orders 202-26-33 and 202-26-32. He authorized PJM to cut data centers off. Shift them to backup generators. That's not a warning shot. That's a circuit breaker that already tripped.

Marcus Vale: Hold on. Chris Wright signed that?

Ben Okonkwo: Chris Wright. Not a regulator who ran on climate. The guy who spent his career in fossil fuels. When he's the one pulling that lever, the argument that this is political theater gets — I mean, it gets complicated fast.

Marcus Vale: Okay, but curtailment during a heat wave — isn't that the system working? You shed load, grid stabilizes, data centers come back online.

Ben Okonkwo: That's — okay, think of it like a highway. Efficiency improvements make individual cars faster. But if you add ten times the cars, traffic still stops. Nvidia's Rubin cooling doesn't reduce the number of cars. The IEA's 240 TWh increase for the U.S. alone by 2030 — that's ten times the cars. The highway already jammed once. In July.

Marcus Vale: So the constraint isn't engineering. It's — wait, the constraint is physical capacity of the grid right now.

Ben Okonkwo: Present tense. And the 50 megawatt threshold in those orders — that's not a future regulatory line. That's already the line where a data center becomes curtailable load. Existing facilities crossed it. The order applied to them in real time.

Marcus Vale: So Nvidia's efficiency argument buys headroom, but the 240 TWh projection fills that headroom before Rubin ships at scale. The moratoriums aren't political panic — they're grid operators talking to politicians and saying the math doesn't close.

Ben Okonkwo: And that's actually where Nvidia gets a partial win — a real one. The Columbus figure, the San Marcos city council voting 4-3 to ban data centers over water — that vote was partly about direct water consumption. Closed-loop cooling on the Rubin architecture addresses exactly that. Not hypothetically. The facility stops pulling from the municipal system. That's a real answer to the thing that showed up in the public records request.

Marcus Vale: Right — that's the kernel. San Marcos bans you over water, you show up with Rubin, the water objection evaporates. Literally.

Ben Okonkwo: But — okay, here's the cut. Closed-loop cooling solves direct facility water use. It does not touch the water consumed at the power plant generating the electricity that feeds the facility. Those cooling towers upstream? Still running. The load Rubin enables actually increases that upstream demand. So Nvidia fixed what shows up in a public records request and left untouched what shows up in a power plant's water permit.

Marcus Vale: Solving the visible problem while the load economics stay identical.

Ben Okonkwo: Exactly — and the Jevons piece compounds it. I mean, I want to be careful here because the specific research on Rubin's efficiency translating to total consumption isn't in yet, but historically — every time compute gets cheaper per watt, total compute deployed goes up faster than the efficiency gained. The IEA's 240 TWh U.S. projection already bakes in efficiency improvements. That number still doubles.

Marcus Vale: So Rubin doesn't reduce the fleet. It makes the next doubling feasible.

Ben Okonkwo: That's the historical pattern. Now — whether this specific cycle breaks that pattern, I genuinely don't know. The evidence isn't there yet. But you'd need a mechanism that actually caps deployment, not one that makes deployment more efficient.

Marcus Vale: Which is — wait, that's actually what the moratoriums might be doing, right? Not killing data centers, forcing a negotiation about the cap. And honestly, whether that negotiation accelerates buildout or hits a hard grid stop — that's the thing that makes or breaks Nvidia's entire efficiency story, and we're not there yet.

Ben Okonkwo: Right — and the outcome of that negotiation is what we should probably sit with, because it changes everything downstream. The Rubin story is either the key that unlocks approvals or it becomes irrelevant the moment the grid runs out of room. Those are very different futures.

Marcus Vale: But the fork is actually showing up in the legislation text, not just in theory. Executive Order No. 62 — Hochul's order — it doesn't just pause permits. It directs creation of a regulatory framework and a Community Investment Framework alongside the moratorium. That's not prohibition. That's a hostage negotiation with a term sheet.

Ben Okonkwo: Right — and that distinction actually matters a lot. A pause that produces rules is structurally different from a ban. San Marcos voted 4-3 to ban outright. New York built an off-ramp into the order itself.

Marcus Vale: Virginia goes further — legislators there are looking at shifting major grid upgrade costs directly onto data centers instead of spreading them across all ratepayers. That's the term sheet getting specific.

Ben Okonkwo: Now that's the mechanism I actually find — I mean, cost-shifting changes the math for hyperscalers in a way that a moratorium alone doesn't. A moratorium you wait out. A bill that says you pay for the transmission upgrade? That reprices the site.

Marcus Vale: Wisconsin. What's the Wisconsin number?

Ben Okonkwo: State utility regulator draft report — 40% increase in electricity demand by 2032. And 72% of peak load tied to just three data center developments. Three. That's not a sector risk. That's a counterparty risk. The grid operator has three customers who can break the system.

Marcus Vale: Okay that 72% figure — if I'm Microsoft, I'm suddenly very interested in whether Virginia's cost-allocation bill passes, because that's the price signal that tells me whether Ohio or Virginia is cheaper.

Ben Okonkwo: And here's where Sanders and Ocasio-Cortez matter — not because the AI Data Center Moratorium Act passes, but because California, Ohio, Utah already have comparable legislation. That's not a progressive coalition. That's grid math showing up across the ideological spectrum. When Ohio and AOC land on the same bill, the argument that this is climate politics collapses.

Marcus Vale: So the defensible claim is — moratoriums are the opening bid. The outcome is either faster buildout with accountability rules baked in, or a hard grid stop that makes Nvidia's efficiency story completely moot. And we genuinely don't have evidence yet that Rubin decouples consumption from buildout at scale. That's the honest version.

Ben Okonkwo: The honest version is also the uncomfortable one. A 240 TWh increase that already bakes in efficiency gains — and we're still talking about a 130% jump for the U.S. alone. Nvidia's efficiency is real. The question is whether efficiency that enables more demand is a solution or just a better-looking version of the same problem.

Marcus Vale: And I mean — I started this saying Hochul's order had a shelf life in months. I'm less sure of that now. Actually, no — I'm sure the moratoriums crack open. I'm just no longer sure the thing that walks out the other side looks like progress.

Ben Okonkwo: That water utility manager in Columbus reading 1.2 billion gallons for the first time — they're still reading the same number. The cooling architecture changed. The number didn't.

Marcus Vale: Yeah. The grid is not waiting.

Ben Okonkwo: Appreciate the sparring — this one took a few turns I didn't see coming.

Nvidia could solve the energy crisis its own chips created — municipalities are banning data centers · Onpode