Stop Arguing the Math
Water, power, and the case for leading with solutions
Steve Williams · babnews.org
There's a pattern in how the AI industry answers its critics, and it's the same on water as it is now on power, and it's the wrong pattern. When a community raises the water question, the industry defends its efficiency ratios. When households watch their electric bills climb, the industry explains that it's really about market design, that data centers caused 63 percent of the increase and not 100, that Texas proves the point. Every one of those claims can be true. I've made the market-design argument myself, because it's correct. And it still loses, because it's an argument, and the public didn't come for an argument. They came to find out what you're going to do about the thing they can see on their bill.
That's the mistake. The industry is fighting a trust problem with a spreadsheet, and you cannot win a trust problem with a spreadsheet. The companies that come out of this era with their social license intact won't be the ones with the most airtight rebuttal. They'll be the ones who visibly added supply, absorbed cost, and said so in plain language. And the useful thing is that this isn't hypothetical. A handful of players are already running that playbook. Most are not, and the gap between them is about to matter a great deal.
Why the math defense loses even when it's right
Being right about the numbers and winning the room are two different games, and the industry keeps confusing them.
When a family's electricity bill rises and a data center explains that the increase is mostly a capacity-auction mechanism rather than the data center's direct fault, every word of that can be accurate and it still reads as evasion. The listener hears a large, well-funded company explaining why the pain they're feeling is technically not the company's doing. That's not reassurance. It's the sound of someone declining responsibility with footnotes. The same is true on water: a defense of power-usage-effectiveness ratios and reclaimed-water percentages answers a question the community didn't quite ask. They asked whether you're taking something they need. A ratio doesn't feel like an answer to that.
The deeper problem is that a defensive posture cedes the initiative. It lets the critic set the terms and reduces the company to responding. And in the vacuum left by companies that only respond, the loudest voice is the angriest one, and the story writes itself as big tech versus the ratepayer. You don't get out of that by winning the argument. You get out of it by changing what you're offering the public to look at.
The power playbook already exists
Here's what makes the defensive crouch so unnecessary: the solutions are real, they're underway, and a few companies are executing them at scale.
The move that most directly answers the ratepayer complaint is bringing your own generation instead of drawing on the shared grid whose scarcity raises everyone's bill. And the industry has committed to it in earnest. Across roughly a dozen announced deals, close to 10 gigawatts of nuclear capacity is now committed to data centers, more capital committed to nuclear in this cycle than in any prior decade of US history. Microsoft locked in the restart of Three Mile Island Unit 1 through a long-term power purchase agreement, which will deliver the first nuclear electrons to an AI data center in 2027, and it chose an existing-reactor restart precisely because it's the fastest path to real megawatts. Amazon invested in X-energy's small modular reactors, Google backed Kairos Power, and Meta committed the largest portfolio, up to several gigawatts across TerraPower, Oklo, and existing operators.
The reason this matters for the household, and the reason it's the message to lead with, is a detail that rarely makes the press release: much of this generation is designed to sit behind the meter, co-located with the data center, bypassing the shared grid and its interconnection queue. Behind-the-meter power is the cleanest possible response to "you're raising my bill," because the data center is adding the supply it consumes rather than competing with households for the existing pool. That's not a defense. It's a solution a ratepayer would actually thank you for, and almost nobody is framing it that way to the public. It sits in energy-transition coverage and investor decks, not in the conversation with the community whose bill is rising.
The honest caveat, which a credible version has to include: these timelines are long. Microsoft's first nuclear power arrives in 2027, and most of the SMR capacity lands in the early-to-mid 2030s. This is direction, not an instant fix. But direction, clearly communicated and genuinely funded, is exactly what earns the patience to keep building. "Here's the supply we're adding and when it arrives" is a fundamentally different posture than "here's why the bill increase isn't our fault."
The water playbook is the same shape
Water shows the identical split between the companies that lead and the companies that defend. The leaders have moved from arguing efficiency to publishing commitments with accounting behind them. Microsoft's water-positive-by-2030 pledge, to replenish more water than it withdraws, sits alongside closed-loop and near-zero-water cooling designs in new builds and, most importantly, public reporting of the specifics through a community-facing infrastructure initiative. The specific behavior that builds trust there isn't the ratio. It's the transparency, showing the numbers and the replenishment projects rather than asking to be trusted on an aggregate.
The pattern across both resources is clear. The leaders add real supply or real replenishment, and they publish it. The laggards stay quiet and let an industry association argue the math on their behalf. And silence, in a debate this charged, is not neutral. It's a decision to let someone else define you.
Consent is the real constraint
Step back and this is the same constraint I keep arriving at from every direction. The binding limit on the AI buildout was never only power, or water, or chips, or capital. Underneath all of them sits consent, the public and political permission to keep building at this pace in these communities. That's the rung that governs all the others, because a buildout the public turns against gets slowed by moratoriums, tax-incentive repeals, and rate-class fights, all of which are already appearing.
And consent behaves differently from the physical constraints. You can't buy it, and you can't win it with a correct argument. It's earned by visibly carrying the cost of your own expansion, adding the power and the water your growth consumes, and communicating that in the language of a neighbor rather than a spreadsheet. The companies treating the water and power backlash as a math problem to be rebutted are quietly spending down the one resource the entire buildout depends on. The companies treating it as a solutions problem to be led on are replenishing it, in both senses.
What leadership actually looks like
So the constructive version, the standard worth holding the industry to, isn't complicated. Lead with the supply you're adding, not the fault you're denying. Put the behind-the-meter generation and the water replenishment in front of the community as the headline, not buried in a sustainability report. Publish the specifics, including the timelines and the parts that aren't solved yet, because credibility comes from the honesty as much as the number. And stop outsourcing the public conversation to a trade association whose job is to win arguments, because winning the argument is not the same as keeping the permission.
The companies that do this will look, in five years, like the responsible builders who brought their own power and stewarded the water, and the public will keep letting them build. The companies that spent those years perfecting their rebuttal will have been technically right and strategically wrong, and they'll find the permission harder and harder to get.
You do not remove a bottleneck. You relocate it. Power and water are real constraints, but the industry is turning the deepest one, consent, into a self-inflicted wound by answering a trust problem with arithmetic. The fix isn't a better spreadsheet. It's leading with what you're adding, out loud, to the people paying the bill. A few already are. The rest are running out of time to notice that the math was never the argument they needed to win.