This one’s for Trevor.

You know Trevor, maybe not by that name, but you have seen them on LinkedIn posting three times a day about AI ethics, finger-wagging at anyone who touches a generative tool, performing outrage for an audience of people who are also performing outrage.

They have a lot of feelings about what AI is doing to creatives and artists and the humans who deserve to be paid for their work, but not a single concrete proposal for actually keeping any of those people employed, paid, or meaningfully in the loop on anything.

I run Antidote Games, and our entire mission is promoting human creativity and making it financially viable, by investing in people’s projects and proving that it is feasible to pay real humans to do the work. If someone comes to us with a project full of AI-generated art they used for prototyping, we will not help them publish it. What we will do is help them hire someone to finish it, polish it, and make it human-made. That is the whole model, investing in human creative output first, because we believe it is both the right thing to do and a viable business. And I am not just saying it, I am writing the checks.

I also build with AI every single day in my professional work, which means I have actual technical authority on this subject, not theoretical concern. My response to that understanding was not to panic about it on LinkedIn. It was to build a business that puts money into human hands and to develop the E² methodology, a framework for responsible AI implementation that keeps humans genuinely engaged in the process rather than replaced by it, using the tools that exist right now without over-committing to a fully automated future nobody has actually built yet.

So when someone like Trevor performs AI outrage on LinkedIn without doing a single hour of the actual work of keeping human talent employed, it is not an abstract frustration for me.

And Trevor’s rants are not hurting the AI industry, they are hurting the people Trevor claims to care about, because when a sociopathic CEO watches someone like Trevor, entitled, loud, factually sloppy, professionally unproductive, the thought that crosses their mind is not that they should take this seriously, it is that they do not want to pay that person anymore and maybe they should just replace them.

You do not get to claim the moral high ground on AI ethics while doing none of the actual labor of building alternatives or supporting artists or engaging seriously with how these systems work, and every bad argument that gets amplified makes the serious conversation harder to have and easier for the people with the most to gain from ignoring it to dismiss entirely.

There is a version of this conversation that is worth having, one with hard questions about labor displacement, decision accountability, access inequality, and concentration of power, and those are real problems that deserve serious people engaging with them seriously.

Instead, a significant portion of public discourse has been captured by claims that do not survive five minutes of contact with actual data, and every person loudly championing those bad arguments is burning credibility that could have fueled real accountability.

This post is about two of the most common offenders, the claim that AI is poisoning our water and the claim that AI is destroying the power grid.

”AI Is Ruining Our Drinking Water”

The logic sounds airtight on the surface: data centers use water, AI runs on data centers, therefore AI is ruining water. But “uses water” and “ruins water” are not the same thing, and the gap between them is exactly where this argument falls apart.

Large data centers do use significant amounts of water, primarily through evaporative cooling where water absorbs heat and gets released as vapor, drawing from municipal supplies, surface water, and in some cases groundwater directly.

The Lawrence Berkeley National Laboratory estimated that U.S. data centers consumed about 17 billion gallons of water directly through cooling in 2023, with an additional 211 billion gallons consumed indirectly through the electricity that powers them, much of which comes from steam-generating fossil fuel plants. Those numbers are real, and in specific communities they translate into genuine local pressure on water infrastructure that is worth taking seriously and regulating.

What those numbers do not translate into is contaminated drinking water, which is the claim that actually gets made. Groundwater contamination, the ruination of water tables that Trevor gestures at, comes from industrial chemical discharge, agricultural runoff, leaking fuel storage tanks, and inadequate sewage treatment, not from server cooling systems.

The water that data centers consume through evaporation is consumed, not poisoned, and those are meaningfully different problems requiring meaningfully different responses.

Scale matters here too. The Food and Agriculture Organization of the United Nations puts agriculture at roughly 70% of all global freshwater withdrawals, a figure consistent across FAO’s AQUASTAT database, the UN World Water Development Report, and the World Bank. All data centers combined sit at a fraction of that total.

The concern worth having is local, specific, and real: in water-stressed regions and communities where a single hyperscale facility can represent a significant share of total local water demand, the siting and regulatory oversight of data centers deserves serious attention. A large hyperscale data center can consume up to 5 million gallons per day, which is comparable to the water use of a town of 10,000 to 50,000 people, so the local impact question is legitimate and worth fighting for.

That is a tractable, enforceable conversation. “AI is ruining our drinking water” is not that conversation, it is a horror movie tagline wearing the costume of environmentalism, and it is making it harder to have the conversation that would actually protect anyone.

”AI Is Destroying the Power Grid”

This one has a stronger foundation, which is exactly what makes the exaggeration more frustrating, because there are real numbers here that get distorted into something they do not actually say.

According to the IEA’s 2025 Energy and AI report, data centers consumed approximately 415 terawatt-hours of electricity globally in 2024, about 1.5% of total global consumption, and the IEA projects that figure roughly doubling to around 945 TWh by 2030 in their base case scenario. AI is the primary driver of that growth, with accelerated servers driven by AI adoption projected to grow at roughly 30% annually. These are real trends with real implications for grid planning, and the IEA is not a fringe source.

What they do not imply is that AI is melting the power grid. The IEA’s own report notes that data center demand growth accounts for less than 10% of total global electricity demand growth between now and 2030, with electric vehicles, industrial output, and air conditioning leading the way.

The entire global data center footprint, running every streaming service, bank, hospital system, cloud storage provider, and AI workload on the planet, is a real and growing number, but it exists in a context that Trevor never provides.

The thing that gets lost most consistently in this conversation is that the 1.5% figure covers all data centers, and AI is a subset of it, so when someone says “AI is destroying the power grid” they are almost always citing a number that includes your Netflix queue, your bank’s transaction processing, every Zoom call, every cloud backup, and the servers running every SaaS product you use at work. Attributing that entire footprint to AI is not an honest framing.

On the supply side, the IEA projects that renewables will meet roughly half of global data center demand growth to 2035, with over 450 TWh of new renewables generation built specifically to serve that demand. Tech companies are among the largest corporate purchasers of renewable energy on the planet, and that is a structural feature of how they are contracting for their infrastructure, not a press release.

The legitimate concern, and it is real, is that in regions still heavily dependent on natural gas or coal, rapid data center expansion may delay the retirement of carbon-intensive generation and lock in near-term emissions. That is a specific, regional, time-bounded problem worth specific policy responses, not evidence that AI is uniquely destroying civilization’s ability to keep the lights on.

Why This Matters Beyond Being Annoying

The people making these bad arguments are not usually acting in bad faith. Many of them genuinely care about what AI is doing to creative work, to labor markets, to the environment, and to the distribution of power in society, and those concerns are legitimate.

The way the concerns get expressed is where the problem is, because when the AI ethics conversation fills up with claims that collapse under basic scrutiny, the people who might have engaged seriously disengage because the discourse seems unserious, and the people who benefit from avoiding accountability get to point at bad arguments as a reason to dismiss all arguments.

Real accountability for AI’s environmental footprint requires accurate baseline measurements, transparent corporate reporting, regional regulatory frameworks that account for water stress and grid composition, and long-term efficiency standards. None of that moves faster because someone tweeted that AI is poisoning the water table.

The conversation worth having is about who controls these systems, who benefits, who bears the costs, and how you maintain decision-making accountability when the systems making consequential decisions are opaque, and those conversations require credibility, which requires accuracy, which requires doing the work before making the claim.

A Note on How This Post Was Researched

Trevor is going to show up in the comments with a counter-source. That is what Trevor does. So here are every source used to build this argument, and here is the part that should be instructive to anyone paying attention.

Compiling this kind of research the traditional way, tracking down primary sources across the IEA, the FAO, the UN, Lawrence Berkeley National Laboratory, and the World Bank, reading them, cross-referencing the claims, and verifying that nothing was being taken out of context, would realistically take weeks of focused time without institutional access or a research budget.

What AI actually did here was help identify what needed to be read, surface the right primary sources quickly, and flag where claims needed to be tightened or dropped entirely because the sourcing did not hold up. The reading, the judgment calls, the understanding of what the numbers actually mean in context, that part was not delegated. The friction between wanting to be informed and actually being informed got removed. That is a meaningful difference.

That is the reality Trevor keeps missing. AI is not the enemy of informed, educated people doing serious work. For a lot of us it is the thing that finally makes serious work accessible without a research budget or an institutional library login.

The people screaming about AI on LinkedIn are screaming on a platform that runs on servers, optimized by machine learning, serving ads targeted by algorithms, while using a smartphone assembled by automation. The outrage is fine. The selective application of it is not.

If you want to have the real conversation about AI, start with accurate information. This is where ours came from.

Sources

Power Consumption

  • International Energy Agency, Energy and AI (April 2025) — primary source for the 415 TWh / 1.5% global consumption figure, the 945 TWh 2030 projection, and the renewables procurement data. Full report at iea.org. Reported by S&P Global Commodity Insights, April 10, 2025.

  • Pew Research Center, What We Know About Energy Use at U.S. Data Centers Amid the AI Boom (October 2025) — cross-referenced the IEA figures for U.S.-specific data center consumption at 4% of total U.S. electricity in 2024.

Water Consumption

  • Lawrence Berkeley National Laboratory, 2024 United States Data Center Energy Usage Report (December 2024) — source for the 17 billion gallons direct and 211 billion gallons indirect U.S. water consumption figures for 2023. Published by the U.S. Department of Energy. eta-publications.lbl.gov

  • Environmental and Energy Study Institute (EESI), Data Centers and Water Consumption (June 2025) — source for cooling mechanism details, the 5 million gallons per day hyperscale figure, and the indirect water footprint from fossil fuel power generation. eesi.org

Agriculture and Global Water Use

  • Food and Agriculture Organization of the United Nations (FAO), AQUASTAT database and One Health Water overview — source for the 70% agriculture freshwater withdrawal figure. Consistent with the UN World Water Development Report (UNESCO, 2024) and World Bank open data.