← Research indexAndy Masley Substack review
Published June 18, 2026Public Substack posts14 posts reviewed
Last-six-month scrape

Andy Masley is building an anti-panic case about AI and data centers.

I found and scraped public posts from Andy Masley’s Substack at blog.andymasley.com. Since the six-month cutoff of December 18, 2025, the public archive surfaced 14 posts, dated March 11 through June 11, 2026. Thirteen are in “AI & the Environment.”

What I found

The scrape used Substack’s public archive API plus public post pages. It does not include paywalled, deleted, email-only, Notes, comments, or hidden posts.

14public posts found since 2025-12-18
91ktotal words across the scraped posts
13/14posts in “AI & the Environment”

Core pattern

The most engaged posts are long, adversarial myth-busting pieces: land use, infrasound, heat islands, and local data-center panic. Masley’s style is numerate, combative, comparison-heavy, and focused on tracing viral claims back to weak methods or bad framing.

Who is Andy Masley, and what is his skin in the game?

This is the credibility check Clay asked for: role, funding, possible conflicts, and how much weight to give him.

Confirmed self-description

Masley describes himself as an independent writer and researcher focused on AI, the environment, animal welfare, effective altruism, public policy, and related topics. His personal site says he is funded by a grant from Coefficient Giving; his Substack About page says Coefficient Giving was formerly Open Philanthropy.

Background

He says he previously ran the effective altruist Washington, DC city network for four years, taught high school physics for seven years, and double-majored in philosophy and physics at Clark University. His YouTube presence is an IB Physics lecture channel, which supports the physics-teacher part of the bio.

Funding and independence

His About page says he is not employed by Coefficient Giving, that his views do not represent the grantmaker, and that the grant does not obligate him to write from any specific perspective. The stated grant purpose is: “To support one year of writing and communications work made publicly available on topics related to AI, public policy, and other topics.”

Other money

His personal site says he does not receive other funding for his writing besides individual paid Substack subscribers and one-off payments for writing in other publications. Substack metadata also lists paid subscriptions as enabled, with paid support framed as encouragement rather than access to paywalled posts.

Tech-company conflicts?

I found no public evidence in the sources checked that he is employed by, contracted by, or directly funded by OpenAI, Anthropic, Google, Microsoft, Meta, Amazon, a data-center company, or another tech company. That is not proof of no conflict; it is a statement about what was visible in his own disclosures and public profiles checked here.

Real incentives to notice

He does have skin in the game: his audience grew through AI/environment debunking; his grant supports AI/public-policy writing; he writes from a broadly effective-altruist perspective; and he has reputational incentives to be the numerate anti-panic guy. That can sharpen his work, but it can also bias him toward counter-panic and toward underplaying harder-to-quantify local concerns.

Credibility verdict

Treat Masley as a smart, numerate, publicly disclosed independent writer — not as a neutral academic, regulator, or industry insider. His comparative math and source-tracing are useful. His conclusions should still be checked against primary sources, especially because his tone is adversarial and his stated niche rewards finding overblown claims.

Primary sources used for this credibility note: andymasley.com, Substack About page, Bluesky profile, and IB Physics YouTube channel. LinkedIn was listed on his site but was not directly readable from this environment.

Major themes

Across the last six months, Masley is making a coherent case about AI, climate communication, and data-center politics.

1. Personal prompt guilt is the wrong target

He argues individual chatbot use adds very little to a person’s carbon or water budget. His June 11 post plugs a calculator intended to make this visible using third-party estimates and citations.

2. Big numbers need denominators

Gallons, megawatts, acres, nuclear-bomb analogies, and tons of CO₂ are not meaningful without comparisons: per user, per county, per acre, per dollar of tax revenue, per alternative land use, or per unit of avoided harm.

3. Data centers are an infrastructure issue

He does think aggregate AI/data-center electricity demand matters. But he wants attention on clean generation, transmission, interconnection, and specific bad siting rather than prompt-shaming.

4. Local harms are site-specific

He concedes real concerns: CO₂, xAI/Memphis-style local air pollution, audible noise near homes, occasional electricity-bill effects, water-rights complications, and transmission siting.

5. Many viral claims are, in his view, broken

He attacks the bottle-of-water claim, land-temperature claims, infrasound illness claims, nuclear-bomb heat analogies, land-use panic, and contextless water stories as misread, inflated, or badly compared.

6. His weakness is tone and overconfidence

Masley is strongest when showing a claim is unproven or badly framed. He is weaker when “unsupported as commonly stated” becomes “fake,” especially for site-specific harms where evidence can be thin.

What he says about AI water and energy

This is the part most relevant to Clay’s teacher-facing AI-water question.

The bottle-of-water claim

In the May 25 post, Masley argues the “AI uses a bottle of water per prompt” claim is wrong by roughly 50–250×. His post includes an important update: after Shaolei Ren contacted him, Masley says some of his reconstruction was incorrect, but that Ren now agrees GPT-4’s actual water cost was likely much lower than 500 mL — around 15 mL total and about 5 mL onsite for a prompt, with current systems likely lower.

That makes this post especially useful: it does not just dunk on the viral claim; it reports a partial correction after contact with the researcher associated with the original estimate.

Training emissions

In “Training AI models doesn’t emit that much,” he argues training is often framed with misleading individual-scale comparisons. He says it should be compared to manufacturing a mass-market product used by hundreds of millions of people. His rough conclusion: even large frontier training runs become small per user when amortized.

Climate communication

His March 11 post asks for more specific, numerate climate tools. This frames the whole series: stop asking whether an action has any footprint and start asking how big it is relative to other actions and systemic interventions.

Best teacher-facing synthesis from Masley

Normal individual chatbot use is probably not a meaningful part of a teacher’s personal water or carbon footprint. The serious issue is aggregate infrastructure: how much new electricity demand AI creates, what generation serves it, whether local datacenters are well-sited, and whether communities manage water, noise, tax revenue, and pollution honestly.

What he says about local data-center impacts

Masley’s data-center posts are mostly a campaign against what he sees as low-quality panic.

Status quo bias

In “A simple trick to fix the data center debate,” he asks whether a county would spend its data-center tax revenue to avoid the externalities. In Loudoun County, he compares roughly $1.3B/year in tax revenue against mitigation costs for water, land, NOx, electricity bills, and emissions.

Land use

He argues land-use objections are mostly fake at national scale: data centers occupy tiny acreage compared with ethanol corn, animal agriculture, food waste, idle farmland, and ordinary exurban development, while generating high revenue per acre.

Heat exhaust

He attacks claims that data-center waste heat is raising surrounding land temperature by degrees. His critique: satellite land-surface temperature mostly detects roofs, pavement, and land-cover change, not server exhaust warming the air.

Infrasound and noise

He sharply separates audible noise, which can be a real nuisance and sleep issue, from sub-audible infrasound illness claims, which he says are not supported by wind-turbine/infrasound research.

Air pollution

His air pollution post concedes real risks from onsite fossil generation and backup generators, especially local NO₂/PM exposure near vulnerable communities. But he is skeptical that datacenters are a national air-pollution catastrophe relative to other sources.

Water

He is skeptical of vague “millions of gallons” stories unless they specify local water stress, share of supply, withdrawal vs consumption, normal operation vs construction damage, and what alternatives or offsets would cost.

Public posts found in the six-month window

Dates, URLs, sections, word counts, and one-line synopses from the scraped archive.

DatePostSectionWordsSynopsis
2026-06-11A plug for a tool I made...AI & the Environment505Introduces a calculator showing ordinary chatbot use adds little to personal emissions or water use.
2026-06-06A simple trick to fix the data center debateAI & the Environment2,486Reframes the debate around whether communities would forgo tax revenue to avoid quantified externalities.
2026-06-04Why I think panic about local impacts...AI & the Environment1,863Surveys local-impact claims and argues most collapse under scrutiny, while asking for counterexamples.
2026-05-25I think I figured out exactly how...AI & the Environment3,541Investigates the “bottle of water per prompt” statistic and says corrected numbers are far lower.
2026-05-20A history of the data center panic - part 1AI & the Environment5,272Traces AI/data-center panic through earlier inflated training-emissions discourse and media incentives.
2026-05-19A crash course on US air pollutionAI & the Environment13,373Explains U.S. air pollution and contextualizes data-center generator and health-damage claims.
2026-05-11Let’s not compare data center heat exhaust...AI & the Environment2,229Rejects nuclear-bomb heat analogies because dispersed continuous heat is not explosive heat.
2026-05-02Data center land use issues are fakeAI & the Environment9,840Argues land-use objections are overstated compared with agriculture, ethanol, and other land uses.
2026-04-24To be clear, I do understand how sound worksAI & the Environment4,874Responds to criticism of his infrasound argument and clarifies audibility vs clinical significance.
2026-04-19Contra Benn Jordan...AI & the Environment25,322Long rebuttal arguing sub-audible data-center infrasound illness claims are unsupported.
2026-04-12Training AI models doesn’t emit that muchAI & the Environment3,729Argues training emissions look less exceptional when compared to mass-market product manufacturing.
2026-04-03What’s blocking Waymo in DC...Artificial Intelligence8,508Maps regulatory obstacles and possible fixes for commercial Waymo service in Washington, DC.
2026-03-31Data centers’ heat exhaust is not raising...AI & the Environment8,362Critiques a “data heat island” paper as likely measuring roofs/pavement rather than waste heat.
2026-03-11A call for more specific and numerate climate communicationAI & the Environment1,266Calls for better tools that compare lifestyle emissions with larger systemic interventions.

Caveats and how I’d use this

Masley is useful, but not neutral in tone. Treat him as a strong anti-panic analyst, not as the final word.

Coverage caveats

  • The public archive surfaced posts only from March 11 to June 11, 2026 inside the six-month window.
  • This excludes paywalled, hidden, email-only, deleted, Notes, comments, and future revisions.
  • Some very long posts were summarized for claims and structure, not every subclaim.

Interpretive caveats

  • His rhetoric sometimes turns “not established” into “fake.”
  • National comparisons can underweight site-specific household harms.
  • Tax-revenue offset logic depends on governments actually mitigating harms.
  • Distribution matters: benefits and burdens may fall on different people.

Best use for Clay

Use Masley as a source for disciplined skepticism: demand mechanisms, denominators, comparisons, and local context. He strengthens the case that teachers should not feel guilty about ordinary AI prompting, while preserving the more serious question of how AI infrastructure should be powered, sited, disclosed, and regulated.