20 Million Chips Today, 200 Million by 2028: What the Times' AI Numbers Mean for Your County

The New York Times' July 29 feature on the AI build-out opens with a hard problem: how do you describe a construction boom this size in numbers a reader can actually hold in their head? Its answer — railroads in the 1800s, the New Deal, the Manhattan Project — is the kind of comparison that sounds like hyperbole until you look at the chip count.

The numbers worth writing down

Every one of those numbers is a proxy for the same underlying fact: the industry believes bigger models need more chips, more chips need more power, and whoever builds the most capacity first wins the race. Amazon's Peter DeSantis, who leads the company's foundational AI models, put it plainly to the Times: "It's hard to get your mind around the scale."

Why "doubling every nine months" matters more than any single project

Most community fights are fought project by project — one rezoning application, one interconnection request, one hearing at a time. That's the right way to fight the project in front of you. But the compute numbers are the reason there will be another one behind it, and another behind that. If the chip count really doubles every nine months through 2028, the pipeline of proposals hitting county planning boards over the next two years won't be a continuation of the last two — it will be several times larger. A county that has fielded one data-center proposal so far should plan for that not being the last.

The backlash the Times itself flagged

Buried in the middle of the piece is a sentence worth pulling out on its own: data centers are shaping up as a major issue in November's midterm elections, part of a "growing national movement" pushing back on the industry — a trend the Times documented in its own April piece. That's a national newsroom confirming what this site's story tracker and moratorium tracker show every week in the headlines: this isn't a scattering of local NIMBY disputes, it's a recognized national pattern, and organizers can now cite a midterm-election-level story to make that case to a skeptical council member.

The bubble question is also a negotiating window

Economists quoted in the piece — including Philippe Aghion, who won the 2025 Nobel in economic science for research on innovation-driven growth — point out that every past infrastructure boom of this shape (railroads, electrification, the dot-com buildout) has been followed by a bust before the technology's benefits fully materialized. Nobody knows if or when that happens to AI. But the possibility cuts two ways for a community negotiating right now:

• Leverage now may be as good as it gets. A developer racing to lock in capacity while the "arms race" framing holds (the Times quotes Oxford economist Carl Benedikt Frey: "if they don't invest, they are acknowledging defeat") is under real time pressure to get local approval fast. That urgency is exactly what makes binding community-benefit terms negotiable before groundbreaking — see the model CBA clauses — in a way they may not be once a project is already operating.
• Decommissioning and stranded-asset language matters more, not less. If capital spending this far ahead of realized revenue does correct, the facility that gets built fastest is also the one a community most needs a binding decommissioning bond and site-restoration clause for. We covered the same stranded-asset dynamic on the gas-plant side of this equation in our post on Meta's Louisiana buildout — the logic holds here too.

What to do with it

1. Don't wait to see if the boom slows before you organize. The Times' own numbers say it's accelerating, not cooling — if a proposal is active near you, the Start here wizard builds a negotiating packet in minutes.
2. Translate "gigawatts of compute" into your own numbers. The $40–60B-per-gigawatt figure is a national average, not what's proposed in your county — run the actual facility size through the impact calculator for water, electricity, and rate-impact estimates specific to your state.
3. Expect federal policy to keep favoring speed. The same week this piece ran, the Times reported on the White House's AI framework pushing to fast-track approvals nationally. As we wrote after Amazon's Pecos County, Texas plant, that shifts real leverage toward local air permits and zoning review — the layers federal fast-tracking reaches slowest.
4. The capital is real, which means the ask can be too. $750 billion in hyperscaler spending this year alone means a "we can't afford community benefits" objection from a developer this size rarely survives scrutiny — see what similar communities have already won.

The Times went looking for a comparison big enough to describe this build-out and landed on the Manhattan Project. Whatever you think of that comparison, the practical takeaway for a community is the same one that document always has: the people who show up to the hearing before the decision is made get a very different outcome than the people who show up after.

Sources

• The New York Times — "The Impending, Inescapable Deluge of A.I." (Jul. 29, 2026)
• Epoch AI — AI chip ownership data (H100 equivalents)
• IDC — AI infrastructure spending forecast to eclipse $1 trillion by 2029
• The New York Times — the national AI backlash movement (Apr. 27, 2026)
• The New York Times — the White House's AI framework (Aug. 4, 2026)
• BIS Working Paper 1367 — past technology infrastructure booms and busts