In which I cite @twallack.bsky.social's excellent WaPo piece on higher ed from this weekend.
"The bachelor's degree is now multiple products that share a name. The hackers priced the cheapest ones first."
One Virginia House district hosts 447 data centers. More than forty-six states. The partisan geography of AI infrastructure runs at two levels — between states AND within them. New post on what that means for the 2026 House map.
kylesaunders.substack.com/p/the-ai-dat...
We live in a world where colleges agree to follow NCAA eligibility rules, but when one of their athletes sues the NCAA to stay eligible and asks a court to override eligibility rules, colleges support their athlete's case.
This is becoming a real issue. My @sportico.bsky.social sports law column:
Degree hacking is niche — but it's the cleanest pricing signal higher ed has seen in a decade. From above: AIM rewrite votes May 22, endowment tax starts July 1, WICHE cliff peaks this year. Four vectors, one direction. kylesaunders.substack.com/p/the-bifurc...
That's why I keep pushing what I'm linking below--an open sourced dataviz/measurement/strategic piece on how different institutions in higher education are positioned, including on AI exposure, to inform and facilitate conversations. I hope you'll look and share. kylesaunders.com/university-m...
Charlie Baker called the college sports governance environment "a fragmented patchwork of state laws." Saturday bonus post inventories what's actually in it: 12+ states, six categories, four months. kylesaunders.substack.com/p/the-colleg...
"NCAA membership is not waiting for others to act." Baker wrote that while asking membership to lobby Congress for a bill the senators say won't pass. 82 min later, Oklahoma granted an eligibility injunction against the NCAA. kylesaunders.substack.com/p/no-one-is-...
Data centers don't just follow cheap electricity. After controlling for power costs and population, bluer states host significantly more. And the buildout is concentrating in swing states.
kylesaunders.substack.com/p/ai-and-pol...
Updated my higher ed structural mapping tool to v1.4. 1,556 institutions, two composite axes, every data point public. The week Hampshire and Goddard announced closures feels like the right time to look at where the stress is actually concentrated. kylesaunders.com/university-map
v1.4 of "Mapping the Structural Divide in Higher Education" is up!
Tried a few new ideas, updated the paper, but the tool remains. All open access.
Search for your institution, learn about AI exposure, and see institutional measures of resilience & market position.
kylesaunders.com/university-m...
Every time a constructivist identifies "bias" in an AI, they're assuming an objective fact the model got wrong.
But the people who built the model trained it to optimize for approval, not truth—and we MUST remember that.
Both pillars are cracking.
kylesaunders.substack.com/p/the-other-...
Every time a constructivist identifies "bias" in an AI, they're assuming an objective fact the model got wrong.
But the people who built the model trained it to optimize for approval, not truth—and we MUST remember that.
Both pillars are cracking.
I spent 5,000 words arguing AI undermines strong constructivism, but now I'm arguing the people who built AI are making the same mistake from the opposite direction.
The constructivists weren't wrong about the mechanism--just the target: the machines.
kylesaunders.substack.com/p/the-other-...
gah. thanks. :)
(yes)
New working paper: "Cheap Power, Competitive Politics." Does a state's political environment predict how many AI data centers it hosts after controlling for electricity costs & population? It does. The map matters more than the megawatts.
Explainer here: kylesaunders.substack.com/p/ai-and-pol...
Klein's piece on Piker is wrong on the coalitional math and wrong on what normalization does inside a party.
kylesaunders.substack.com/p/the-normal...
Fifty years ago the academy developed a reasonable caution: be careful about objectivity, power shapes inquiry, perspectives matter.
Then the caution mutated into dogma. And AI is stress-testing the dogma.
New piece tracing how it happened -- Heidegger to Habermas to Haraway to the present crisis:
14/
Replication data and code:
kylesaunders.com/datacenter-r...
More political context in my AI & Politics series:
kylesaunders.substack.com/p/ai-and-pol...
13/
Data sources:
DC counts — DataCenterMap.com
Power — EIA Electric Power Monthly
Margins — Dave Leip Atlas, 2024 certified
Population — Census Vintage 2024
Education — ACS 1-yr 2024
Cooling — NOAA CDD 2025
12/
Working paper — 50 states, OLS, deliberately parsimonious. County-level and panel designs come next.
But the pattern is clear: the AI backlash sits in the states where it has electoral consequences.
11/
Education and cooling demand? Not significant once population, power, and politics are in the model. The story is simpler than the industry narrative suggests.
10/
Virginia is the anomaly. 665 data centers — most in the country. Mid-range electricity (10.6¢). Harris +5.8. Its dominance comes from fiber infrastructure, not cheap power.
9/
Georgia is the proof of concept. 140 data centers, Trump margin of 2.19 points. Democrats already ran on data center backlash there. The AP called it an electoral gift.
8/
The swing states tell the political story.
8 states within a 5-point margin host 575 data centers. Georgia: 140. Nevada: 120. Pennsylvania: 90. Michigan: 80.
These aren't abstractions. They're zoning fights in competitive districts.
7/
Cheap power and Trump margin correlate at r = –.59. Cheap electricity concentrates in red states. But the model separates these effects.
Bluer states have more data centers than their power prices alone would predict.
6/
Full working paper:
kylesaunders.com/datacenter-r...
