Topics Pioneers Insight Method Author
Back

Why the Tech Right Lost Big This Week

·Hard Fork

Rendering the long image…

Topics AI Data Center Power Bottlenecks Trillion-Dollar AI Capex and Bubble Risks Frontier AI Regulation and Existential Risk
Pioneers Ezra Klein Erin Griffith Theodore Schleifer

Summary

  • AI’s political backlash now threatens the industry’s permission to build. Ezra Klein cites exit polling in which 56% called AI and data centers a crucial or major voting factor, 57% expected AI to hurt society versus 24% expecting benefits, and 76% opposed a local data center. The Republican position—“let them rip”—left an opening for opponents who understood the technology and rejected its trade-offs.
  • Texas’s election result puts a substantial share of next year’s planned AI capacity at risk. Erin Griffith says 10–15 gigawatts of roughly 45 gigawatts expected online next year could be endangered because investors assumed Texas’s construction moratorium would disappear after the election. Continued restrictions could tighten compute supply, raise prices, and weaken growth across the “AI trade.”
  • Slower construction could also protect investors from a capital-spending bubble. Klein argues that companies may be overbuilding because nobody wants to fall behind, while the Nasdaq’s Wednesday rebound contradicted investors’ initial catastrophizing. Externally paced investment could push labs toward economically useful deployment: “Not at all obvious to me that a slowdown in data center construction from where we are right now is bad for that.”
  • Political spending produced sharply different returns, with Musk becoming an electoral liability. Theodore “Teddy” Schleifer estimates Musk spent $470–$500 million; roughly half, perhaps more, centered on Texas and Ken Paxton’s losing campaign. Klein calls Musk an “attentional super object” whose money amplified the accusation that he was buying the election, although Schleifer says Paxton probably would have lost without it.
  • California’s wealth-tax result rewards targeted spending without removing national tax risk. Sergey Brin contributed over $100 million to an effort whose wealthy backers raised about $333 million against the measure. Klein nevertheless predicts a wealth-tax proposal will become “table stakes” for Democrats in 2028, arguing federal administration and jurisdiction make a national version stronger than a state measure.
  • AI lobbying is encountering stronger resistance than the crypto playbook anticipated. The Brockmans delivered $25 million of a pledged $50 million to Leading the Future, then withdrew the second installment as Greg Brockman called the contribution a “distraction”; Meta also retracted roughly $20–$25 million in state political giving. Crypto’s deterrent suffered too: Sherrod Brown won despite $45 million in Fairshake attack ads.
  • Schools and energy permitting are the next concrete tests of political consent. Griffith describes Google conditioning school Search access on acceptance of AI products—“You don’t want our AI? Well, fine, you don’t get our Search either”—before promising to work through the issue. Klein’s nearer legislative watch is a bipartisan permitting bill that could accelerate energy construction while making data centers bear more of their costs.
  • The next funding cycle could pit AI acceleration money against newly liquid AI-safety donors. Schleifer expects Musk to remain involved—“I would be shocked if Musk did not support Vance”—while an anticipated Anthropic IPO could enable people in its community to contribute $1–$3 million apiece. His central limitation: increasingly opaque nonprofit funding means public donor rankings reveal only part of the contest.

Deep dive

1. AI became an electoral liability beyond Trump’s other problems

  • Ezra Klein separates the tech right’s responsibility from Trump’s tariffs and war in Iran, but argues it helped produce his unpopular AI position. Exit polling he cites found 56% considered AI and data centers a crucial or major voting factor; 57% thought AI would hurt society, versus 24% expecting benefits; 76% opposed a data center in their community.
  • The Republican prescription—“let them rip, build as many of them as you can”—created identifiable local targets, including Jon Husted in Ohio. Max Read’s chicken-and-egg question remains useful: was Trump dragging down AI, or was AI dragging down its political sponsors? Klein sees both an unattractive proposition and an untrustworthy rollout.
  • Erin Griffith identifies the industry’s mistaken inference: because people use Facebook, Amazon, iPhones, and a ChatGPT she says has a billion users worldwide, investors assume opposition cannot matter. The tech right’s posture became, “We’re so powerful and rich that we don’t have to convince you.” The election challenged that equation between usage and consent.

2. The inevitability argument fails to answer whether people want AI

  • Klein’s alternative begins with regulation as a “trustworthy” and at least “semi-democratic container” for managing trade-offs. His Abundance question—“What do we want more of, and how do we get it?”—exposes the missing premise: Silicon Valley knows how to fund more AI but has not convinced people to want it. That is more than a messaging problem.
  • Jon Ossoff’s victory speech linked Silicon Valley’s bunker-building titans to their own warnings about human extinction. Klein says lab employees are taking concerns directly to legislators such as Ossoff. Politicians are therefore absorbing the industry’s internal doubts alongside its public promises, rather than simply responding to an outside caricature.
  • El-Sayed’s discussion of distillation impressed the panel because it challenged the China-race argument technically. Klein’s skier-and-boat analogy: a boat keeps accelerating because the skier it is towing remains close behind. “Well, it’s because you’re pulling the skier.” Klein sees younger politicians acquiring enough expertise to dispute the industry’s premises.
  • Klein reports insiders saying, “If I could, I would push the button. I would stop it for now,” while invoking China and competitors as reasons they cannot. Griffith distinguishes investors: they are less worried about existential risk and more worried regulation will constrain data centers and compute. Their concern itself contradicts technological inevitability.

3. The data-center revolt is concentrated where construction is concentrated

  • Griffith preserves a salience caveat: data centers had ranked near the bottom five of roughly 30 issues in polling discussed the previous week. Yet a direct yes-or-no question produced emphatic opposition. That does not establish how many voters they brought to the polls; it does establish a continuing political obstacle.
  • Klein calls the issue “spiky”: Texas, Ohio, Georgia, and Arizona had substantial construction and strong Democratic performances. Texas, which he describes as building the most data centers, was particularly exposed in its governor’s race. Ohio Democrats targeted Ramaswamy and Husted. The wedge could attract Republicans; it need not explain a race somewhere like Alaska.
  • Griffith says investors were especially alarmed by Texas because 10–15 gigawatts of roughly 45 gigawatts expected online next year could now be endangered. They had expected the pre-election moratorium to lift afterward. Constrained capacity would mean more expensive compute and slower growth for frontier labs, chip suppliers, neoclouds, and cloud resellers.
  • Read does not assume construction stops entirely: projects could take longer or move to states such as Louisiana or Oklahoma. Klein’s explanation for local resistance remains blunt: facilities are “big and ugly,” “don’t create jobs,” and face the ordinary difficulty of winning acceptance for industrial construction.

4. Restraining construction might improve the economics of AI

  • Griffith heard predictions of a market collapse, economic damage, and China winning. Klein notes Nasdaq futures initially fell, but the market bounced when trading opened Wednesday. His contrarian explanation is a suspected “collective rationality problem”: each company spends to avoid being left behind, producing aggregate overinvestment. He floats Oracle’s demise as one possible consequence.
  • Externally paced capital spending could prevent fragile entrants from launching or surviving, while pushing Anthropic and OpenAI toward diffusion and economic usefulness. Klein’s condition is alignment between construction, product value, and adoption. A capital-spending bubble is bad for markets; so is creating dangerous AI before useful AI. Slower building is therefore not obviously harmful to investors.

5. Musk’s visibility and the tech right’s candidates became weaknesses

  • Schleifer estimates Musk spent $470–$500 million, with probably half or somewhat more concentrated on Texas and Paxton. Klein calls that spending “hugely counterproductive”: the money sought attention, but Musk’s status as an “attentional super object” made the closing message that a billionaire was buying voters’ election.
  • Schleifer is more cautious about causality: Paxton probably would have lost even if Musk spent nothing. Read adds that Musk’s prominence in Texas and national right-wing politics might have made him a campaign character regardless.
  • Ramaswamy’s final Ohio ad featured Amy Acton’s mother calling her daughter a liar after Acton described childhood abuse by her stepfather. Klein calls the tactic grotesque and dishonorable. His broader diagnosis is that Ramaswamy and JD Vance “speak Trump with an accent”: imitation produces cruelty without Trump’s political feel, like a “GPT-3-level LLM.”
  • Klein argues the tech right worked better as a wealthy faction behind a charismatic coalition leader than as its public face. Its preferred candidates understand its intellectual agenda, but that does not make them appealing politicians. “It doesn’t understand and often doesn’t seem to like human beings”—including the voters it needs.

6. California rejected a particular wealth tax, not necessarily the category

  • Klein was surprised the California wealth-tax proposition failed amid pervasive anti-billionaire sentiment. Griffith thinks voters distrusted its design: opponents called it poorly written, difficult to sustain, and a one-time measure. Both distinguish hostility toward billionaires from confidence that a particular ballot proposition would improve California’s finances.
  • Schleifer supplies the spending mechanism. Brin put over $100 million into Building a Better California; wealthy opponents collectively raised about $333 million, while the union-backed proponents raised almost nothing. At recording, Schleifer said the measure seemed headed for defeat. “They were able to demolish this thing financially”—a clear counterexample to claims that money never works.
  • Gavin Newsom’s opposition mattered to Klein, as did fears that billionaires would leave the state. He predicts wealth-tax proposals become “table stakes” for Democrats in 2028, including Newsom. A national measure would have the IRS and federal jurisdiction behind it, with relocation abroad a less credible threat than moving between states.

7. A political reset requires a more desirable technological future

  • Griffith expects companies may hire lobbyists and cultivate Democrats, but doubts right-wing cultural loyalties will vanish quickly. Reid Hoffman is taking a victory lap while calling for money out of politics despite his own large donations. She does not expect the much-discussed “big forgetting” to happen rapidly.
  • Klein sees an opening for “a normal tech CEO,” exemplified by Satya Nadella: someone visibly running a company rather than becoming a political influencer. He does not see a left-wing equivalent of the All-In group emerging to replace the right-wing network.
  • Read contrasts the moment with 2008, when Democrats and Silicon Valley shared an optimistic story about the internet, social connection, green energy, and recovery. Klein asks what Democrats positively want from technology now. Drug development could provide one answer, but the party needs more than opposition to billionaires, job losses, and utility costs.
  • Klein rejects a successful-AI future where humans become “well-kept pets” of superintelligent systems as intrinsically uninspiring. With children aged seven and five, he finds their future preparation for work unclear. The difficulty is substantive: people may reject even the future builders describe as going well.

8. Schools and energy permits will test whether the industry can negotiate

  • Griffith highlights reporting that Google would turn off Search for schools unwilling to use its AI products. Chromebook penetration makes that consequential. Google subsequently said it would work through the issue, but she sees the original posture as a familiar refusal to accommodate customers: “I’m taking my ball, and I’m going home because you don’t like me.”
  • Read expects keeping AI away from children and education to become a major issue for millennial parents over the next two years. A company appearing unresponsive gives campaign managers and legislators an obvious point of intervention.
  • Klein is watching a bipartisan permitting bill involving Democrats Sheldon Whitehouse and Martin Heinrich and Republicans Shelley Moore Capito and Mike Lee. He acknowledges parts he does not understand. Broadly, it would ease construction of green and non-green energy while trying to internalize more data-center costs. Amendments could become a vehicle for the backlash.
  • At his latest check, Senate movement was expected within a week or two, with negotiations extending over several months. The lame duck may be decisive: Klein says the theory is that failure then makes passage unlikely afterward. A Democratic Congress could make agreement harder despite Democratic sponsorship, particularly with more natural gas being built to power data centers.

9. Prediction markets became ordinary without proving predictive superiority

  • Read’s surprise is how little drama Kalshi and Polymarket generated despite ubiquitous advertising and television visibility. They seemed broadly to get the election right, although he puts “right” in quotation marks because interpreting probabilities is an epistemological question. His main observation is normalization within the election-news apparatus.
  • Klein pushes back with Polymarket’s Mississippi Senate market: actual information arrived, then the lines abruptly reversed. That demonstrated fast reaction, not necessarily advance insight. Griffith adds that markets often track polls. Read accepts the exception while retaining his narrower point about their increasingly routine presence.

10. Donor returns undermine the myth of automatic political conversion

  • The top 20 donors supplied about $2 billion, more than $800 million connected to technology. Marc Andreessen and Ben Horowitz themselves spent $100 million. Schleifer questions his beat’s assumption that money reliably buys wins: Musk’s roughly $300 million in disclosed 2024 spending was perceived as successful, but that narrative did not establish how much it caused.
  • Hoffman’s largest publicly identified investment was $17 million supporting James Talarico, including a competitive Democratic primary. Talarico’s unexpectedly large general-election victory vindicated Hoffman’s judgment about the candidate, while raising a counterfactual Schleifer keeps open: perhaps Hoffman did not need to spend anything in the general election.
  • Fairshake’s 2024 success narrative followed fundraising of roughly $100–$150 million from Coinbase, Ripple, Andreessen Horowitz, and others. Its biggest November 2026 test was Brown: $45 million in September–October attack ads failed to defeat him. Brown had lost in 2024, partly, Schleifer thinks, because of roughly $40 million in crypto spending.
  • The timing illustrates the threat model. Democrats blocked the Clarity Act in September; the next day came the punishment campaign against Brown. Super PACs can bank money as a “permanent looming threat.” Brown’s victory punctures that deterrent, although Schleifer stresses crypto won other races and should not be treated as universally defeated.

11. AI’s stronger public salience disrupted the copied crypto playbook

  • Leading the Future reproduced Fairshake’s model with AI donors, notably Andreessen Horowitz and Greg Brockman. Schleifer identifies the crucial difference: people hold stronger opinions about AI. Crypto’s low salience can make accepting its support strategically attractive; AI money can bring a more substantial political cost.
  • Greg and Anna Brockman pledged $50 million in 2025, matched by a $50 million commitment from Andreessen Horowitz. The firm delivered its commitment. The Brockmans gave the first $25 million, then backed out of the second installment a year later. Internally, Greg said his support had become a “distraction.”
  • Meta also retracted approximately $20–$25 million from state political giving as the politics worsened. Read sees employee pushback returning after its suppression in recent years. The implication for executives is attachment risk: a public donation can become a story inside the company as well as outside it.
  • Schleifer adds a major qualification: retreat from disclosed giving need not mean retreat from politics. He says less is disclosed than ever, especially through Democratic nonprofits. Newly formed intermediaries can suddenly fund $30 million super PACs. Fear of Trump’s threats and investigations helps explain secrecy, but leaves outsiders unable to identify much of the underlying money.

12. Musk’s next campaign may face a growing AI-safety donor base

  • Schleifer traces Musk’s attempted lower profile to Wisconsin’s spring 2025 judicial election, when his spending and personal involvement made him central to Brad Schimel’s defeat during DOGE. Musk subsequently floated a third party without implementing or recanting it. This cycle, he said little publicly and deleted a “Vote red today” post after election night.
  • Schleifer reads Musk’s warning about a possible Democratic trifecta in 2029 as a signal he will participate in the 2028 Republican primary. Musk’s relationship with Vance has remained good despite difficulties with Trump. “I would be shocked if Musk did not support Vance.” Other industry donors’ choices will depend more on candidates’ positions.
  • AI-safety money could become “a force in the Democratic primary,” Schleifer predicts: donors around Anthropic, effective altruism, and parts of the Democratic world favor slower development or take existential risk seriously. Dario Amodei could appear on a future top-donor list. Eric Schmidt and Dustin Moskovitz already combine Anthropic investments with substantial Democratic giving.
  • Read anticipates an Anthropic IPO creating more wealthy donors. Schleifer says influence need not require billionaire status: people in the Anthropic community could give $1 million, $2 million, or $3 million. Democratic fundraisers are watching for newly liquid people willing to host candidates such as Josh Shapiro. The political consequences remain prospective, and undisclosed giving will complicate any accounting.