Data Center Opposition Is Shifting Toward Demands for Local Guarantees

Democrats won majorities in the House and Senate, while local voters also pushed back against new data centers. Lon Harris said more than a dozen Ohio towns voted to halt or restrict development, and similar measures passed in Kansas, Florida, and California. Texas governor-elect Gina Hinojosa campaigned on pausing construction until the legislature could establish rules. Texas, Georgia, and Ohio together account for more than 40% of planned U.S. data-center capacity, Harris said.

The political signal arrived alongside market anxiety. After an initial post-election rally, technology stocks fell roughly 5% to 6% on Thursday. Harris cited reports that OpenAI leaders were discussing whether to reset expectations for next year’s capacity, and that Microsoft had canceled a roughly $3 billion deal with UK data-center developer Nscale. For investors and builders, the results made it harder to assume that local opposition would remain only a campaign issue.

Jason Calacanis did not dismiss that opposition. He argued that campaigning and governing create different incentives: candidates can run against development, then negotiate with developers once elected. In August, he had predicted that after the election, officials would announce agreements under which data centers would pay for electricity, address local concerns, and contribute visible benefits to their communities.

The prediction was not that opposition would vanish, but that it could become a negotiation over terms. Harris read from an X post by investor Gavin Baker describing Hinojosa’s position. Baker said she had indicated that already-permitted Texas facilities could finish construction, while newly permitted projects expected to come online in 2028 and 2029 could face conditions. In Baker’s account, those might include paying taxes, preventing increases in consumer power prices, protecting water availability, using union labor, and giving Texans a seat at the table. Baker called the terms reasonable and said they could preserve jobs and property-tax revenue in rural towns. The post described Hinojosa’s position; it was not presented as a finalized set of rules.

Calacanis agreed with the general direction: developers should be good neighbors. He said they should make sure water and electricity are sustainable, avoid pollution, pay taxes, create jobs, and contribute to the local community. If companies meet those conditions, he expects communities to compete for projects and developers to improve their offers. The bargain, in his telling, is not simply to permit construction; it is to make local costs and benefits explicit.

That argument shaped his view of President Donald Trump’s record. Calacanis gave Trump an A to A+ on AI and data-center policy because he had made growth, energy, and infrastructure priorities and defended projects when opposition was rising. Harris pressed him on whether Trump had made data centers more partisan by putting himself at the center of the issue. Calacanis agreed that he had turned the issue into a wedge, forcing Democrats to take the other side.

The two men were not equally confident about what comes next. Calacanis expects Democrats to govern toward the center because they need jobs and economic growth, and because a prolonged slowdown would be blamed on them. Harris was less sure that voters who supported anti-data-center candidates would accept a compromise once those candidates took office. The election did not settle whether new facilities will be built. It made the conditions for building them more politically important.

Calacanis characterized Thursday’s market drop as an overreaction. He interpreted the post-election rally as reflecting hopes that the Iran war would end and tariff uncertainty would ease, which he said would be bullish for stocks. He also argued that divided government could provide checks and balances. He described the decline as a possible buying opportunity, calling that view “incredible non-investment advice.” Harris said he had bought some shares during the dip.

The political compromise Calacanis expects would address immediate local impacts. But a deal over water, power, taxes, and jobs does not resolve a broader question raised in the discussion: who benefits from AI, and what happens to workers whose jobs may be displaced?

Lon Harris recalled an earlier Calacanis argument that AI could retire jobs across trucking, taxis, delivery, factories, and Amazon warehouses. Calacanis said those workers would need to be redeployed, and asked whether the technology’s gains would create a permanent underclass of people who had lost their jobs. In that archived discussion, he described a divide within the Republican coalition between technology companies and wealthy owners, on one side, and people worried about jobs and concentrated wealth, on the other.

That tension complicates a simple reading of data-center opposition as opposition to growth. The facilities are part of the AI buildout, but voters may associate that buildout with job loss, concentrated power, and pressure on local resources. The proposed local bargain can address some costs of construction. It does not, by itself, settle who shares in the gains or bears the disruption.

For Calacanis, the case for data centers is not just that AI matters or that construction supports economic growth. Residents need to know whether a particular project will impose costs on them. Water use, electricity prices, pollution, noise, and a facility’s proximity to homes are concrete concerns; broad claims about national competitiveness do not resolve them.

That is why he welcomed OpenAI’s commitments announced after the election. As Harris described them, OpenAI pledged to pay the full cost of powering its data centers and said that for facilities it owns and operates it would seek community input, publish construction plans, and avoid nondisclosure agreements with public officials. A separate policy blueprint encouraged lawmakers to enact similar commitments as law. Harris said Meta, Microsoft, Google, and Anthropic had also outlined commitments of their own.

Calacanis did not call OpenAI’s proposal revolutionary. He saw it as useful communication, but said companies would have to keep making the case and show that their commitments meant something. His practical theory of leadership was to listen, repeat back what residents have said so they know they have been heard, and then repeat the company’s commitments until people remember them.

That repetition can sound tedious. Calacanis illustrated it with the example of a company repeatedly saying that a data center will not raise local electricity bills, will use recycled water, and will pay taxes. But he argued that repetition is part of the work: a company cannot make one announcement and assume people who distrust the project will be reassured.

Harris added that data centers need not remain a partisan identity marker. He compared them with big-box retail, where communities have long debated public incentives and whether taxpayers are paying for supporting infrastructure. Calacanis agreed that asking who pays for a project is a valid criticism. A community can oppose a particular facility or demand stronger rules without rejecting AI altogether; companies, in turn, cannot treat a general argument about growth as an answer to a town’s specific concerns.

The point is not that local objections will disappear once a company communicates better. It is that the argument for building has to address effects residents can observe and specify who is responsible for them. In the compromise Calacanis expects, permission depends on companies agreeing to pay, limit disruption, and contribute locally.

The election’s local disputes were part of a wider argument about who bears the costs of technological change. Harris and Calacanis described affordability, opposition to corruption, and opposition to new wars as prominent themes in the election. Calacanis put the cost of living first and said the Iran war was intertwined with it. He also agreed that voters were concerned about the state of democracy and the perception of grifting. On possible conflicts involving Trump’s investments, he said he would wait for investigations to establish whether the president’s money was held at arm’s length or whether he had used access to make trades.

The two men also discussed the Democratic campaign’s emphasis on affordability rather than cultural-war disputes. Calacanis argued that candidates gained ground by returning to practical concerns instead of defining voters or opponents through caricature. He cited Texas Senate candidate James Talarico as an example of someone who, after winning, spoke respectfully about political opponents and technology leaders. Talarico said he respected Elon Musk and would like to meet with him despite Musk having spent heavily against him; he also said he looked forward to governing with Senator Ted Cruz.

Calacanis saw that posture as an invitation to govern rather than prolong the campaign fight. He argued that Republicans should stop describing the left as uniformly socialist or culturally extreme, just as he had criticized Hillary Clinton’s “deplorables” framing of Trump supporters. Harris said Talarico seemed like a normal person rather than the caricature used against him. Their point was not that partisan divisions had disappeared, but that candidates could gain ground by speaking to practical concerns without inflaming every cultural dispute.

The data-center question fits that wider frame but does not collapse into it. A promise to create jobs or grow the economy may not answer whether a particular town will face higher costs; a promise to limit local impacts may not answer how AI’s gains will be distributed. The local bargain is more concrete, and perhaps easier to negotiate, than the question of who benefits from the technology over time.

Calacanis’s tariff argument illustrates the method he said he used to assess policy: look beyond politicians and commentators to people directly affected. He recalled criticizing Trump’s meme coins early in the administration, arguing that a president and his family should not launch speculative ventures while the government was still setting crypto policy. He also said he had raised concerns about corruption and loss of focus before those issues became central to public discontent.

On tariffs, Calacanis said the administration had identified real trade imbalances but handled them chaotically. In an archived discussion, investor David Sacks described the approach as a series of trial balloons rather than a planned strategy: instead of sorting trading partners by importance and explaining the intended negotiations, Trump would react to a proposed figure with a much higher one. Calacanis agreed that the administration had a legitimate subject to negotiate, but argued it could have done so privately, with a clear strategy and less public escalation.

To understand the tariffs’ effects, Calacanis said, he spoke with businesses. He pointed to Steven Borrelli, chief executive of clothing company Cuts, whom he had interviewed about tariffs’ effect on the company. Calacanis said that when Republican friends accused him of reflexively opposing Trump, he went back to people facing the policy in their businesses. Harris emphasized that Borrelli was a Trump supporter, making his account worth considering even for people who assumed criticism of the policy came only from political opponents.

Calacanis described the approach as a way to reach a view and withstand pressure from people who disagreed: get first-hand experience, do the homework, and do not substitute group loyalty for evidence. He said that process explained how he could support Trump’s data-center policy while criticizing other choices. A person can support a policy when its case appears strong, criticize poor execution, and withhold judgment where the facts are not established.

The discussion of Anthropic’s first content-licensing agreement, with the Associated Press, raised a parallel question about who should share in the value AI companies create. Harris said the agreement covered election reporting and that Claude users asking for polling updates would receive current information provided by and attributed to the AP. The agreement was signed shortly before the midterms, and its financial terms had not been announced. Harris described the polling service as a planned use.

Calacanis argued that using a publisher’s reporting to provide a competing real-time product is different from using material as training data. In his view, when a company monetizes responses drawing on current reporting, the publisher’s work is serving a direct commercial purpose and should be paid for. He presented that as his legal conclusion, not as a settled ruling. Harris noted that Anthropic had previously maintained that its use of publisher content was permissible under fair use, and that it had not made a licensing deal with a digital publisher before the AP agreement.

The discussion separated several disputes that can otherwise get collapsed together: whether content was obtained unlawfully, whether a site’s terms of service were violated, whether material may be used to train a model, and whether a model’s output reproduces protected work. Calacanis argued that the first two issues had seen clearer industry responses, while questions about training and output remained unsettled. Harris agreed that disputes over pirated books had not resolved the separate question of whether training on lawfully obtained material is permissible.

Calacanis proposed a broad revenue-sharing arrangement as one possible way to reduce the conflict. If AI companies were generating very large revenues, he argued, they should share a meaningful portion with publishers and other creators whose work makes their services more useful. He suggested 30% as an illustrative figure, then discussed a smaller 3% pool—$3 billion a year in his example—that could be distributed across newspapers, magazines, books, and platforms such as Substack.

He was describing a possible market, not a negotiated formula: rights holders could participate or opt out, rather than repeatedly litigate over how their work was used. Calacanis argued that platforms could make participation straightforward, with publishers receiving most of the revenue and the platform retaining a smaller share. The proposal was meant to make licensing attractive enough that publishers would choose to participate.

The analogy to data centers is not exact, but the underlying question is similar: companies should not expect to capture the benefits while leaving affected parties to bear the costs. For infrastructure, that means paying for power and addressing local impacts; for AI products, Calacanis’s proposal was to pay for the content that makes the products more useful. Neither proposal, on its own, settles how much is fair or who should set the terms.

Harris raised Apple’s reported decision to reduce the daily AI spending allowance for engineers as a question about the costs of AI use. He said Apple had lowered one daily cap from $300 to $200 and was considering further cuts, with some allowances falling to $100; employees outside engineering could face a limit as low as $25 a day.

Calacanis connected the limits to constraints on data centers and energy. If access to AI depends on limited compute, he argued, companies may not be able to treat tokens as effectively unlimited. He said his own team used a $40-per-month plan, but heavy users could consume far more in tokens than the subscription price suggests. In his view, flat-rate plans encourage adoption; as usage grows, enterprise pricing could rise to $100, $200, or $300 per employee per month.

He compared that variation with other software licenses: some employees use a tool constantly, while others barely use it. A company facing higher prices might review actual usage rather than pay the same amount for every seat. That would make AI spending more like an operating decision than a standard software subscription.

Apple, Calacanis suggested, might pursue a different route by running models locally on its own hardware rather than paying cloud providers for every request. Harris agreed that Apple could let users choose between cloud-based models and models running on devices. Calacanis said Apple might arrive later than competitors but still make a strong product if its hardware and software supported local AI well.

The discussion did not establish what AI access will ultimately cost. It did make the cost question harder to separate from the infrastructure debate: data centers require power and capacity, while the services built on them may be priced according to how much people use them. If demand grows, both the local effects of building and the price of access become part of the bargain.