The All-In Hosts Debate the Midterms and the AI Infrastructure Backlash

Jason Calacanis opens a post-election discussion with Chamath Palihapitiya, David Sacks and David Friedberg. They plan to examine the midterms, the remainder of President Donald Trump’s term and the implications of a sharp selloff in AI-related companies.

The panel records on Friday morning after Democrats won control of both chambers of Congress and gained governorships, including Texas. Calacanis asks what drove the Republican losses and what Democrats did successfully.

Sacks calls the result a major defeat and says he underestimated its scale. He attributes the outcome primarily to the Iran war and its economic consequences, particularly higher fuel and transportation costs that he says fed concerns about affordability.

He argues that inflation and mortgage costs worsened after the war began. Other hosts challenge his earlier optimism about growth. Sacks says he expected Republicans to fare better partly because employment and growth figures looked relatively strong to him.

Sacks says his forecast was wrong because voters’ experience of affordability mattered more than aggregate economic statistics. He also argues that the war fractured the Republican coalition, including relationships with influential podcast hosts and other media figures.

Calacanis cites reactions from Republican figures and asks Friedberg whether messaging or underlying conditions drove the result. Friedberg argues that the problem was more fundamental: in his view, government spending and the failure to reduce it contributed to the cost of living.

Friedberg recalls hopes for government efficiency and says the effort did not produce the spending reductions he wanted. He argues that many people still felt unable to afford housing, healthcare and other essentials, regardless of positive aggregate growth.

He also thinks concern that technology companies and wealthy investors capture the gains from AI became politically potent. He displays a Microsoft–Gallup survey comparing views of AI across countries and asks whether the resulting political pressure will persist.

Palihapitiya says he got the election wrong and recognizes that his own experience as an asset owner shaped his perspective. He argues that a broad group of voters has moved between parties because its economic circumstances feel broken.

He contrasts investors who benefited in different interest-rate environments with wage earners facing higher living costs. In his view, the result reflected dissatisfaction with the economy more than a move toward a particular ideology or a single issue.

He points to spending and redistricting outcomes as evidence for his interpretation, arguing that candidates emphasizing affordability connected with voters across party lines. He now favors an economic-populist agenda and says he has revised his assumption that gains to asset owners would reach everyone.

His proposed changes include addressing tax loopholes, student debt and taxes for lower-earning households. He predicts that voters will continue to swing against incumbents if they do not see economic improvement.

Calacanis plays an earlier clip in which he predicted Democratic gains, citing prices and the war. Sacks responds that the war and the inflation associated with it were decisive, and argues that ending the conflict could relieve fuel costs.

Calacanis disagrees that the decline began with the war. He points to the administration’s tariff rollout, immigration enforcement and other controversies as earlier sources of dissatisfaction, and describes the president’s execution as chaotic.

Calacanis also returns to what he calls a divided economy: people without substantial assets may not feel included in growth. He argues that the administration did not persuade those voters that it was working on their concerns.

Asked what could change the president’s trajectory, Sacks shows charts comparing approval, gasoline prices and political expectations. He argues that the war was the central turning point and says ending it would be his first priority.

He cites a voter survey on the importance of the Iran war and predicts that ending the conflict would reduce oil and fuel prices. These are his interpretation and forecast; Palihapitiya challenges the idea that the issue can be reduced to the war alone.

Palihapitiya says fuel prices are one visible expression of a broader economic problem. The hosts discuss Texas governor-elect Gina Hinojosa’s emphasis on utility costs and community concerns about data centers.

Friedberg agrees that voters may view both the war and data-center controversy through their own economic circumstances. He worries that a divided government will struggle to resolve budget and spending issues over the next two years.

Friedberg then outlines a concern that governments may seek greater control over AI through regulation of computing infrastructure. He presents a hypothetical in which widely distributed AI tools shift power away from centralized authorities.

Calacanis instead suggests that some data-center demands are bargaining over local benefits. He proposes a political deal involving healthcare, housing, minimum wages and taxes, and controversially suggests broad pardons before negotiations.

Palihapitiya argues that future political coalitions will have to address economic fairness. He also defends parts of Trump’s immigration and industrial agenda and suggests trading policy priorities to reach an agreement.

Sacks points to past presidents who recovered after midterm defeats and contrasts them with the Iraq-war-era Republican losses. His conclusion is that the administration should change course on Iran rather than escalate.

The discussion turns to an AI-related market selloff. Calacanis cites a loss of roughly $4 trillion in market value, notes that the market subsequently rebounded somewhat, and asks what uncertainty about data-center projects means for companies and investors.

He notes speculation about the prospective valuations of Anthropic and other AI companies, then asks whether incoming governors will slow data-center projects or bargain for better terms for their communities.

Palihapitiya describes a contested SpaceX presentation in Arizona and says the opposition reflects both economic frustration and concerns about the terms offered to local communities. He questions whether the largest companies need tax incentives for investments that could generate substantial returns.

His forecast is that slower construction would benefit large infrastructure owners with existing power and capacity while putting pressure on smaller cloud providers and model developers that rent compute. He names Amazon, Meta, Microsoft, Google and SpaceX in his discussion.

He describes a choice for infrastructure owners: continue renting capacity, or reserve more of it for their own products. In his view, a constrained supply could increase costs for leading model developers and give competitors more time to catch up.

Palihapitiya speculates that a reported termination option in a SpaceX–Anthropic compute arrangement could allow repricing if capacity becomes much scarcer. He presents this as a scenario, not as an announcement that either company will terminate the agreement.

Calacanis argues that if construction becomes difficult in the United States, companies could build abroad, shifting investment and jobs elsewhere. Friedberg raises a more expansive concern that policymakers could seek to restrict AI development or use itself.

Friedberg argues that restrictions on software and computing could concentrate power in the state. Calacanis regards the most extreme version as unlikely, while Friedberg says he is warning about a gradual change in the balance between government and citizens.

Sacks predicts a major push for federal AI legislation and expresses concern about surveillance and censorship. Palihapitiya disagrees that government control is the central outcome, emphasizing instead the leverage of companies that already control power and data centers.

The panel disputes whether regulation would primarily favor model developers or large infrastructure owners. Palihapitiya argues that compute renters remain exposed if the supply of power and data centers becomes constrained.

Calacanis gives his own firm as an example of growing AI use. He estimates its spending and productivity gains and predicts that organizations will increasingly use local models as well as remote services; he presents these as his own estimates and forecasts.

Sacks separates federal questions about AI rules from state and local decisions on infrastructure. He worries that a pause on new permits could leave some partially built projects waiting for approvals, while acknowledging that the details of future policy are not yet settled.

His base case is that fully permitted projects can proceed in 2027, but that the pipeline for later projects could narrow. Calacanis and Palihapitiya argue that governments and companies may instead negotiate over local benefits, or companies may choose other locations.

Calacanis says he has been buying shares of SpaceX and Tesla. Palihapitiya revisits his hypothetical about repricing scarce compute, while Calacanis argues that longer-term commercial relationships may affect the outcome as well as spot prices.

Sacks says large infrastructure investments need predictable conditions and forecasts a materially smaller project pipeline by 2028. The panel’s assessments remain predictions about permitting and capital spending, not settled outcomes.

In the science segment, Friedberg introduces an OpenAI preprint that claims to refute the exponential time hypothesis. He explains the distinction between solving a problem and checking a proposed solution, and describes the paper’s claim of a subexponential algorithm for a satisfiability problem.

Friedberg emphasizes the novelty and potential significance of the result, but also says outside researchers are still evaluating the lengthy paper. A claimed subexponential algorithm does not by itself establish that P equals NP or that every problem is computable.

The hosts discuss what progress in mathematics might mean for AI’s benefits and for competition between countries. Sacks argues that the technology has significant potential benefits while acknowledging uncertainty about its future effects.