All-In Hosts Dissect Midterm Losses, Trump's Approval Slide, and AI Data Center Fears

The episode opens with a land acknowledgment for the Comanche, Kikapoo, Tonkawa and other peoples associated with the Austin, Texas location. Host Jason Calacanis then jokingly says the show is searching for a fifth bestie, citing the panel's composition as too white and too male, and frames the search as a DEI initiative.

The hosts then discuss the agenda. Sacks offers SaaS news, and a segment called Science Corner is set to cover the exponential time hypothesis. The hosts say the livestream for election night drew about 400,000 viewers, and a correction later gives the figure as 500,000 live viewers.

Chamath corrects the earlier audience figure, saying the livestream drew about 500,000 people live on election night. He compares that to other streams, saying far-right streamers drew about 87,000 and far-left streamers about 250,000. He says the All-In figure was larger than most cable news channels.

Chamath's quote states: "We got 500,000, which was bigger than most cable as well."

Chamath calls the result a shellacking and a blowout, saying it was the biggest midterm loss since the Watergate midterms. He says the vice president, JD Vance, explained the defeat well, and adds that the war and its economic consequences were the cause.

Chamath describes the economic chain. Gas prices rose from about $3 before the war to over $4, and diesel rose further. Those costs fed into food prices and transportation, and fertilizer costs also increased. He says inflation had fallen to 2.4% in January, down from 9% under Biden, but later rose to 3.7%.

He says the market had expected multiple rate cuts that year and that the administration had forecast 5% GDP growth. The higher inflation instead led to no rate cuts and then a rate hike, pushing mortgage rates to around 7.5%.

Chamath states: "It was this war. And the only thing I would add to that is it was this war and the economic consequences that follow from the war."

The hosts discuss a group of podcasters and media figures who had supported Trump and then broke with him over the war. Chamath says losing them mattered because Republicans depend on the podcast circuit, and that many of those listeners voted protest in the midterms to signal opposition to the war.

The panel reacts to quotes from Ted Cruz, who called the result a Watergate-level repudiation, and Chamath's own description of it as close to an extinction-level event for the GOP. Friedberg is asked to respond to these reactions and to the question of why the result became a tsunami.

Chamath states: "We lost a huge part of our base. And I would say that the Republicans in particular are dependent on the podcast circuit."

Chamath says he got the midterm reading totally wrong and identifies his blind spot as viewing events through his own experience as an asset owner and his narrow self-interest. He says the broader electorate's main concern was the economy, not the culture wars.

He argues money did not decide the election and that gerrymandering failed, noting that of 15 Republican seats redrawn, Democrats won 11 and Republicans only one. He says the vote was about affordability and how the economy works, not about Iran, Israel or AI. He concludes that a 2028 winner will need an economic populist platform with reasonable standards of economic fairness.

Chamath states: "I got it totally wrong."

Chamath says he believed asset owners building assets and investing would benefit everyone else, but says that is clearly not happening and he has to update his priors. He proposes closing loopholes, stopping federal underwriting of loans, getting rid of student debt, and zeroing out federal taxes below roughly $150,000.

He predicts that until wage earners feel ownership of the economy, Democrats and Republicans will keep swinging away from the status quo. He warns that if Democrats hold both houses and spend their time impeaching Trump rather than legislating economic fairness, they could lose the way Republicans did.

Chamath says Sacks is right that the endless war and inflation were the last straw, but disagrees on what came before. He argues the polls and prediction markets were accurate, while many on the right wrongly said they were manipulated. He cites a Silver Bulletin chart showing Trump's net approval fell from above +5 at the start of the term to -28, a swing of more than 30 points.

Chamath attributes part of the decline to the Liberation Day tariffs, which he says spiked inflation and which the Supreme Court ruled against. He also points to ICE enforcement, including masked agents and absent body cameras, as a factor. He says he has shifted to voting roughly 55% Democrat and 45% Republican because of what he calls left-wing lunacy.

Sacks presents charts. He says Trump's net approval maps closely to gas prices. He also shows that Senate prediction markets for 2026 showed Republicans expected to win by a large margin throughout the first year, and that the April 2025 tariff rollout did not move that outlook. He says the odds began to change in March 2026, after the war began on February 28, 2026.

Sacks also cites a survey in which about 74% of respondents said the Iran war was a crucial or major factor in their congressional vote, with those voters breaking heavily for Democrats. He notes this includes independents and some younger Republicans, while older Fox News viewers remained more supportive of the war.

Sacks states: "Trump net approval maps to the price of gas."

The hosts discuss Gina Hinojosa, identified in the episode as Texas's new governor-elect, who they say won unexpectedly. In a New York Times interview they cite, she declined to engage with culture-war questions and instead made the case for economic populism, tying it to data centers.

The speakers argue that data centers are unpopular because they raise electricity bills and use water, and because people perceive oligarchs profiting while they pay more. One speaker says the oligarchs get tax breaks while voters face higher utility bills, which the hosts connect to a sense that ordinary people are unheard.

Friedberg argues that actors seeking central planning and authority will react strongly against AI, which shifts power away from them. He says calls to stop AI development are effectively calls for those actors to control AI, because it is too powerful for them not to control.

He predicts that regulators will focus on compute rather than chips or models, since software is hard to regulate and its use is constrained by access to compute. He expects data centers to be treated like utilities and says the government will seek a tax benefit, while data center policy is partly performative and aimed at a better deal.

A speaker proposes a phased increase to $15 an hour, raised over four years in two-year steps. He says only about 10 to 15 million people earn under $15 an hour, so the cost is manageable, and that it would counter the narrative that Republicans do not care about the bottom half of society. He identifies himself as a moderate independent and anticipates being called a 'libtard' for the idea.

He also lists other ideas he would support, including a ban on trading stocks while in office and on doing business with foreign countries, and a housing plan to incentivize new home construction. He says if Trump solves two or three of five such issues, it would redeem his presidency, and that ending the war is essential.

Sacks offers three historical examples. Clinton lost the House in 1994, when Newt Gingrich became speaker, but won re-election in 1996 after making course corrections. Obama lost seven Senate seats and 63 House seats in 2010 and still won re-election in 2012.

Sacks contrasts these with George W. Bush, who lost badly in the 2006 midterms over the Iraq war. Instead of adjusting, Bush doubled down with the surge, and Republicans lost again in 2008. Sacks argues that Trump should course-correct and exit the war now to have a successful next two years, and warns that neocons encouraging a doubling down would repeat the 2006 to 2008 pattern.

The hosts turn to markets after a week when stocks including Nvidia, SpaceX, Tesla, AMD and Micron fell, with close to $4 trillion in market value erased. One speaker argues that the deals behind data center buildout were gratuitous, questioning whether the largest seven or eight companies needed tax breaks to invest tens of billions. He expects incoming governors to demand directly measurable economic impacts, which would slow buildout.

He says a slowdown would benefit hyperscalers such as Amazon, Meta, Microsoft, Google and SpaceX, because they have existing capacity. He says it would hurt neoclouds, which are renters, and foundation model companies that are short on capacity. He argues hyperscalers could allocate incremental compute to their own products, making them kingmakers.

The speaker states: "It sets the hyperscalers to be a kingmaker and they have this interesting game theory now."

Chamath argues that if governors tighten rules on big tech hyperscalers and neoclouds, data centers could be built in Canada, Mexico, Singapore, Malaysia and other countries. He says the Middle East is off the table for now because of the war but could be used once it ends, including Saudi Arabia, the UAE and Qatar.

He predicts the next large model could be trained outside US soil, that the 300,000 to 400,000 construction jobs would go away, and that the tax revenue would be collected locally abroad because data can move quickly around the world. He concludes: "We're going to be a net importer of tokens."

Asked whether the US could become a net importer of tokens, the discussion then turns to proposals some on the left have floated to ban AI development, deployment and use, which Chamath describes as a far-out prediction that is being floated.

Freeberg argues that the Second Amendment's purpose was to keep a check on power, and that if the state monopolizes AI and no one else has it, that check disappears. He says the founding intention of the country was individual liberty and citizens' rights, not state power.

Jason responds that this is a Doom scenario and that he sees about a 1% chance of something like this happening. He says Freeberg's perspective serves as a backstop against overreach and asks Chamath and Sacks for probability estimates on whether such a scenario is possible.

Freeberg argues that a regulatory deal may cut in favor of large companies, with government gaining power and companies gaining regulatory protection. Chamath counters that if AI is legislated broadly, the supply constraint would hurt the renter class of foundation model companies, leaving real control with Meta, Amazon, Google, Microsoft and SpaceX.

Freeberg says foundation model companies are not secure because they are renters rather than owners of data center capacity. He notes that Google recently took over a CoreWeave data center from a renter, which Chamath explains is because Google controls the power. Chamath argues regulatory capture could only work if the government also granted free electricity and ignored energy rules, which he says no one has the money or political will to do.

Sacks says the hosts should separate federal AI safety rules from state and local infrastructure rules, and that infrastructure will be handled at the state level. He says people are too optimistic about the election's effect on data centers, arguing Gina Hinojosa said there would be no new permits until legislation passes, and that she did not say existing projects are protected.

He cites a semi-analysis analyst who noted that projects under construction may still need permits such as air pollution permits, which can come only at the end. Sacks warns that if no new permits are passed until a new regulatory regime takes effect, the result could be stranded, half-finished projects.

The hosts discuss a clause in which Musk could end his Anthropic arrangement after 90 days. One speaker says this is not about the working relationship or the quality of either company but a spreadsheet question: if spot prices spike beyond a threshold, the economics force the decision.

Another speaker argues Musk has aligned himself closely with the Anthropic team and that his acrimony with OpenAI and Sam Altman, along with Anthropic's loyalty and success, could lead him to seek a middle ground rather than move to spot pricing, particularly for space data center plans. The first speaker responds that the point is about existing contracts becoming negative in economic value, forcing renegotiation.

Freeberg explains the core question: whether a problem can be solved by a computer and how long it takes. He describes P problems, which can be solved in polynomial time, and NP problems, which can be verified quickly, using a Sudoku example. He notes the open question of whether P equals NP.

He describes the 1999 exponential time hypothesis, which claimed that a problem called 3-SAT takes exponential time and thus that P is not NP. He says an OpenAI paper of 175 pages claims 3-SAT can be solved in subexponential time, disproving the hypothesis, and that the work was done end to end in under 10 days. The panel notes that the proof has so far held up, though it had not been fully verified by third parties.

Sacks says AI or superintelligence will create many good things, and that fear around it may be partly justified because powerful technology frightens people. He says the future is unknown and cannot be fully predicted, but he expects many benefits.

He argues that if the US does not build the data centers, they will be built in other countries and the US will pay to import tokens. He concludes that if the US does not pursue AI, China will benefit.