All-In: Midterm Blowout, $4T AI Sell-Off and the Data Center Freeze

The All-In hosts dissect the 2026 midterm blowout, the $4 trillion AI sell-off and why new governors could freeze data center construction across the US. In this episode of the All-In Podcast, recorded the Friday after the election, Jason Calacanis, Chamath Palihapitiya, David Sacks and David Friedberg argue over why Republicans lost so badly, whether President Trump can rescue his last two years, what a halt on new permits does to the AI build-out, and why OpenAI’s claimed break of the Exponential Time Hypothesis matters.

TLDW

Democrats swept both chambers and flipped states like Texas, Kansas and Iowa, in what Sacks calls the worst midterm loss since Watergate. Sacks blames the Iran war and the gas and diesel inflation that followed it, Chamath says a post-ideological swing bloc is punishing whoever holds power for a broken economy, and Friedberg points to the end of DOGE and unchecked spending. The second half turns to markets: close to $4 trillion of AI market value lost in one day, a Polymarket drop in Anthropic IPO odds, and a looming freeze on new data center permits that Chamath thinks turns Amazon, Google, Meta, Microsoft and SpaceX into kingmakers. The show closes with Friedberg’s science corner on OpenAI’s claimed disproof of the Exponential Time Hypothesis.

Thoughts

A data center freeze does not hurt the AI industry evenly, and Chamath’s map of who wins is the sharpest analysis in the episode. If new permits stop, the companies that already own powered capacity (Amazon, Meta, Microsoft, Google and SpaceX) hold the scarce asset, while the neoclouds that rent it out and the frontier labs that rent it in are the ones exposed. That gives the hyperscalers three choices: keep renting as today, quietly keep new compute for their own models so Gemini and others can catch up while OpenAI and Anthropic get squeezed, or reprice. The pressure point he names is specific: the SpaceX and Anthropic compute deal reportedly lets either side walk on 90 days’ notice, and if spot prices double or triple, a rational landlord reprices. Calacanis argues Elon Musk’s relationship with Anthropic would soften that. Chamath’s reply, that “the spreadsheet will dictate the answer,” is the more useful frame for anyone modelling AI lab margins in 2027. Google taking over a Crusoe site, which he mentions in passing, is the first sign of the renter class being absorbed.

Sacks’s point about permits is the one investors should not wave away. Texas governor-elect Gina Hinojosa has said no new permits until new legislation passes, and he notes, citing SemiAnalysis, that projects under construction often still need late permits, such as air pollution approvals, before they can open. Even without a hostile governor, he says nobody has a reason to sign anything: both parties ran attack ads against politicians who approved data centers, so no ambitious official or civil servant will stick their neck out. His base case is that fully permitted projects finish in 2027 and the pipeline thins sharply in 2028, on $50 billion projects nobody will start under that kind of uncertainty. Calacanis’s counter is that the work simply moves to Canada, Mexico, Malaysia or the Gulf and the US becomes a net importer of tokens. Both can be true. The cost of a permitting freeze is not that AI stops; it is that the jobs, taxes and the training of the next frontier model happen somewhere else.

The best argument about the election itself is between Sacks and Chamath, and they are less opposed than they sound. Sacks has the data: Trump’s net approval tracks the price of gas, the Senate prediction market held steady through Liberation Day and only broke after the war began on February 28, 2026, and 74% of voters called the Iran war crucial or a major factor. Chamath, admitting he got the election “totally wrong,” says money and gerrymandering both failed (Democrats won 11 of 15 seats redrawn for Republicans) and that a swing bloc has now voted for Biden, Trump and Democrats in turn, punishing whoever fails wage earners. Gas is the trigger and the broken wage economy is the gunpowder. Sacks’s fix (end the war and Brent falls from about $110 to $70) works only if Chamath is wrong that prices were already the symptom of a deeper problem.

Friedberg’s framing of AI as “the ring of Sauron” sounds like a joke and is really a forecast. His claim is that the push to stop AI is actually a push to control it, and that because software is hard to regulate, the lever will be compute: data centers end up treated like utilities, and whoever grants access decides who gets the technology. Sacks reaches a similar destination from the other side, predicting a Democratic push for a federal department of AI that reviews frontier models, with the closed labs trading compliance for regulatory protection. Chamath dissents usefully: any rules that constrain data centers mostly empower the companies that own the power. Put together, the three views say the real AI policy fight of 2027 is about megawatts, not models.

The science corner is the counterweight to the gloom, and it deserves a careful reading. Friedberg says OpenAI published a 175-page paper showing that 3-SAT can be solved in subexponential time, which would disprove the Exponential Time Hypothesis proposed in 1999, and that the work took days or even hours. He is also clear that it has not been fully verified by third parties. His explanation runs a little ahead of the math: the hypothesis was never a proof that P does not equal NP, so disproving it removes a strong conjecture rather than settling the question. But his larger point stands. If an AI system really cracked a 27-year-old open problem, that is exactly the kind of benefit voters who see AI as an oligarch’s project are not hearing about, and it arrives the same week those voters elected governors promising to slow AI down.

Key Takeaways

Democrats took both chambers, with 54 to 56 Senate seats and at least 250 House seats expected, and flipped Texas, Kansas, Iowa and South Carolina races. Republicans lost ground with every voter group, including white men.
Sacks blames the Iran war and its economic fallout: inflation rose from 2.4% in January to 3.7%, expected rate cuts became a hike, and mortgages reached about 7.5%.
Young men swung 36 points against Republicans relative to 2024, and Sacks ties that to losing podcast hosts such as Tucker Carlson, Joe Rogan and Theo Von over the war.
Chamath says ideology no longer decides elections. A swing bloc punishes whoever holds power for an economy that rewards asset owners and fails wage earners.
Friedberg dates the problem to June 2025, when Elon Musk left and DOGE ended, leaving a $7 trillion budget with a $2 trillion annual deficit.
A Gallup survey shows 93% of Chinese citizens expect AI to help their country, against 36% of Americans, fourth from last.
Calacanis, citing Nate Silver, says Trump went from about +5 net approval to -28, with Liberation Day tariffs, ICE raids and the Epstein files hurting before the war.
Sacks’s fix is to end the war: Brent could fall from about $110 to $70, and history shows Clinton and Obama recovered from midterm losses while Bush’s surge after 2006 led to defeat in 2008.
Close to $4 trillion of market value disappeared in one day, the largest one-day dollar loss ever, and Polymarket odds of an Anthropic IPO this year fell from 95% to 71%.
Chamath expects a data center slowdown to favour hyperscalers with existing capacity and hurt neoclouds and frontier labs that rent compute.
Sacks expects permitted projects to finish in 2027 and few new ones to start, because no politician gains from approving a data center.
Calacanis says his firm spends about $50,000 a year on AI subscriptions for roughly $2 million in value, and predicts offshore data centers and local models on desktops.
Friedberg reports that OpenAI claims a subexponential-time algorithm for 3-SAT, disproving the Exponential Time Hypothesis, pending outside verification.

Chapters

3:04 The Results and Sacks on the War

Calacanis runs through the results: Democrats sweep Congress, win at least 29 governor races and flip deep red states including Texas. Sacks calls it the worst midterm since Watergate and blames the Iran war and its effect on gas, diesel, food and fertilizer prices. He admits he expected a “red spritz” because jobs and GDP were strong. JD Vance’s post, he says, got the cause right.

11:10 Friedberg: The End of DOGE and the Oligarch Narrative

Friedberg replays his 2024 prediction of a rise in socialist movements and dates the turn to June 5, 2025, when DOGE ended. Government spending, he argues, drives the cost of living, and voters who are not getting ahead conclude someone else is: the AI companies. A Gallup chart puts Americans near the bottom of the world on AI optimism.

18:19 Chamath’s Blind Spot: A Post-Ideology Electorate

Chamath says he got it totally wrong because he saw the economy through the eyes of asset owners. Campaign money and gerrymandering both failed, and centrists talking only about affordability won. His new framework is economic populism: close loopholes, end federal loan underwriting, and cut federal taxes to zero below about $150,000. Until that happens, he expects voters to keep swinging violently between parties.

25:29 Calacanis: The Polls Were Right

Calacanis plays his own clip predicting the sweep and argues the polls and prediction markets were right all along. Using Nate Silver’s approval chart, he lists Liberation Day tariffs, masked ICE raids, the Epstein files and grifting allegations as the slide before the war. He calls Trump’s ideas right and his execution chaotic.

32:36 Can Trump Salvage His Presidency?

Sacks shows approval mapping to gas prices and the Senate market breaking only after February 28, 2026. His three fixes are end the war, end the war and end the war. Chamath counters that gas, like the price of eggs in 2024, is a window into a broader broken economy. Sacks describes how Gina Hinojosa won Texas on economic populism and tied data centers to higher utility bills.

40:44 The Ring of Sauron, Pardons and a Deal

Friedberg expects budget gridlock and predicts the government will try to control AI through compute, treating data centers as utilities. Calacanis proposes blanket pardons followed by a deal on taxing margin loans, healthcare, a $15 minimum wage, a ban on stock trading in office and housing. Chamath argues that history may judge Trump on the closed border and reshored manufacturing, as with Reagan. Sacks cites Clinton in 1994, Obama in 2010 and Bush in 2006 as the choice between correcting course and doubling down.

52:50 The $4 Trillion Sell-Off and Hyperscaler Kingmakers

Nvidia, Tesla, AMD, Micron and SpaceX fell together in the biggest one-day dollar loss in market history, and Anthropic’s IPO odds slipped. Chamath describes SpaceX staff facing a hostile crowd in Cochise County, Arizona, over a 6 gigawatt data center. He expects incoming governors to end tax breaks, slow the build-out and hand leverage to the hyperscalers, with the SpaceX and Anthropic 90-day out as the deal to watch.

61:54 Could AI Be Banned?

Friedberg says calls to ban AI development and use are growing on the left and puts the odds well above 10%, framing control of AI as a Second Amendment-style question about state power. Calacanis puts it near 1%. Sacks says a major Democratic push on AI legislation is certain and predicts a federal department of AI and a regulatory deal with the closed labs. Chamath says data center limits will mostly entrench the companies that own power.

72:01 Agents, Offshore Compute and the Permit Problem

Calacanis describes his firm’s AI spend and returns and predicts offshore data centers plus local models on high-end Mac Studios. Sacks separates federal AI safety fights from state and local infrastructure fights and warns that “no new permits” could strand half-built projects. He expects meaningful capex in 2027 and a much thinner pipeline in 2028. Chamath and Calacanis spar over whether SpaceX would reprice Anthropic’s compute.

84:07 Science Corner: The Exponential Time Hypothesis

Friedberg explains P and NP problems with a Sudoku example and describes the 1999 Exponential Time Hypothesis about 3-SAT. OpenAI’s 175-page paper claims a subexponential algorithm, produced in days or possibly hours, and is still being checked. Sacks closes on AI’s benefits and the risk of importing tokens if the US will not build.

Notable Quotes

“This was a blowout. It was, I believe, the biggest midterm loss since the Watergate midterm loss.”
David Sacks, opening the midterm postmortem

“I do not think Democrats or Republicans matter. That is my real takeaway after this election.”
Chamath Palihapitiya, on the swing bloc deciding elections

“There is a small percentage of people, the four of us on this pod and all of our friends, who are the asset owners of America. And we have benefited in every weather condition.”
Chamath Palihapitiya, on why he misread the electorate

“AI is almost like the ring of Sauron. Okay. And like both sides want the ring of Sauron.”
David Friedberg, on why calls to stop AI are really calls to control it

“So I think all the game theory, Jason, literally comes down to this one deal and whether Elon decides to rip it because if he rips it, watch out.”
Chamath Palihapitiya, on the SpaceX and Anthropic compute contract

“In this last election, you had both Democrats and Republicans running campaign attack ads on any politician who approved any data center.”
David Sacks, on why new data center permits may dry up

“This is a spreadsheet and the spreadsheet will dictate the answer.”
Chamath Palihapitiya, on whether relationships can override compute pricing

Watch the full All-In episode here.