Midterm Blood Bath, $4T AI Sell Off, Data Center Freeze, What’s Next for Trump
Topics
Macro Asset Allocation in High-Interest Regimes
Trillion-Dollar AI Capex and Bubble Risks
GPU Depreciation and Cloud Market Shakeup
Frontier AI Regulation and Existential Risk
AI Inference Cost Optimization
AI Advances in Mathematics and Formal Verification
Pioneers
David Sacks
Summary
- The hosts frame the Democratic sweep as an affordability revolt, but disagree over whether ending the Iran war can reverse it. Sacks links inflation’s rise from 2.4% to 3.7% to fuel and transportation costs, arguing that ending the war could take Brent from as high as $117 toward $70 and reopen the path to rate cuts. Calacanis calls getting out of the war “my number one, my number two, and my number three”; Chamath and Calacanis see a broader economic failure.
- Chamath abandons his belief that asset-led growth is reaching wage earners and makes economic fairness the 2028 electoral test. Asset owners prospered through cheap capital, higher rates, tariffs and subsidies, while paycheck earners remained squeezed: “I have to update my priors.” His prediction is continued electoral reversals against incumbents unless either party delivers meaningful relief.
- Friedberg sees government spending as the structural affordability problem that Washington refused to confront. He points to DOGE’s collapse, a $7 trillion annual budget and $2 trillion deficits, arguing that spending inflates assets while eroding workers’ purchasing power. “The more the government spends, the more expensive things get”; he worries Congress and the sitting president will not reach a clean budget agreement, leaving living costs at risk in 2027–28.
- The reported $4 trillion AI sell-off turns data-center permitting into a major risk for infrastructure spending and IPO expectations. Calacanis says OpenAI’s Polymarket odds of going public fell from 95% to 71%, alongside valuation expectations moving toward roughly $1 trillion. Sacks’s base case preserves meaningful construction spending in 2027, but he worries the 2028 pipeline could close or shrink under “conditions of great uncertainty.”
- Chamath argues that constrained compute could strengthen infrastructure owners while trapping frontier-model companies and neoclouds. Existing hyperscalers could reserve capacity for themselves or charge renters more; he singles out the SpaceX–Anthropic agreement’s mutual 90-day termination right as a potential repricing trigger. Calacanis suggests a negotiated midpoint, Chamath points to Elon’s relationship with Anthropic, and Friedberg says, “This is a spreadsheet and the spreadsheet will dictate the answer.”
- The AI-policy debate divides over how much control government will seek and who regulation would ultimately empower. Sacks assigns a major Democratic legislative push a 100% probability and predicts a bargain exchanging regulatory protection for government power. Friedberg warns about compute becoming a regulated utility; Calacanis puts an extreme state-control scenario around 1%, while Chamath argues supply restrictions would mainly entrench hyperscalers.
- Calacanis predicts a two-year shift toward offshore data centers and local models. He estimates $40,000–$50,000 in annual AI subscriptions creates roughly $2 million of value at his firm, and sees adoption plus agents driving vastly greater demand. He is considering $400,000 of desktop hardware for local-first inference: “We want to have unmetered intelligence.”
- Friedberg’s scientific counterweight is an OpenAI complexity-proof claim that he acknowledges remains unverified. He describes a 175-page paper challenging the exponential time hypothesis, completed in less than 10 days, with some reports claiming hours. He presents it as evidence of AI’s research potential while explicitly saying it is “not fully verified by third parties yet.”
Deep dive
1. The Republican defeat becomes an argument about affordability
- Calacanis reports Democrats taking both chambers and winning at least 29 governor races: 54–56 Senate seats and 250–260 House seats, with Alaska and Mississippi unresolved. Texas, Kansas, Iowa and South Carolina flip after Trump won them by double digits two years earlier. Sacks calls it the biggest midterm defeat since Watergate: “There’s no sugarcoating it.”
- Sacks’s chain runs from war through fuel, transport and fertilizer to lost purchasing power. Gas moves from around $3 to above $4, costing households, he says, $200 more monthly; January’s 2.4% inflation becomes 3.7%, rate-cut hopes become a hike, and mortgages reach roughly 7.5%. His prewar 5% GDP-growth expectation draws “Aggressive. Unbelievable. Unrealistic” from Calacanis. The failure is “this war and the economic consequences that follow from the war.”
- His admission—“I was wrong”—concerns the defeat’s severity. He cites 4.1% unemployment, first-half growth around 2.1%–2.5% and third-quarter GDPNow tracking at 3.7%. He expected something resembling Biden’s inflation-heavy “red spritz” midterms. Instead, inflation eroded the first Trump year’s roughly 3.1% gain in real after-tax household income; voters judged their own purchasing power.
- War also fractures the coalition, Sacks says, costing Republicans Carlson, Rogan, Theo Von and Megyn Kelly: “they are our media.” Sacks cites a 36-point young-male swing; he also cites 34% calling Iran crucial and 40% a major voting factor. JD Vance’s post called for a reset centered on ending the war, tackling living costs, putting America first and changing the relationship with Israel.
2. Friedberg traces the backlash to an abandoned spending overhaul
- Calacanis revisits his post-2024 prediction of rising socialist movements despite deregulation and AI-driven growth: some companies and communities would win, others lose. Friedberg says DOGE promised government efficiency and positive returns on spending. When Musk left on June 5, 2025, Friedberg considered that effort finished and wrote, “China just won.”
- Friedberg’s March 2025 Washington visit supplies the disappointment: Musk was struggling, while members of both parties showed little interest in spending efficiency or government’s share of GDP. July’s One Big Beautiful Bill reduced some entitlement costs without changing fundamental spending. Friedberg’s tally remains a $7 trillion annual budget and $2 trillion deficit; he says Congress was never on board with the necessary overhaul.
- His causal claim is categorical: “The more the government spends, the more expensive things get.” He links COVID’s doubling of spending to asset inflation rewarding owners while squeezing paycheck earners. With this Congress and the sitting president, he worries a clean 2027–28 budget agreement will be difficult; continued $2 trillion deficits could worsen costs. Watch the bond market, he says. Voters seeking relief can produce socialist policies without choosing socialism ideologically.
- AI becomes the beneficiary people resent when growth bypasses them. Friedberg cites Gallup figures of 93% Chinese optimism about national benefits and 93% about daily-life improvements, against 36% American optimism, fourth from bottom—below Afghanistan and slightly above Malawi and Bangladesh. He allows skepticism about Chinese polling. His point is receptivity to an “oligarch narrative”; whether that backlash persists remains uncertain.
3. Chamath rejects the assumption that asset gains reach wage earners
- Chamath says, “I got it totally wrong,” identifying a blind spot rooted in his own interests. Asset owners benefited from cheap capital at zero rates and pricing power at high rates, across tariffs and subsidies; wage earners lacked that resilience. “I thought trickle-down economics could still work.” His update: “It’s clearly not happening. It’s not true. I have to update my priors.”
- His evidence against an ideological reading is failed spending and gerrymandering. Money poured into Texas and Ohio without delivering the intended wins; of 15 Republican-redrawn seats, he says Democrats won 11 and Republicans only one. He sees centrist affordability candidates prevailing, explicitly disputes Friedberg’s socialism framing and says Iran, Israel and AI were not the central explanation.
- Since COVID, Chamath sees a swing bloc granting mandates to Biden, Trump and then Democrats, demanding economic repairs. His 2028 call is economic populism built on “reasonable standards of economic fairness,” from either party. Incumbents who do nothing will keep losing. Democrats must now legislate that fairness or concentrate on impeaching Trump and risk reversal if Republicans supply the economic agenda.
4. Calacanis challenges the war-only explanation with an earlier decline
- Calacanis points to Trump’s net approval falling from roughly +5% to −28%, with deterioration predating Iran. He says polls and prediction markets were right. Liberation Day’s tariffs raised costs without balancing the budget; he recalls speaking with the Trump-supporting CEO of Cuts Clothing during a discussion of the tariffs. Trump should have worked through Congress rather than pursuing the route that lost at the Supreme Court.
- Summer 2025 brings Calacanis’s next breaks: ICE enforcement by masked agents without visible badges or body cameras, and Trump reneging on sending out the Epstein files, he argues. Those episodes alienated podcast audiences before the war; grifting allegations and inflation deepened the damage. “Trump didn’t make people feel like he was working for them.” He credits the ideas while criticizing chaotic execution.
- Sacks counters with Senate-control markets: Republican odds remained strong through 2025, then deteriorated after the war began on February 28, 2026. Brent’s 10% election-day fall, without a change in the Strait of Hormuz, illustrates his recovery mechanism. If the war ends, he expects crude—which has been as high as $117—to fall toward $70. Lower inflation could permit rate cuts and restore Trump’s approval into the 40s.
- Calacanis uses 2024’s egg-price controversy as an example: eggs were “a window into a broad problem,” and gasoline can serve the same purpose. Fixing one visible price may miss structural hardship. Sacks agrees affordability is central but sees healthy employment and growth as grounds for recovery without restructuring the economy. If the war continues, however, he expects Republicans to keep bleeding support.
5. Trump’s proposed reset combines economic bargains with an end to war
- Calacanis proposes a controversial reset: pardon Trump’s family, Trump himself and administration officials immediately, then tell Democrats, “Let’s work. Let’s make a deal.” He wants equity-backed loans above $10 million taxed at 50% of the capital-gains rate, credit for repayment and the balance collected when shares sell. The aim is revenue and an answer to claims that wealthy borrowers pay no tax.
- His remaining proposals cover healthcare, a phased $15 minimum wage over four years, banning stock trading in office and business with foreign countries, and incentivizing housing construction. He estimates 10 million–15 million workers earn below $15, making the wage measure manageable and a signal that the bottom half matters. Solving two or three of those five issues, alongside ending the war, could redeem the presidency.
- Chamath first suggests zero federal taxes below “probably $150,000,” later proposing $125,000 or under; he calls the revenue loss “a nothing burger” and would close loopholes. He pairs ending federal underwriting of loans with getting rid of student debt. Preserve border closure through an economic bargain, he argues. Tariffs remain conditional: if promised investment materializes and revives domestic manufacturing, their historical record could improve as gains compound.
- Sacks contrasts course correction with escalation. Clinton lost the House in 1994 and won in 1996; Obama lost seven Senate and 63 House seats in 2010 and won in 2012. Bush lost in 2006 over Iraq, chose the surge, and Republicans lost again in 2008. Sacks sees renewed escalation repeating that failure; Calacanis calls ending the war “my number one, my number two, and my number three.”
6. The AI sell-off exposes a dispute over local benefits
- Calacanis describes nearly $4 trillion erased across Nvidia, SpaceX, Tesla, AMD and Micron: a dollar record, explicitly not a percentage record, followed by a partial rebound. Polymarket’s OpenAI IPO odds fall from 95% to 71%; he relays Wall Street Journal valuation expectations near $1 trillion versus $1.5 trillion–$2 trillion and earlier $3 trillion–$4 trillion talk. Anthropic’s IPO faces a darker backdrop. He has been buying SpaceX and Tesla “through this whole insanity.”
- Sacks’s Texas example connects electoral economics to infrastructure: the incoming governor rejects culture-war questions and links data centers to utility bills and water use. Calacanis adds that the oligarchs are getting tax breaks while residents get higher bills. Chamath concedes the bargaining problem: should seven or eight powerful companies need tax relief for $30 billion–$50 billion investments generating $345 billion–$500 billion? Incoming governors may demand directly measurable benefits before agreeing.
- Chamath’s illustration is SpaceX’s proposed 6-gigawatt Arizona facility. Presenters Riley and Katie promise electricity costs down 40%, property taxes down 20%, no water use and revenue to retire a county bond, yet the audience shouts over them. He calls it “an emotional issue that is not tethered to the facts”: economic division that technical and fiscal promises alone fail to defuse.
7. Scarce compute could transfer leverage from model companies to owners
- Chamath’s call favors hyperscalers with existing capacity—Amazon, Meta, Microsoft, Google and SpaceX—over neocloud renters and compute-starved frontier labs. Anthropic and OpenAI improve on 60–90-day loops, but lack the combined fortress balance sheets, profitability and quasi-infinite compute that would make them “home free.” His warning is that the data-center renter class is “very much at risk.”
- Owners can keep renting or reserve incremental capacity for themselves. Google could prioritize Gemini while rivals face higher prices, deteriorating user experience and slower iteration. Chamath points to Google taking over a renter’s data center; Friedberg says the hyperscalers have critical ingredients, specifically power. Calacanis adds that concentrating compute on training could determine who produces the best next model.
- Chamath watches the SpaceX–Anthropic agreement’s mutual 90-day termination right. If SpaceX exits and reprices scarce capacity, he argues, that revenue could rise two or three times and Anthropic would have little choice but to pay. Other hyperscalers would reassess yields; neoclouds could spend remaining debt and equity securing whatever compute they can, because being shut out risks bankruptcy.
- Calacanis suggests Musk could negotiate a middle ground. Chamath points to Elon’s close working relationship with Anthropic; Sacks cites acrimony between Elon and Sam Altman and argues that relationships will inform the decision. Friedberg says, “This is a spreadsheet, and the spreadsheet will dictate the answer.” Sacks’s point is that if constraints push GPU-hour and megawatt-month prices sufficiently high, even long-term offtake can become negative expected value. Capacity could go to Google or Meta; the argument does not require selling it to OpenAI.
8. AI regulation divides the hosts over state power and practical incentives
- Friedberg calls AI “the Ring of Sauron”: both political sides want its power. Even a million data centers running open-source models could weaken centralized authority without an oligarch. His prediction, allowing “I could be wrong,” is regulation of compute and treatment of data centers like utilities. Chips and software are harder to police; compute access offers a practical control point.
- Sacks frames a model as code and software, making regulation a free-speech question. He argues that existing laws cover harmful uses, such as theft and illegal weapons. His Second Amendment analogy concerns unequal capability: a state with more powerful AI gains power without the same check. He says the concern is that the state might have access to capabilities that citizens and businesses do not.
- Sacks gives a major Democratic AI-legislation push a 100% probability. His forecast for its shape is a federal AI department preapproving frontier models, with misinformation rules becoming censorship. The quid pro quo gives government more power and the closed-model duopoly regulatory protection. He predicts surveillance and censorship enhanced by superintelligence; this is his interpretation of regulation’s destination.
- Calacanis first floats 10% as a handicap for an extreme scenario, then says the chance of something like it is about 1%. Friedberg says the concern is not binary, but gives no probability. Chamath cites California legislation vetoed by Gavin Newsom and a patchwork of state laws. Chamath disputes the priority and political logic of nationalization: entrenching an oligopoly contradicts voters’ economic demands. Infrastructure restrictions, he argues, ultimately empower hyperscalers, because model companies cannot obtain protection from shortages of electricity and money merely through regulatory capture.
9. Permitting uncertainty threatens the construction pipeline after 2027
- Sacks separates federal AI-safety rules from state infrastructure permits. Texas’s incoming governor has said no new permits will be issued before legislation, Sacks says, without unequivocally protecting existing construction. A partly built facility may still need air-pollution or other late approvals. He allows reasonable or unreasonable implementation: “I’m not saying that’s necessarily going to happen,” but stranded half-finished projects are possible.
- His base case preserves meaningful capex in 2027 as fully permitted projects finish; by 2028, the pipeline could close or shrink substantially. Both parties attacked politicians for approving data centers, leaving ambitious officials little incentive to approve more over the next four years. Gigawatt-scale projects costing around $50 billion require certainty before developers risk injunctions, labor action or community opposition halfway through construction.
- Sacks says a slowdown need not be a conspiracy, but believes Democrats could cause the damage and blame Trump. Calacanis sees an opposing incentive: tanking growth creates an electoral vulnerability, whereas negotiating electricity, schools and local benefits rewards constituents. Their disagreement is whether governors chiefly want better terms or permitting politics itself blocks capacity.
10. Calacanis expects demand to escape through offshore and local compute
- At his 20-person venture firm, Calacanis estimates annual AI subscriptions of $40,000–$50,000, mainly Grok and Claude, create roughly $2 million in value and about $1 million in incremental revenue. Automating investment workflows and media production increases the firm’s operating velocity.
- He estimates only 10%–20% of corporate workers use AI daily, leaving five to ten times more potential users. Agents abstract away coding and setup while using 50 times as much compute. His resulting demand call is 500–1,000 times today’s need.
- His firm is considering $20,000 Mac Studios with 512 GB, totaling $400,000 for 20 people, running local models before falling back to frontier services. He says their current use is only 70%–80% of what frontier models can do. The objective: “We want to have unmetered intelligence.”
- His two-year prediction is offshore data centers plus desktop models. Canada, Mexico, Singapore and Malaysia can host capacity; ending the war could reopen Saudi Arabia, the UAE and Qatar. Fiber and Starlink support serving users across borders, he argues. America could become a net token importer, losing local tax revenue and construction jobs he describes as paying $300,000–$400,000.
11. Friedberg’s scientific optimism rests on a proof still under review
- Friedberg starts with solving a problem quickly versus checking its answer quickly. Sudoku is his example: finding the solution can be hard while checking it is easy. He uses that distinction to introduce P versus NP and the exponential time hypothesis, proposed in 1999 for 3-SAT.
- He says a 175-page OpenAI paper provides a subexponential-time approach and “disproved” that hypothesis. The work took less than 10 days end to end, with some reports claiming hours, against a 27-year-old question. He qualifies the claim: “It’s not fully verified by third parties yet.” Others are still evaluating its novel approach.
- Friedberg interprets the result as keeping P = NP possible and, in his framing, the prospect that every problem could be solved by computer. He calls it “one of these moments of humility” for mathematicians. Sacks returns to his broader wager: AI could deliver enormous benefits, and China will pursue them if America does not.