Featuring
Spice rack
Will safety fears and local opposition stop the United States from building small nuclear reactors?
Original point: Sacks predicts that SMRs will not be built because communities will reject even a small nuclear plant near them, and he argues that an accident is eventually inevitable.
What everyone argued
Chamath Palihapitiya
Chamath warns that the corporate announcements vary sharply in substance and says many advanced-reactor promises remain pre-commercial. He also rejects Friedberg's claim that nuclear is the only plausible answer to future electricity demand, pointing to solar, storage, grid reform, materials science, and continued efficiency gains.
Jason Calacanis
Jason argues that the United States has enough low-density land to site reactors far from cities and that modern designs can deliver abundant electricity without exposing dense communities to old-reactor risks. He keeps pressing for safety evidence and eventually finds common ground with Sacks around remote sites whose host communities consent.
David Sacks
Sacks says pro-nuclear enthusiasm is a luxury belief because the risks will fall on poorer host communities, predicts that local opposition will stop projects, and refuses to accept that new designs eliminate accident risk. He later concedes that he would not oppose a project in a willing community or a remote area without nearby residents.
David Friedberg
Friedberg argues that rising electricity demand makes nuclear economically necessary and that safety fear has blocked a mature, unusually low-fatality technology. He challenges Sacks to use data, points to operating SMRs abroad, and says U.S. competitiveness will suffer if China deploys abundant firm power while America refuses it.
Winner circle
Friedberg and Jason carry the core question. The existing international safety and operating record, followed by U.S. permits and construction, defeats Sacks's absolute claim that local fear will stop SMRs altogether. Chamath earns an important technical caveat: the specific U.S. advanced designs still have to prove operation and commercial economics. The pro-build side wins, but only with host consent, design-specific evidence, and honest accounting of accident consequences.
Commentary
Chamath Palihapitiya
Assumptions and fact checks
Solar, storage, grid reform, and new materials will provide enough capacity before electricity becomes a binding constraint on AI and industrial growth.
Why it mattersThose resources can contribute substantially, but the timing, transmission buildout, firm-capacity needs, and growth in demand remain uncertain. Chamath identifies alternatives without quantifying whether they meet the same reliability and scale requirements.
The announced nuclear deals did not involve money actually changing hands.
CheckThat description fits some conditional offtake or exploration arrangements, but not the whole group Jason introduced: Amazon led an approximately $500 million equity financing in X-energy. Google's Kairos agreement was a future power-purchase commitment tied to reactors planned for 2030 and later.
Kairos's proposed next-generation commercial reactors were not yet functioning power plants.
CheckGoogle described successive demonstrations before a first commercial unit targeted for 2030. The NRC permits cover low-power Hermes 2 test reactors, not an operating commercial fleet.
Jason Calacanis
Jason improves his argument by searching for a consent-based compromise rather than dismissing public fear. He weakens it when distance figures become a substitute for a full safety and grid analysis.
Assumptions and fact checks
Remote siting and long transmission lines can resolve most public-safety and consent objections without undermining project economics.
Why it mattersRemote siting can reduce population exposure and conflict, but long transmission, workforce access, water, emergency planning, tribal and local consultation, and cost still require site-specific proof.
Amazon announced a $500 million investment connected to advanced small modular reactors.
CheckAmazon says it led an approximately $500 million financing round in X-energy to support SMR technology and manufacturing capacity.
Modern advanced-reactor projects were moving through U.S. approval rather than being categorically blocked.
CheckThe NRC issued construction permits for two Kairos Hermes 2 test reactors in November 2024, and DOE documented active Hermes construction at Oak Ridge in April 2026.
David Sacks
Sacks identifies the strongest political obstacle—host-community consent—but initially frames residents as passive victims and substitutes fear for comparative risk analysis. His later acceptance of remote or willing sites retreats from the absolute opening prediction and becomes the most defensible part of his case.
Assumptions and fact checks
A serious accident is inevitable if enough SMRs are deployed.
Why it mattersNo technology has zero risk, but inevitability requires evidence about design-specific failure probabilities and consequences. Sacks offers neither and does not compare nuclear risk with fossil generation, grid scarcity, or other alternatives.
Poorer communities will bear reactor risks without meaningfully choosing or benefiting from the projects.
Why it mattersEnvironmental-justice and bargaining-power concerns are real, but the outcome depends on consent, ownership, tax revenue, jobs, regulation, emergency planning, and the site. It cannot be inferred from community income alone.
David Friedberg
Friedberg wins the comparative-risk argument by demanding evidence and connecting reliability to economic tradeoffs. The casualty and deployment statistics should have been stated with source definitions and uncertainty; overstating clean facts gives opponents an easy reason to distrust an otherwise strong case.
Assumptions and fact checks
Nuclear is necessary because renewables, geothermal, storage, and transmission cannot scale fast enough to meet industrial and AI demand.
Why it mattersFirm clean power is valuable and demand is rising, but necessity depends on future load, efficiency, storage, transmission, gas with carbon controls, geothermal, and build times. Friedberg makes the constraint concrete but does not prove exclusivity.
Commercial SMRs were already operating in China and Russia.
CheckThe IAEA reported two designs in commercial operation in 2024: Russia's Akademik Lomonosov and China's HTR-PM.
SMRs were operating in India and about 65 were being built.
CheckThe IAEA's 2024 status material identified commercial operation in Russia and China, not India, and described only a small number of SMR plants under construction. Friedberg appears to have mixed proposed designs or conventional reactor plans with operating and under-construction SMRs.
The Three Mile Island accident caused zero deaths.
CheckDOE says the accident caused no injuries, deaths, or direct health effects and exposed nearby residents to a very small additional dose.
Fukushima caused one death.
CheckThat single number is misleading. WHO reports no acute radiation injuries or deaths from radiation exposure, while evacuation and relocation caused broader mortality and health harms that cannot be summarized as one death.
Chernobyl caused 46 deaths.
CheckWHO reports 28 acute-radiation deaths among workers in the first three months and more than 6,000 thyroid-cancer cases among those exposed young, with 15 deaths reported by 2011; long-run modeled excess cancers and fatalities are much larger but uncertain. Forty-six is not a defensible complete total.
Was J.D. Vance's 2020 certification position a legitimate concern or a media distraction?
Original point: Jason asks whether Vance's statement that he would not have certified the 2020 result should concern voters.
What everyone argued
Jason Calacanis
Jason repeatedly asks Sacks and Friedberg to answer the narrow question: should voters worry that Vance would have handled the 2020 electoral count differently from Pence? He resists Sacks's attempt to redirect the exchange to a separate Vance interview about immigration.
David Sacks
Sacks says Jason is fixated on a topic no persuadable voter without 'TDS' cares about and redirects attention to Vance's effective answer to Martha Raddatz about alleged gang activity in Aurora. His position is that media combat and current voter concerns matter more than revisiting certification.
David Friedberg
Friedberg says Vance is trying to broaden the question from one sentence about who won into concerns about social-media moderation, election administration, voter verification, and public trust. He argues those systems deserve transparent scrutiny even while declining Jason's requested yes-or-no judgment.
Winner circle
Jason wins. His factual premise was accurate, the office's role made the question important, and neither opposing speaker addressed the legal or evidentiary burden for alternative slates. Sacks was partly right about electoral salience, but he answered 'will voters care?' when the debate asked 'should this concern them?' Friedberg offered valuable context without closing that gap.
Commentary
Jason Calacanis
Jason does the essential debate work: he notices that Sacks is answering a different question and refuses the substitution. A one-sentence explanation of the vice president's ministerial role would have made the exchange harder to evade.
Assumptions and fact checks
A vice-presidential candidate's view of the electoral count is important enough to press even if polling suggests voters prioritize other issues.
Why it mattersThe vice president presides over the count, and the peaceful transfer of power is a core constitutional function. Popular salience and institutional importance are different questions.
J.D. Vance said he would not have simply certified the 2020 result and instead would have sought alternative slates from states.
CheckVance publicly said he would have asked states to submit multiple or alternative slates rather than proceeding as Pence did. He also repeatedly refused to say Trump lost the 2020 election during the 2024 campaign.
Joe Biden won the certified 2020 Electoral College vote.
CheckThe National Archives records a 306-232 Biden victory; objections to Arizona and Pennsylvania were not sustained, so their certified votes were counted.
David Sacks
Sacks performs the exact substitution the analysis rules should penalize: he attacks the salience and messenger rather than the strongest version of the institutional concern. His electoral prediction aged better than his argument.
Assumptions and fact checks
If few persuadable voters prioritize a candidate's position, the position is not a serious concern.
Why it mattersElectoral popularity does not determine constitutional validity or institutional risk. It may predict campaign impact, but that is an adjacent claim.
A strong answer in an immigration interview offsets the need to answer questions about election certification.
Why it mattersThe two issues have different facts, powers, and burdens. Performance on one does not resolve the other.
David Friedberg
Friedberg gives a serious account of why election trust can erode, which is more useful than Sacks's dismissal. He still bridges from broad process concerns to the certification question without supplying the missing evidence or legal mechanism.
Assumptions and fact checks
Concerns about moderation, voting rules, or verification can justify refusing a clean answer about the certified winner.
Why it mattersThose concerns can support prospective reforms or specific legal challenges. They do not erase the obligation to distinguish verified defects from generalized distrust after states certify results.
The vice president has discretion to demand alternative slates during the electoral count when broader election-system concerns remain.
CheckCRS describes the presiding officer's duties as largely ministerial and says the vice president has no unilateral authority over which certificates are presented. Objections are handled by the House and Senate under federal law.
Should Uber have bought Expedia to become a travel super-app?
Original point: Chamath calls the proposed acquisition stupid because Expedia's booking interface and third-party data could be disintermediated by AI agents, making a large control premium reckless.
What everyone argued
Chamath Palihapitiya
Chamath says Uber should not spend about $30 billion on an interface layered over widely licensed travel inventory when AI agents may route customers directly to suppliers. He argues Uber can test travel or buy strategically closer assets for far less capital.
Jason Calacanis
Jason argues that hotels and Vrbo fit Uber users' travel journey, payment credentials, membership, and existing ride behavior better than flights do. He favors a hotel or travel tab and emphasizes proprietary lodging inventory and cross-promotion.
David Sacks
Sacks argues that ride hailing is an immediate, low-friction task while vacation booking is a different state of mind. He doubts cross-selling can support the cost cuts in Friedberg's model, though he later accepts that a separate Uber Travel product or rebranded asset might have value.
David Friedberg
Friedberg builds the acquisition steelman: Dara knows Expedia, Uber can cross-sell to its installed base, overlap could reduce G&A, and lower customer-acquisition spending could lift Expedia from roughly $3 billion to as much as $6 billion of EBITDA. He treats that as an explanation of the board's possible rationale, not a confident recommendation.
Winner circle
Chamath and Sacks win the acquisition question. Chamath correctly separated useful inventory from an expensive, AI-exposed corporate shell, while Sacks exposed the unproven marketing-synergy assumption. Jason and Friedberg deserve substantial credit because the product they imagined effectively arrived in 2026. The elegant outcome was partnership: Uber got the travel tab, Expedia kept the booking machinery, and nobody had to pretend a $30 billion purchase was the only route.
Commentary
Chamath Palihapitiya
Chamath best separates the strategic feature Uber wants from the company Uber would have to buy. His 'few million dollars' recreation claim is too glib, but the capital-allocation conclusion aged extremely well.
Assumptions and fact checks
AI agents will commoditize consumer travel-booking interfaces faster than they erode the value of inventory, service, and supplier relationships.
Why it mattersThe interface is becoming more contestable, but reliable booking still needs contracted supply, live availability, payments, changes, refunds, and support. The 2026 API partnership supports modular access more clearly than total commoditization.
Jason Calacanis
Jason sees the product before the market ships it. The missing move is to separate 'put Expedia in Uber' from 'put Expedia on Uber's balance sheet.'
Assumptions and fact checks
Uber's travel context and stored payment credentials will generate meaningful lodging conversion.
Why it mattersAirport and out-of-town ride behavior creates relevant intent, and Uber chose to launch precisely this product. Conversion economics remain undisclosed, so the size of the benefit is still unproven.
Uber had roughly 150 million monthly active customers at the time.
CheckUber reported 161 million monthly active platform consumers in Q3 2024, so the rounded figure was conservative.
Hotels and Vrbo inventory could be integrated directly into the Uber app.
CheckUber's April 2026 Expedia partnership launched hotel booking in Uber with access planned to more than 700,000 properties and Vrbo vacation rentals later in the year.
David Sacks
Sacks correctly attacks the synergy rather than the spreadsheet arithmetic and forces a product-level explanation. His strongest contribution is that lower marketing spend cannot be assumed merely because two apps share a broad category.
Assumptions and fact checks
Users will reject lodging offers because opening Uber usually signals an immediate ride need.
Why it mattersThat is true for the core ride flow but not for every surface or moment. A distinct travel tab and pre-trip context can preserve task focus, which is how Uber ultimately implemented the product.
David Friedberg
Friedberg offers the best steelman and clearly labels it as such. The model would be decision-grade only after cohort tests showed incremental hotel conversion and how much paid acquisition could actually disappear.
Assumptions and fact checks
Uber could remove about 30% of Expedia's marketing and half its G&A without materially reducing bookings or damaging the travel operation.
Why it mattersSome overlap and cross-sell are plausible, but Expedia's marketing buys demand and its corporate functions support a distinct regulated, global travel business. The size and losslessness of the cuts needed direct experiment evidence.
Expedia was spending about $8 billion annually on sales and marketing and producing about $3 billion of adjusted EBITDA.
CheckExpedia's 2024 filing reports $6.846 billion of direct and $781 million of indirect selling and marketing expense, plus $2.934 billion of adjusted EBITDA. Friedberg's rounded figures were close.
Expedia's annual G&A expense was about $720 million.
CheckThe 2024 filing reports $805 million of G&A. A trailing or annualized figure available during Q3 may have been lower, but the episode's number understates the completed-year expense.
Chamath supplies the debate's best commercial-readiness check. His case would be cleaner if he had begun with the distinction he eventually reached: a few SMR designs operate, but the particular U.S. advanced designs in these announcements still need demonstration, licenses, supply chains, and cost proof.