Jason Calacanis
B4 claims reviewed
The election recap and tariff argument hold up. The pardon proposal and ease of moving compute need qualification.
Evidence balance 7.5/10AI analysis of claims on the All-In Podcast
The affordability backlash, strong jobs report and risk of data-center delays have solid support. The larger errors concern household gas costs, the price of tax cuts and the meaning of the reported computing breakthrough. Several political arguments leave out evidence that complicates the story.
Episode: Midterm Blood Bath, $4T AI Sell Off, Data Center Freeze, What's Next for Trump
4 claims reviewed
The election recap and tariff argument hold up. The pardon proposal and ease of moving compute need qualification.
Evidence balance 7.5/106 claims reviewed
Gets the jobs report right and raises credible oil and permitting risks. The gas estimate and selective political framing reduce the score.
Evidence balance 5.5/105 claims reviewed
Reports the AI survey and research announcement accurately. The spending explanation and claim about universal computation go too far.
Evidence balance 5.5/105 claims reviewed
The affordability and compute-scarcity arguments are credible. The tax-cost claim and interpretation of the Crusoe sale are weaker.
Evidence balance 6/10| Speaker | Overall | Accuracy 30% | Coherence 20% | Evidence support 20% | Calibration 15% | Evidence balance 15% | Claims |
|---|---|---|---|---|---|---|---|
| Jason Calacanis | 75 / B74.75 before rounding | 7.5 | 7 | 7.5 | 8 | 7.5 | 4 |
| David Sacks | 65 / C66 before rounding | 6.5 | 6.5 | 7 | 7.5 | 5.5 | 6 |
| David Friedberg | 60 / C60.75 before rounding | 6 | 5.5 | 6.5 | 7 | 5.5 | 5 |
| Chamath Palihapitiya | 65 / C66 before rounding | 6.5 | 6.5 | 7 | 7 | 6 | 5 |
Credit. Acknowledges several causes of the election result and treats offshore expansion as a forecast.
Deduction. Overstates how much pardons could sideline investigations and how easily large facilities can relocate.
Considers both the war and earlier dissatisfaction, and acknowledges local governments’ incentives. No concrete selective presentation was found in these four claims.
Credit. Accepts the war’s contribution while adding evidence of earlier price pressure. Did tariffs raise prices before the war?
Credit. Acknowledges that governors may seek better terms for communities before approving projects. Could US restrictions push data centers overseas?
No concrete selective presentation was found in the reviewed sample.
Credit. Admits the election miss and clearly qualifies the risk to unfinished projects.
Deduction. Overstates average gas costs and omits relevant evidence when assigning political responsibility.
Includes household pressures and different project statuses, but underplays tariffs, bipartisan AI proposals and the pre-election permit freeze.
Credit. Acknowledges that healthy hiring can coexist with pressure on household budgets. Was October’s jobs report strong?
Credit. Distinguishes fully permitted projects from those still waiting for approvals. Could permit freezes strand unfinished data centers?
Deduction. The war-centered explanation gives too little weight to prewar tariff costs and voters’ corruption concerns. Both appear in the evidence and broaden the explanation of the defeat. Did the Iran war drive the Republican defeat?
Deduction. The account omits bipartisan sponsorship and the bill’s specified catastrophic-harm thresholds. Those features weaken the claim that oversight itself demonstrates a Democratic plan to suppress dissent. Does the AI agenda point to a censorship state?
Deduction. The closing attribution of sole responsibility to Democrats omits Abbott’s September permit freeze. The bottleneck was already developing under Republican policy. Could permit freezes strand unfinished data centers?
Credit. Flags survey limitations, ongoing proof review and uncertainty about future government control.
Deduction. Misstates the growth in spending and the implications of P = NP, while overstating what safety bills imply about broad bans.
Acknowledges limits around China and outside proof review. The inflation and regulatory arguments omit material alternative explanations and limits.
Credit. Flags the difficulty of interpreting survey responses from China. Does Gallup show a large US–China gap in AI optimism?
Credit. Makes clear that outside researchers are still evaluating the announced result. Did OpenAI report a refutation of the Exponential Time Hypothesis?
Deduction. The spending-centered explanation leaves out commodity shocks and constrained supply. Those factors help explain why prices moved differently across sectors and why inflation later eased. Does government spending explain the rise in living costs?
Deduction. The examples are presented as evidence for broad bans without explaining their narrower model-safety requirements. The scope matters to the forecast of restricted public access. The government exception still deserves scrutiny. Do existing AI bills support a forecast of broad bans?
Credit. Updates his election view and explains how scarce compute could change bargaining power.
Deduction. Understates the revenue at stake and uses the Wyoming acquisition to support a displacement story the announcement does not establish.
Considers several responses to scarce capacity, but underplays the war in the election and the developer’s role in the Wyoming sale.
Credit. Considers continued renting and use of capacity for a supplier’s own products as separate possible responses. Would a building slowdown favor owners of existing compute?
Credit. Addresses the counterargument about long-term relationships and clarifies the market conditions needed for repricing. Can SpaceX use a 90-day exit to reprice Anthropic’s compute?
Deduction. The strong dismissal of Iran underplays its prominence in the voter surveys. Economic grievances and opposition to the war can reinforce each other. Was the election mainly a demand for economic relief?
Deduction. The displacement story leaves out Crusoe’s stated pre-development role and continuing Google partnership. That context offers a straightforward commercial explanation for the sale. Did Google’s Wyoming purchase push Crusoe aside?
| Speaker | Facts | Opinions / causal arguments | Forecasts |
|---|---|---|---|
| Jason Calacanis | 1 | 2 | 1 |
| David Sacks | 2 | 1 | 3 |
| David Friedberg | 3 | 1 | 1 |
| Chamath Palihapitiya | 1 | 2 | 2 |
20 claims · 4 speakers
Democrats won both chambers of Congress and broke through in Texas, Kansas, Iowa and South Carolina.
AP called Democratic control of the Senate on November 3 and the House early November 4. Its chronology includes Democratic Senate wins in all four states Jason names.
Jason gives ranges for the final seat totals while votes are still being counted. That is appropriate. This assessment covers congressional control and the named breakthroughs; it does not treat his provisional totals as certified results.
AP’s published calls directly confirm the selected results. The remaining count can change the margins.
A reversed race call or certified result that changes control or one of the named victories.
The Iran war and the higher living costs it caused explain the scale of the Republican loss.
The Fed describes an energy-price surge after the conflict began. AP’s voter survey links the result to living costs, tariffs, Iran and corruption; corruption was crucial to more than half of voters.
Sacks gives a credible explanation for part of the backlash and openly admits his earlier prediction was wrong. His narrower account gives too little weight to concerns that preceded the war. The evidence supports several causes working together.
The energy shock and voter concerns are documented. A survey cannot isolate how many seats each cause changed.
The war-centered explanation gives too little weight to prewar tariff costs and voters’ corruption concerns. Both appear in the evidence and broaden the explanation of the defeat.
Voter-level research showing that war-related changes account for most of the swing after earlier dissatisfaction is considered.
The new jobs report showed healthy hiring, upward revisions and unemployment edging down to 4.1%.
BLS reported 165,000 additional payroll jobs, compared with a 48,000 monthly average over the previous year. August and September were revised up by a combined 84,000. Unemployment moved from 4.2% to 4.1%.
Calling this a strong monthly report is reasonable. Sacks also acknowledges that good employment figures can coexist with pressure on household purchasing power, so his broader remarks do not simply dismiss that pressure.
The same-day release confirms the figure, direction and revisions. Later revisions could change the hiring estimate.
Substantial revisions to payroll growth or the unemployment estimate.
Government spending is the main cause of rising living costs; the doubling of spending during COVID explains why asset owners gained while wage earners struggled.
Treasury reports federal outlays of $4.447 trillion in 2019 and $6.552 trillion in 2020, an increase of about 47%. Even 2025’s $7.010 trillion is about 58% above 2019. Bernanke and Blanchard identify demand shifts, commodity prices and supply constraints as contributors to pandemic inflation.
Spending can push demand beyond available supply, so Friedberg’s basic mechanism is sound. The actual increase is well short of doubling, and spending alone cannot explain the timing or size of price increases across housing, energy and goods. He labels this his view and welcomes disagreement; the deduction concerns the explanation and figures.
The outlay comparison is direct. Inflation research supports several interacting causes, though their relative importance remains debated.
The spending-centered explanation leaves out commodity shocks and constrained supply. Those factors help explain why prices moved differently across sectors and why inflation later eased.
A specified spending measure and period that support doubling, plus evidence that fiscal changes explain the other price pressures.
Gallup finds roughly 93% optimism about AI in China and 36% in the United States, putting the US near the bottom of the countries surveyed.
Gallup reports 93% in China and 36% in the US expecting AI to help their country. The US sits near the bottom of the 37-country sample. China’s daily-life figure rounds to 93% as well.
The questions cover adults who know about AI. That qualification helps interpret the percentages without changing the large gap Friedberg describes. He also flags uncertainty about surveying China. The poll cannot establish his separate suggestion that an outside influence caused American pessimism.
Gallup’s table confirms the rounded figures and broad ranking. It measures stated expectations.
A corrected table or evidence that the questions were materially different across countries.
Voters swung against the party in power because the economy was failing them; affordability mattered more than party ideology.
AP-NORC found the economy and living costs were top issues for 85% and 84% of voters. AP’s election survey also found dissatisfaction with both parties, while many voters who disliked Democrats still voted for them.
Chamath’s central interpretation fits the polling, and he explicitly revisits his earlier view. His stronger dismissal of Iran and other issues goes too far. Those concerns can reinforce economic frustration. Failed spending or gerrymandering strategies also cannot establish that ideology stopped mattering.
Two voter surveys support the economic emphasis. They cannot fully separate motives or identify every voter who switched parties.
The strong dismissal of Iran underplays its prominence in the voter surveys. Economic grievances and opposition to the war can reinforce each other.
Detailed voter research showing that other issues explain substantially more of the swing than economic dissatisfaction.
The average household is spending about $200 more each month on gasoline because of the war.
BLS puts average 2025 spending on gasoline and other fuels at $2,503 a year, about $209 a month. EIA’s gasoline figures rise from a $3.10 average in 2025 to $4.35 in September 2026 and $4.07 on November 5. With unchanged fuel use, those increases imply roughly $65–84 extra a month.
This is an approximate check using a survey average and fuel prices, with other fuels a small part of the spending category. It shows why $200 extra is too high for the national average. Heavy drivers can face larger bills. Higher freight and grocery costs are additional burdens, but Sacks specifically describes what households pay at the pump.
The estimate depends on fuel use staying similar, yet the gap is far larger than ordinary rounding.
Current household spending data showing a roughly $2,400 annual increase in direct fuel purchases.
Tariffs had already raised consumer prices and contributed to dissatisfaction before the Iran war began.
Fed researchers estimate that 2025 tariffs raised core goods prices by 3.1% through February 2026 and the broader core PCE price level by 0.8%. AP’s voter survey also identifies tariffs among voters’ economic complaints.
Jason’s point directly answers the suggestion that affordability trouble started with the war. He also accepts that the war added to it. The exact effect on votes is harder to separate, but the earlier price increase is well supported.
The research isolates tariff exposure before the conflict, and the survey establishes that voters noticed the issue.
Revised price research that removes the estimated tariff effect or voter evidence showing it had little relevance.
Ending the Iran war could bring Brent back toward $70, lower fuel costs and help Trump recover politically.
EIA links high oil prices to disrupted Middle East flows and depleted inventories. Its October outlook expects Brent to fall from about $105 in late 2026 to an $84 average in 2027 as flows improve. The forecast still assumes supply constraints.
A fuller peace settlement could produce a larger decline than EIA’s constrained-supply case, so $70 is a plausible scenario. Repair times, inventories and producers’ responses will affect the speed. Lower fuel prices could help politically, though other grievances would remain. The outcome is still open.
The supply mechanism is strong. The size and timing depend on what ending the war actually restores.
A settlement that fails to restore exports, persistent infrastructure damage, or supply cuts that keep prices high.
Trump should pardon himself and his administration now to remove, or at least weaken, the Democrats’ ability to pursue endless investigations.
CRS explains that presidential pardons cover past federal crimes and leave impeachment and state offenses outside their reach. Congress can investigate pardon decisions. Whether a president can pardon himself remains legally unresolved.
Jason qualifies the idea as weakening investigations “to a certain extent,” which deserves credit. Removing some criminal exposure could change the political stakes. It would still leave Congress able to investigate, and the self-pardon part assumes a disputed power. The proposal’s practical effect is smaller than the initial pitch suggests.
The scope of federal pardons is established; self-pardons and the political response remain uncertain.
A controlling court ruling on self-pardons or a specific investigation whose purpose would disappear after a valid pardon.
Eliminating federal taxes for people earning $125,000 or less would barely affect revenue and could be offset by closing a few loopholes.
IRS data for 2023 show $270.25 billion in total income tax from returns with positive adjusted gross income below $100,000, or 12.6% of the total. The proposed $125,000 threshold includes additional taxpayers.
The fairest reading is federal income tax. The lower $100,000 band alone shows a substantial revenue base. A current cost estimate would also need the filing rules, phaseouts and offsets. Calling the revenue negligible is the material error.
The IRS table directly measures the narrower group, which is enough to test the claim of a trivial cost.
A specified proposal with credible offsets or a much narrower exemption that makes the net cost small.
Slower data-center construction would strengthen hyperscalers with existing capacity and squeeze AI companies that depend on rented compute.
Anthropic’s filing shows large compute commitments to several suppliers. Texas has already paused data-center permits, and Google’s Wyoming acquisition includes a developed power strategy.
Chamath describes a credible supply bottleneck and clearly frames it as his expectation. Existing capacity becomes more valuable when substitutes are delayed. The advantage will vary by contracts, location, available power and the ability to switch hardware; some suppliers also need new construction. That limits a blanket prediction about every company’s prospects.
The contracts and permitting constraints support the mechanism. They do not determine future prices or each company’s exposure.
New capacity arriving faster than expected, weaker demand, or efficiency improvements that ease the bottleneck.
SpaceX and Anthropic have mutual 90-day termination rights; if compute becomes scarce enough, SpaceX could use them to demand much higher prices.
Anthropic’s filing says either party can cancel on 90 days’ written notice after initial non-cancelable periods ending in January and March 2027. It also discloses substantial commitments to other suppliers.
The clause makes repricing a credible risk. Chamath later explains that he means a sufficiently large market-price increase, which is a reasonable condition. The initial protected periods matter to timing. A doubling or tripling of prices, and Anthropic having no alternative, remain assumptions about future bargaining.
The filing verifies the clause and its timing. It cannot tell us how either company will exercise that option.
An amended contract, substitute capacity, or actual negotiations that change the expected bargaining power.
If US states hold up data centers, investment and AI computing could move overseas, potentially making the US a net importer of AI services.
Google announced a $15 billion Malaysia expansion for 2027–28, including new campuses and more than 1 GW of clean-energy capacity. Its plan also includes grid, water and local-workforce investments.
Jason’s relocation forecast follows a familiar commercial incentive. His later description of moving anywhere as extremely easy understates the physical work involved. Fast networks help deliver results, while power, chips and approvals still determine where capacity can be built. The evidence supports some relocation; net-importer status depends on the balance of US and overseas growth.
An announced overseas expansion supports feasibility. Its eventual scale and the US trade balance in AI services remain open.
Deployment and cross-border usage data showing whether foreign capacity is replacing US projects or mainly serving new local demand.
Politicians may ban AI use as well as development; California’s vetoed bill and other regulations show a serious risk of government control over access.
The SB 1047 veto message describes safeguards for large models and catastrophic risks. The bipartisan FRONTIER proposal requires evaluations and reporting. Its catastrophic-risk definition excludes lawful federal-government activity, a concrete distinction worth scrutinizing.
Friedberg later says the concern is a gradual loss of individual power, and rejects an all-or-nothing reading. That qualification matters. Government exemptions can justify concern about unequal treatment. These examples still leave a large gap between safety oversight and broadly banning private AI use. His risk estimate needs that gap explained.
The documents establish narrower requirements and a government exception. Their long-term political consequences remain uncertain.
The examples are presented as evidence for broad bans without explaining their narrower model-safety requirements. The scope matters to the forecast of restricted public access. The government exception still deserves scrutiny.
Specific proposals that substantially reserve capabilities for the state or prohibit ordinary private AI use.
Democrats will push AI legislation that ultimately uses model approvals to build a more powerful surveillance and censorship state.
The FRONTIER proposal shows real interest in federal oversight, with Democratic and Republican sponsors. Its text focuses on severe physical, cyber and property harms, and contains a government-activity exception.
Sacks identifies a credible risk: regulators can gain influence over which systems reach users. The exception also warrants scrutiny. He goes further by attributing a shared censorship goal to Democrats and treating that outcome as the direction of travel. The cited proposals do not establish either conclusion. This is assessed as a forecast, with credit for the underlying institutional concern.
The legislative mechanism can be examined now. Future misuse and a whole party’s intentions cannot be inferred from it alone.
The account omits bipartisan sponsorship and the bill’s specified catastrophic-harm thresholds. Those features weaken the claim that oversight itself demonstrates a Democratic plan to suppress dissent.
Proposed approval rules targeting lawful dissent, surveillance mandates, or evidence that safety powers are being used that way.
Google taking over Crusoe’s Wyoming data center shows hyperscalers displacing the companies that rent compute.
Crusoe says Google bought the site after Crusoe completed pre-development work on location, approvals and power. It also says Crusoe remains the developer of a separate Google campus in Texas.
Google did acquire the Wyoming site. The release gives no evidence of an operating tenant being pushed out or Crusoe losing capacity it expected to keep. It fits a developer preparing a site for a buyer. The announcement alone cannot establish the distress story Chamath uses it to support.
The transaction and development role are explicit. The full commercial terms are not public in the announcement.
The displacement story leaves out Crusoe’s stated pre-development role and continuing Google partnership. That context offers a straightforward commercial explanation for the sale.
Deal terms or statements showing forced displacement, a canceled tenancy or a distressed sale.
Projects under construction may still need approvals, so a freeze could delay completion and sharply reduce the 2028 pipeline.
TCEQ generally requires air permits before construction, while allowing some early site work and limited exceptions. Abbott had already ordered a halt to data-center permits in September pending grid and water audits.
Sacks carefully distinguishes fully permitted projects from those still awaiting decisions and says the outcome could go several ways. That makes his delay forecast reasonable. His closing claim that Democrats would be solely responsible overlooks the existing Republican freeze. How long new restrictions last will matter more than the election result alone.
The permit sequence and existing pause establish the risk. Their effect on 2028 depends on project details and future policy.
The closing attribution of sole responsibility to Democrats omits Abbott’s September permit freeze. The bottleneck was already developing under Republican policy.
Grandfathering rules, a prompt end to the pause, or project-level evidence showing the pipeline already has its required approvals.
OpenAI produced a disproof of the Exponential Time Hypothesis, with outside researchers still evaluating the result.
OpenAI’s October 26 release says its system found a refutation on October 16 and formalized it in Lean the next day. It names an external group led by Impagliazzo, Paturi and Williams to examine and explain the work.
Friedberg correctly conveys the announced result and explicitly says outside verification is still underway. Formal verification strengthens the announcement, while expert examination checks the formulation and implications. This review confirms that account; it has not independently verified the proof or the reported hours of computation.
The primary announcement supports the selected claim and its qualification. Independent proof verification is beyond this review.
An identified flaw in the formal statement, proof or implementation, or a correction by the researchers.
Refuting the Exponential Time Hypothesis keeps P = NP in play, which would mean every problem in the universe can be solved by a computer.
Cook’s formal statement concerns problems whose candidate answers can be checked in polynomial time. Subexponential time can still be much slower than polynomial time. Undecidable problems remain outside those classes.
Refuting ETH would be a major result, but P = NP was already unresolved. Even a proof that P = NP would concern a specific class of computational problems. It would not supply algorithms for undecidable problems or show that every physical question can be computed. That changes the conclusion far more than a loose analogy would.
The distinction follows from the definitions of the complexity classes and computability, independently of whether the new ETH result holds.
A narrower statement limited to efficiently checkable problems. The claim about every problem would require a fundamentally different argument.
No claims match these filters.
Published November 7 at 18:27 UTC; recorded Friday, November 6. All four regular hosts participate.
The full recording runs 1:33:03. Claim links use automated-caption timestamps.
Use Show transcript on the recording. The full automated English transcript was read.
AP’s own race-call chronology. Some races were still being counted when the panel recorded.
Findings from the AP Voter Poll, conducted across all 50 states before and through election day.
October 15 report covering the economy, living costs, Iran and other issues.
November 6 release, including revised August and September payroll estimates.
April 8 research note estimating price effects of tariffs implemented through November 2025.
Discusses tariff costs, the Middle East conflict, supply constraints and consumer prices.
Budget outlays in billions of dollars, with the prior-year comparison.
The 2025 statement reports $7.010 trillion in outlays.
Research on demand, commodity prices, supply bottlenecks and labor markets.
September 22 report. The optimism questions cover adults who are aware of AI.
Average annual spending per consumer unit, including gasoline and other fuels.
October 6 forecast, completed October 1. Includes 2025 prices and September 2026 fuel prices.
Daily snapshot accessed November 8: regular gasoline $4.07 per gallon and Brent $97.96. This page updates over time.
Explains the reach of federal pardons and congressional oversight.
Article II, section 2 discussion: self-pardons remain an unresolved legal question.
Table 1.1, total income tax column. Dollar amounts in the spreadsheet are in thousands.
Contractual commitments describe SpaceXAI termination rights and commitments to other suppliers.
November 4 announcement of $15 billion planned investment during 2027–28, with power and water commitments.
September 29, 2024 explanation of the proposed safeguards and reasons for the veto.
July 23, 2026 announcement of tiered oversight, evaluations and reporting requirements.
Definitions and oversight provisions; this is a proposal, with no assumption that it will become law.
November 6 announcement explaining Crusoe’s completed pre-development work and Google’s acquisition.
Explains when permits are required and which early site activities may proceed before issuance.
September 21 directive pausing permits pending grid and water audits.
October 26 announcement with research materials and an external group examining the result.
Formal problem statement and the distinction between efficient computation and computability.
Each dimension is scored from 0–10. The weighted total is multiplied by ten and rounded to the nearest five, with halves rounded up.
Grades use that rounded score: A ≥80, B ≥70, C ≥60, D ≥50, F <50.
Accuracy tests factual correctness; coherence tests whether the conclusion follows; evidence support tests the strength of the sources. Calibration asks whether the speaker’s certainty matches the evidence.
Balance assesses cherry-picking, omitted counterevidence, inconsistent standards and distorted alternatives. Higher scores mean fairer treatment of evidence. Each deduction names the affected claim, the missing evidence and why it changes the conclusion.
Political disagreement and presumed intent do not count. A false claim or weak source alone does not establish cherry-picking.
Scores assess selected arguments. They do not measure a speaker’s overall reliability or percentage of correct statements. Confidence ratings are subjective; no statistical intervals are calculated. Automated-transcript attribution remains provisional. Small differences are not statistically meaningful, and no independent second rater has checked the scores.
New reviews select up to 20 significant claims, without speaker quotas. The sample is editorial, not random or exhaustive. All-In #286–292 retain their original 14-claim samples. An absent or unassessable speaker has no score and leaves a gap in the chart.
Primary sources support the assessments. Current product and methodology pages are labeled when used retrospectively. Unresolved forecasts are assessed for their support. All-In recording links identify chapter starts or video timestamps. Pivot uses transcript timestamps, which can differ from ad-supported audio. GoodFellows uses automated-caption timestamps; its speaker attribution has not been audio-checked.
This review: 20 selected claims. Published Nov 7, 2026; reviewed Nov 8, 2026.
Full automated English transcript read; speaker attribution provisional and not audio-checked.
Rubric v2.0. Earlier scores and correction records remain in the review data.