All-In PodcastE135: Wagner rebels, SCOTUS ends AA, AI M&A, startups gone bad, spacetime warps & more
CHAPTERS
- 0:00 – 2:43
Friedberg takes the mic: feisty intro and bestie banter
Friedberg fills in as moderator and opens with a deliberately provocative, satirical intro about “conspiracy” narratives. The group riffs on travel, time zones, and inside jokes before transitioning to the week’s headline topics.
- •Friedberg substitutes as moderator; playful roasting of the hosts
- •Running jokes about media narratives, censorship, and “mums the word”
- •Quick check-in on Jason’s voice and recent heavy podcast schedule
- •Set-up for the show’s main segments
- 2:43 – 8:04
Wagner mutiny recap: what happened and why it stopped short of Moscow
Friedberg summarizes the Wagner Group’s march, seizure of Rostov-on-Don, and abrupt stand-down after a Belarus-brokered deal. Sacks argues it was a real mutiny triggered by Wagner’s forced integration into Russia’s defense ministry structure.
- •Timeline: Rostov occupation, convoy toward Moscow, sudden negotiation and halt
- •Trigger: MoD contracts would reduce Prigozhin’s income/status and autonomy
- •Sacks rejects the “staged” theory; frames it as mutiny with coup optionality
- •Deal mechanics: Lukashenko mediation, exile/immunity claims, Wagner future uncertain
- 8:04 – 23:04
Interpreting the rebellion: propaganda, public support, and escalation risks
The panel debates whether the mutiny signals regime fragility or consolidation, and how reliable any “Putin support” measures are under authoritarian conditions. Sacks emphasizes the danger of wishing for chaotic succession in a nuclear state and warns the event could push Putin toward escalation.
- •Debate over Russian public opinion: polling skepticism vs. ‘rally-around-flag’ thesis
- •Propaganda culture moment (patriotic song) used as anecdotal evidence
- •Risk lens: replacing Putin could yield a worse hardliner with nuclear control
- •Sacks’ prediction: pressure may incentivize more aggressive mobilization/escalation
- 23:04 – 25:13
From Ukraine to SCOTUS: setting the table for affirmative action rulings
Friedberg pivots from geopolitics to his personal UC Berkeley experience during the end of affirmative action in California. He lays out the Supreme Court decisions in the Harvard and UNC cases and the core allegation of discrimination against Asian-American applicants.
- •Friedberg’s Berkeley context and reference to Bakke/UC Regents history
- •SCOTUS decisions: Harvard (6–2) and UNC (6–3) against race-based admissions
- •Students for Fair Admissions as plaintiff; Asian-American discrimination focus
- •Immediate implication: admissions criteria and applications must change
- 25:13 – 30:47
What changes next: legacy, athletics, and spillover into corporate DEI/ESG
Chamath argues the ruling will trigger downstream challenges not only to admissions forms but also to legacy and athletics preferences. He extends the logic into private enterprise—predicting increased legal scrutiny of race-based recruiting programs and parts of ESG/DEI frameworks.
- •Harvard admission-rate-by-decile chart illustrating perceived unfairness
- •Prediction: legacy and athletics admissions become the next legal battleground
- •Corporate implications: race-targeted hiring/recruiting initiatives may be challenged
- •Claim: changes will begin slowly in higher ed, then accelerate into business practices
- 30:47 – 36:25
Values debate: meritocracy vs. fairness, and “fix the pipeline earlier”
Jason and others frame affirmative action as a clash between equality of opportunity and equality of outcomes, arguing college admissions is too late to correct unequal starting conditions. The group converges on earlier interventions—childcare, pre-K, K–12 quality, and school choice—as more durable solutions.
- •Meritocracy tension: what ‘achievement’ should mean in admissions decisions
- •Jason’s view: invest earlier (childcare/pre-K/elementary) rather than college ‘band-aids’
- •Discussion of real-world DEI frictions and perceived quota-like behavior
- •Sacks emphasizes school choice/competition and critiques union-driven monopolies
- 36:25 – 50:51
Legacy admissions hypocrisy and Chamath’s “transparent rate card” proposal
Chamath cites research on the scale of athlete/legacy/donor-related admissions advantages and calls the system structurally unfair if it claims meritocracy. He proposes radical transparency—publish a “rate card” for donor/legacy pathways—while the panel debates federal funding and quasi-public status.
- •NBER stat cited: large share of Harvard admits tied to athlete/legacy/donor categories
- •Argument: if you can’t be race-conscious, you should also remove non-merit preferences
- •Chamath proposal: publish explicit donor price tags to make the system honest
- •Federal funding as key distinction between private club autonomy and public accountability
- 50:51 – 56:58
AI M&A shockwave: Databricks buys MosaicML and the race for the AI toolchain
Friedberg breaks down Databricks’ $1.3B MosaicML acquisition and why the real price may differ due to private-stock valuation assumptions. Sacks shares firsthand context on Mosaic’s prior fundraising and frames the deal as strategic—enterprise infra companies are assembling an end-to-end AI stack amid GPU scarcity.
- •Deal terms nuance: cash + private stock valued off Databricks’ 2021 round
- •Sacks: Mosaic had a Series B term sheet; rapid ARR growth and strategic attractiveness
- •Why Mosaic matters: training/customization efficiency and GPU utilization as scarce resource
- •Broader trend: data platforms “level up” from storage/ETL into model training and inference
- 56:58 – 1:00:28
What AI enables: natural-language access to enterprise data and new startup surfaces
Jason and Friedberg describe how LLMs make previously uneconomic products viable by slashing transcription/analysis costs and enabling natural-language interfaces. The conversation expands from infra M&A to practical use cases, model selection beyond OpenAI, and the coming wave of AI-native startups.
- •Shift: business users query data directly without needing data scientists
- •Startups now test multiple models (not always OpenAI) before choosing a stack
- •Example: ingesting and analyzing public meeting recordings at scale becomes viable
- •Expectation: more AI acquisitions as incumbents race to avoid commoditization
- 1:00:28 – 1:09:27
Inflection’s $1.3B round: GPU economics, round-tripping, and hype-cycle warnings
Friedberg introduces Inflection AI’s funding and planned 22,000 H100 cluster. Chamath analyzes the economics—capex intensity, implied enterprise value for a chatbot product, and the possibility of “round-tripping” between cloud providers, chipmakers, and startups.
- •H100 pricing and capex math: cluster costs approach ~$1B with infrastructure
- •Chamath skepticism: heavy equipment spend as unusual share of startup enterprise value
- •Concern: strategic investors funding startups that then buy their compute/services
- •Prediction: early-cycle froth peaks when hype is highest and facts are minimal
- 1:09:27 – 1:17:32
Startups gone bad: IRL’s fake users, Byju’s distress, and VC diligence failures
Friedberg tees up IRL’s shutdown after a board probe found most users were fake, alongside Byju’s scramble for capital and governance conflict. Chamath argues these outcomes reflect weak checks-and-balances, inexperienced board members, and the incentives created by oversized venture funds.
- •IRL: 95% fake users allegation, SoftBank-led unicorn era cautionary tale
- •Byju’s: valuation markdown, emergency fundraising, founder control pressure
- •Chamath critique: poor diligence culture and inability to ask hard questions
- •Friedberg: ‘trust but verify’—bank statements, customer calls, real metrics matter
- 1:17:32 – 1:27:17
Fund size as destiny: late-stage fundraising collapse and the “zombiecorn” purge
Sacks and Chamath argue that massive funds forced unnaturally large checks into early-stage companies, magnifying mistakes. They cite reports of late-stage mega-funds struggling to raise capital and predict many unicorns will either go to zero or face painful down rounds.
- •SoftBank dynamic: $100B funds require giant checks; small checks aren’t worth the time
- •Late-stage fundraising ‘dead’: examples of Insight/Tiger target reductions and shortfalls
- •“Zombiecorns”: estimate that a large fraction of unicorns are functionally stalled
- •Return distribution: many zeros, some money-back outcomes, few true outliers drive averages
- 1:27:17 – 1:36:57
Science Corner: NANOGrav, pulsar timing arrays, and spacetime ripples
Friedberg explains NANOGrav’s 15-year pulsar dataset and how tiny timing variations can indicate a background of gravitational waves from supermassive black hole dynamics. He connects the finding to general relativity and describes how improved measurements could reveal a “gravitational-wave fingerprint” of the universe.
- •Pulsars as precise cosmic clocks; long baselines reveal subtle timing correlations
- •Interpretation: spacetime itself undulates due to gravitational waves
- •Potential sources: supermassive black hole binaries shaping a low-frequency background
- •Analogy to the CMB: possible map/fingerprint of the universe’s large-scale structure via gravity
