All-In PodcastTariffs, Trump's Economic Endgame, Market Chaos, Bitcoin Reserve, CoreWeave IPO
CHAPTERS
- 0:00 – 23:00
Cold Open, Intros, and Trump Week ‘Tsunami’ Recap
The episode opens with banter, introductions of Chamath, Friedberg, Jason, guest Joe Lonsdale, and a tease that Sacks will join later. Jason then rattles through a breathless recap of a hyperactive Trump week: Epstein file flop, Zelenskyy drama, crypto-reserve tweets, wild crypto price moves, tariff whiplash, NATO and Department of Education rumors, GPT‑4.5 and Qwen releases, LLaMA stagnation, and massive market volatility.
- •Joe Lonsdale reintroduced as a right‑leaning VC with defense tech ties.
- •Jason outlines an incredibly dense week of Trump actions and market reactions, including crypto tweets that sent specific coins soaring.
- •Mentions GPT‑4.5’s muted reception, Alibaba’s Qwen open‑source model outperforming, and signs that Meta’s LLaMA may be plateauing.
- •Markets are experiencing some of the wildest swings in 20 years, with ‘Mag 7’ compression and capital flight from Europe.
- 23:00 – 45:00
What’s Trump Really Doing With Tariffs?
The besties tackle Trump’s erratic tariff announcements and walk-backs. Lonsdale sees them as leverage over border security and fentanyl; Chamath frames them as part of a deeper plan to reset the dollar’s role, deficit funding, and long-term U.S. industrial vibrancy. Friedberg offers a ‘grandmaster’ theory linking tariffs, income-tax cuts, and spending cuts to re‑onshore manufacturing and transition from income to consumption taxation.
- •Lonsdale says tariffs on Canada/Mexico are negotiation tools to force fentanyl crackdowns.
- •Chamath argues tariffs can rebase reliance on the dollar, reshape foreign reserves, and make U.S. firms more competitive.
- •Friedberg’s ‘three legs’: tariffs (raise import costs, push onshoring), lower income taxes (free capital for new factories), reduced government spending (shift workers to private sector and counteract tariff inflation).
- •Discussion of moving from taxing income to taxing consumption, and whether that’s more ‘fair’ and pro‑growth.
- •Concern about current ultra‑low unemployment (~4%) and where workers for new domestic industries will come from.
- 45:00 – 1:01:00
Government Bloat, DOGE Audits, and Cutting $5 Trillion
The panel drills into the size and structure of government spending, including the indirect army of contractors and NGOs. DOGE audit data showing thousands of unused software licenses becomes a symbol of systemic waste. Lonsdale blasts Congress for lacking the backbone to match Elon's aggressive cuts at the agency level, and Chamath warns that now comes the hard phase where specific programs like the Inflation Reduction Act (IRA) must be surgically re‑evaluated.
- •Friedberg notes that while federal headcount growth is significant, the bigger story is 30% of GDP flowing through government to contractors and NGOs.
- •DOGE finds ~35,855 ServiceNow licenses used by only 84 people and 11,000 Acrobat licenses with zero users—Jason calls for clawbacks and procurement investigations.
- •Lonsdale argues Congress is only talking about $1–2T in 10‑year cuts when at least $5T is needed.
- •Chamath explains how IRA repeal is nontrivial: 90%+ of new U.S. electricity in December came from renewables, heavily dependent on IRA tax equity incentives; gutting it jeopardizes AI compute growth because of looming power shortages.
- •They argue for starting with ‘chainsaw’ cuts (obvious waste) then moving to cautious, surgical revisions of big programs like the IRA.
- 1:01:00 – 1:20:00
Kleptocracy Fears, Defense Procurement, and Open Standards
Responding to claims that Silicon Valley figures entering government is a new kleptocracy, Lonsdale and Chamath argue the real goal is fair competition in defense and federal contracts. They detail how incumbents like Raytheon have long written RFPs to entrench themselves, and call for transparent, outcome-based acquisition with open standards and published scoring so new tech can win on merit.
- •Lonsdale describes Epirus outperforming legacy defense primes by ~9.5x in range on hardened drones at far lower cost, yet losing due to RFPs written around old tech.
- •He wants contests defined by outcomes, not inputs, so newer architectures can compete.
- •Chamath says people will always suspect influence‑peddling unless RFPs, specs, eval criteria, and test data are all public and appealable.
- •They advocate acquisition reform modeled on early 20th‑century pistol competitions (simple outcome specs, rapid trials) instead of 700‑page RFPs.
- •Press, whistleblowers, and social transparency (like DOGE’s feed) are seen as vital safeguards.
- 1:20:00 – 1:44:00
Super PACs, Free Speech, and Citizens United
The group debates whether to cap mega-donations and roll back Citizens United. Lonsdale and Friedberg defend super PACs as free speech and a counterweight to unions, celebrities, and professional lobbies. Chamath and Jason, concerned about gerrymandering and downstream capture, lean toward returning to pre–Citizens United limits to revitalize local and state democracy.
- •Lonsdale: super PACs equal the ability to fund speech (books, websites, organizers), not just ‘buy politicians’; limiting them just reweights power to unions and institutional interests.
- •Jason pushes for caps (e.g., $5–25M per person and per cycle) and public funding for final candidates to reduce the appearance of impropriety.
- •Chamath highlights massive post–Citizens United spending by Soros, Zuck, the Kochs, and now Elon; sees biggest problem as frozen local/state races via gerrymandering.
- •Friedberg worries any content-based cap requires a ‘censorship board’ to decide what speech is political vs issue advocacy; he instead favors robust free expression and cracking down on clearly abusive tactics (e.g., literally paying people to vote).
- 1:44:00 – 1:58:00
CoreWeave IPO and the GPU Arbitrage Question
The conversation shifts to CoreWeave, a ‘NeoCloud’ GPU provider racing toward an IPO at a >$30B valuation. The besties marvel at its growth and technical decisions (bare‑metal instead of hypervisors; aggressive GPU/power lockups), but worry about concentration risk and structural headwinds as hyperscalers pour tens of billions into their own GPU fleets.
- •CoreWeave runs ~32 data centers with ~250K NVIDIA GPUs and ~$1.9B 2024 revenue but nearly $1B loss and ~$8B debt.
- •Chamath credits a key design choice—no hypervisors, direct-to-metal access—for enabling them to beat AWS/GCP/Azure for certain AI workloads.
- •Primary existential risk: misjudging GPU useful life; if economics assume 10 years but tech/energy curves force replacement in 5, the capital stack breaks.
- •Lonsdale notes founders’ commodities-trader mentality; they locked in GPU, power, and data-center supply early, giving them a strong but possibly transient edge.
- •Friedberg likens CoreWeave to early ‘speed doubler’ dial-up accelerators—lucrative during an infrastructure gap but vulnerable once broadband (here, hyperscaler GPU clouds) catches up.
- 1:58:00 – 2:16:00
Markets, Bonds, and Trump’s Main Street Gambit
Chamath lays out a three-market framework—U.S. long bonds, U.S. equities, and European bonds/equities—and argues Trump is privileging Main Street over Wall Street. Tariffs, war posture, and rhetoric about not watching the stock market are interpreted as deliberate choices to tame inflation, lower Treasury yields, and force Europe to internalize Ukraine war costs.
- •Chamath sees MAGA’s core coalition as working/middle-class non‑asset‑owners, tech optimists, and patriotic business owners—none centered on stock indexes.
- •Letting equities fall curbs asset-driven consumption, helping fight inflation and pushing investors into Treasuries, lowering yields just in time for massive U.S. refinancing needs.
- •Europe’s decision to self-fund Ukraine defense via new borrowing is pushing their yields higher; UK is using accounting tactics to hide debt/GDP impact, but bond markets are pushing back.
- •Lonsdale agrees war is inflationary and bond markets ‘hate war’; he emphasizes rate cuts’ importance for Main Street industries like housing and real estate services.
- •Friedberg estimates ~60% chance Trump 2.0 is less sensitive to short-term market noise and more focused on using lower rates to lock in sustainable financing for the $10T coming due.
- 2:16:00 – 2:31:00
Ukraine, NATO, and a Resource-Abundant Future
The episode dives into Ukraine strategy and NATO’s future. Lonsdale supports a tough stance on Putin but wants Zelenskyy to approach peace talks more humbly and realistically, acknowledging likely territorial loss. On NATO, he’s emotionally attached but deeply skeptical of Europe’s trajectory; Friedberg counters with a techno‑optimist vision where resource-tech progress makes many geopolitical conflicts over land and minerals obsolete.
- •Lonsdale: Putin is clearly wrong to invade, but the previous U.S. administration mismanaged deterrence; now a negotiated peace that sacrifices some Ukrainian territory may be the only realistic path.
- •He’s critical of Zelenskyy’s posture in D.C. and notes Ukraine’s corruption reputation; stresses ‘peace through strength’ as the right doctrine.
- •On NATO, Lonsdale views Europe as increasingly dysfunctional, especially on free speech and crime, but still values U.S.–U.K. ties and wants leverage to force European reforms, not an abrupt exit.
- •Friedberg explains why AI, thorium reserves, and improved mineral discovery/mining could make energy and materials effectively abundant for millennia, reducing incentives for empire and conflict.
- •In a future of abundance, he questions the need for U.S. policing over ‘some jungle’ or foreign plot of land and floats a leaner, multipolar order with more limited U.S. security commitments.
- 2:31:00 – 3:18:00
LLMs, Grok’s Rise, and the Limits of Benchmarks
The panel surveys the rapidly evolving AI model landscape. Chamath praises Claude for code and Grok for consumer UX, notes Alibaba’s Qwen and DeepSeek as standout open models, and argues we’re at the point where classic benchmarks are overfitted and less informative. Joe and Jason share how cheap inference is enabling new always‑on AI workflows in email and enterprise tools.
- •Chamath: Claude 3.7 is ‘exceptional’ for codegen; Grok‑3 is his default for consumer tasks, especially because of its X integration (tweet‑context summaries with one click).
- •GPT‑4.5 feels underwhelming largely because evaluations like SWE‑Bench and IMO are saturated—models are trained directly on them, obscuring true generalization.
- •He calls for independent, rotating benchmarks run by third parties, analogous to FAA or NCAP testing, to restore trust in capability claims.
- •Lonsdale says competition among models is great for builders; his portfolio companies use multiple LLMs (e.g., Grok, Cognition) and are seeing 10–15% monthly improvements.
- •Jason notes that Superhuman now drafts and summarizes everything by default because AI is cheap enough; he predicts a new job class of ‘AI trainers’ and fact‑checkers, especially recruiting laid‑off journalists.
- 3:18:00 – 3:35:00
Sacks Explains Trump’s Bitcoin Reserve and Stockpile
In the second half, David Sacks joins from the White House to detail Trump’s new Bitcoin Strategic Reserve and Digital Asset Stockpile. He delineates Bitcoin’s special treatment, explains how seized digital assets will be audited and centralized, and addresses accusations that he’s personally profiting from policy. The discussion then turns to crypto market structure, disclosure, and how to avoid the government ‘picking winners.’
- •Bitcoin Strategic Reserve: only BTC, treated as ‘digital Fort Knox’; EO explicitly prohibits Treasury from selling it; goal is long-term preservation, not active trading.
- •Digital Asset Stockpile: centralized custody for all other seized digital assets (ETH, alt‑coins, etc.); Treasury has discretion to rebalance and manage the portfolio like a fund, including selling or reallocating.
- •Historic misstep: U.S. once held ~400K BTC but sold ~200K for ~$360M; that tranche alone would now be worth >$17B. New policy aims to avoid repeating that error.
- •EO mandates the first government-wide audit of all digital assets across agencies; once court cases are final, seized assets move into reserve/stockpile.
- •Sacks emphasizes Bitcoin’s uniqueness: no issuer, unknown founder, massive market cap, never hacked—a ‘$2T bug bounty’ that has never been collected.
- •Treasury/Commerce can pursue ‘budget-neutral’ ways to acquire more BTC; Jason floats a tiny transactional ‘crypto tax’ as one possible mechanism, though Sacks is tax-skeptical.
- 3:35:00
Sacks on Divesting Crypto and Frameworks for Regulation
Sacks defends himself against accusations of self‑dealing by outlining his complete divestment from crypto and related funds prior to joining the administration. The conversation then focuses on how to design crypto regulation around disclosure and market structure, including how to define securities, collectibles, and the role of FIT‑21 and the SEC’s new leadership.
- •Sacks reveals that he and Craft sold roughly $200M of crypto (about $85M attributable to him) before day one of the administration, paying taxes and exiting positions to avoid conflicts.
- •He also sold LP interests in crypto‑exposed funds (Bitwise, Multicoin, Blockchain Capital, and Jason’s fund) at discounts, emphasizing that public service is costing him money.
- •He calls allegations of ‘bag pumping’ slanderous, noting accusers presented no evidence and were effectively alleging crimes.
- •Regulatory philosophy: government shouldn’t ‘pick winners’ beyond recognizing Bitcoin’s unique status; core is robust disclosure—token cap tables, insider holdings and vesting, minting mechanisms.
- •Market structure bill FIT‑21 (by Rep. French Hill) is expected to be reintroduced, defining categories (security vs commodity vs collectible) and their rules; SEC Commissioner Hester Peirce is leading crypto work; Paul Atkins has been nominated as SEC Chair.
- •Sacks favors tough enforcement for fraud (lying about tokens’ features/value) but broad freedom to trade honestly described assets, including meme/collectible coins as long as their lack of intrinsic value is clearly disclosed.
