All-In PodcastE174: Inflation stays hot, AI disclosure bill, Drone warfare, defense startups & more
CHAPTERS
- 0:00 – 17:20
Summit Announcements And Global City Decay Observations
The hosts open by announcing the All-In Summit 2024 in Los Angeles and describing upgrades to the attendee experience. Chamath then debriefs a Paris AI conference and segues into a broader discussion of Western cities’ decline—petty crime, drugs, and civic mismanagement in Paris, London, San Francisco, and other liberal urban centers.
- •All-In Summit 2024 returns to UCLA in LA, Sept 8–10, with one unified ticket tier, transportation, concierge services, and expanded programming.
- •Chamath’s Paris trip included a fireside chat with Groq founder Jonathan Ross, who outlined architectural differences vs. NVIDIA, especially on inference efficiency and supply-chain independence.
- •He highlights Ross’s claim that Groq could capture ~50% of available inference compute by end of next year, with far lower power and cost.
- •Chamath draws parallels between NVIDIA’s current playbook and Intel’s 80s–90s dominance via defining metrics (e.g., clock speed), arguing NVIDIA’s choices around HBM and optics may be unsustainable.
- •Conversation pivots to Western city disorder: rising petty crime, visible drugs, vandalism, and how soft-on-crime policies and decriminalization have degraded quality of life.
- 17:20 – 32:40
Progressive Urban Policy, Crime, And Democratic Course Correction
The besties dissect how decriminalization, reduced prosecution of misdemeanors, and generous subsidies have reshaped street life in blue U.S. cities, especially San Francisco. They debate whether recent elections and DA changes mark a bottom and how long it takes for urban turnarounds, using New York’s 70s–80s recovery as a historical analog.
- •Sacks attributes urban chaos to a decade-plus of policies: decriminalization, decarceration, Soros-backed DAs reluctant to prosecute misdemeanors, and a Prop 47 threshold that turned many felonies into misdemeanors.
- •San Francisco’s subsidies for street living and open drug use have effectively ceded public space to addicts and vagrants.
- •Recent SF elections ousted more radical supervisors and installed a tougher DA, suggesting voters have had enough, but entrenched one-party machinery makes rapid change difficult.
- •Chamath recalls NYC’s long turnaround that took two mayors (Dinkins and Giuliani) and broken-windows policing, arguing Western capitals like Paris and London now face comparable knife crime and lawlessness.
- •They note that no major Western city has yet fully embraced aggressive policing and punishment again, so a structural fix remains pending.
- 32:40 – 43:00
Cloud Conferences, Multi-Cloud Reality, And AI Infrastructure Competition
Friedberg reports back from Google Next in Las Vegas, describing the massive ecosystem around cloud providers and how enterprises increasingly use multiple clouds. The trio unpack how commoditized infra is driving a race to the bottom on price, while differentiation and margins migrate to higher-level AI and developer services.
- •Google Next hosted ~30,000 attendees, with entire convention centers filled by cloud partners, integrators, and consultancies like Deloitte, Capgemini, and PwC.
- •Most large enterprises run multi-cloud—spreading workloads across AWS, Azure, and GCP based on price and specialized services, rather than choosing a single provider.
- •Basic services (compute, storage) are becoming commodities; profit pools are shifting to AI models, ML services, and managed higher-level tooling.
- •Cloud vendors court global consultancies and developer ecosystems to drive adoption and lock-in via proprietary developer services and technical assistance.
- •Friedberg’s big takeaway: the cloud market remains far more open and competitive than many assume; no hyperscaler monopoly is locked in.
- 43:00 – 54:00
Hot CPI Print, Fed Dilemma, And Political Fallout For Biden
The hosts dive into March’s hotter-than-expected inflation data, its implications for Fed policy, and the 2024 election. They argue that the three-month streak of upside surprises has killed the narrative of a smooth glide to 2% and discuss how persistent inflation will squeeze consumers, destabilize real estate, and end Biden’s hope for a ‘Fed put’.
- •March CPI at 3.5% YoY marks the third consecutive overshoot; prior ‘seasonal quirk’ excuses no longer hold.
- •Markets and the Fed’s December dot plot expected three 2024 cuts; now cuts may be pushed to late 2024 or vanish entirely.
- •Sacks notes Larry Summers warned in 2021 that Biden’s $2T American Rescue Plan was unnecessarily stimulative; inflation spiked to 5% by that summer, validating his concerns.
- •Summers recently argued that if borrowing costs are fully accounted for, peak ‘effective’ inflation was closer to 18%, explaining why consumers feel far worse than headline data imply.
- •Higher-for-longer rates hurt mortgage affordability, car loans, and personal credit, and may trigger corrections in housing and other rate-sensitive assets.
- 54:00 – 1:05:00
Energy, OPEC+, And Structural Pressures Keeping Inflation Elevated
Chamath presents charts showing rising oil prices and OPEC+ production cuts that more than offset increased U.S. output. He frames these moves as geopolitical ‘votes’ against U.S. policy that feed back into domestic inflation and rate policy, while also highlighting liquidity contraction via reverse repos.
- •Oil prices have rebounded in 2024, reversing much of last year’s relief and directly feeding into broader inflation through energy, transport, and industrial costs.
- •The U.S. has tried to flood the market with more oil, but OPEC+ has systematically cut production by greater volumes, raising global prices.
- •Chamath interprets OPEC+ behavior as a strategic choice by producer nations to oppose U.S. economic and foreign policy, recognizing that high energy prices hurt incumbents electorally.
- •He notes plummeting reverse repo balances as a sign of liquidity being drawn out of the system, another constraint on easy-money hopes.
- •Taken together, these factors make persistent inflation and potential rate hikes increasingly plausible heading into the election.
- 1:05:00 – 1:16:00
Debt Spiral Dynamics: Banks, CRE, And The End Of Easy Money
The conversation shifts to systemic knock-on effects of higher rates: stress in commercial real estate, regional banks, and the federal balance sheet. The hosts describe ‘extend and pretend’ strategies in CRE, ballooning debt service costs in Washington, and the possibility that the 40-year era of falling rates is over.
- •Commercial real estate sponsors who levered up at low rates and high valuations now can’t refinance; banks are capitalizing unpaid interest and extending maturities (“extend and pretend”).
- •If cuts don’t arrive soon, more sponsors will abandon properties and banks will have to foreclose and liquidate at distressed prices; a St. Louis office building dropped from $200M to a few million in value.
- •On the sovereign side, $7.6T of U.S. debt is rolling from ~2% to ~5%, adding over $200B/year in interest costs just from that tranche.
- •Annual interest already exceeds $1T and may hit $1.6T, surpassing defense spending while deficits remain above $2T; the U.S. is essentially borrowing to pay interest.
- •A long Fed funds chart suggests the Volcker-era to COVID-era downtrend in rates may be reversing structurally, removing a multi-decade tailwind to equity and bond valuations.
- 1:16:00 – 1:26:00
Biden’s Political Response: Stimulus, Vote Buying, And Inflation Feedback
Friedberg asks how the Biden administration will respond politically to persistent inflation in an election year. The group anticipates further vote-buying attempts—like expanded student loan forgiveness—that may provide short-term relief but exacerbate inflationary pressures, underscoring the administration’s bind.
- •Biden publicly maintains that rate cuts are still coming, but timing is slipping; his credibility on economic management is at risk.
- •The administration has already tried large student loan forgiveness plans; after the Supreme Court blocked one, Biden announced a new version targeting ~30M borrowers.
- •Sacks argues these measures are effectively attempts to buy votes but come late and on top of years of heavy spending: ARP, IRA, CHIPS Act, infrastructure, and the nearly passed Build Back Better.
- •Chamath warns that further fiscal stimulus in this environment risks creating more inflation, not less, worsening the macro backdrop even as the White House tries to help voters.
- •They also flag the risk of political ‘wag-the-dog’ distractions (e.g., foreign crises) when domestic economic levers are constrained and unpopular.
- 1:26:00 – 1:37:00
Adam Schiff’s AI Disclosure Bill And The Fair Use Fight
The hosts analyze Rep. Adam Schiff’s Generative AI Copyright Disclosure Act, which would force AI developers to register all copyrighted training data. They question both its timing and motives, suggesting it serves Hollywood studios and big tech via regulatory capture, and argue that courts must first settle core fair use questions.
- •The bill defines consumer-facing generative AI and would require developers to submit lists of all copyrighted training data to a federal registry.
- •Sacks views it as premature: fair use boundaries for AI training versus human reading haven’t been litigated yet, and entrepreneurial solutions (e.g., rights marketplaces) are emerging.
- •He frames Schiff as a powerful California fundraiser likely acting on behalf of Hollywood studios and major rights-holders, with quiet assent from big tech firms that can easily comply.
- •Chamath notes reports that OpenAI transcribed over a million hours of YouTube to train GPT‑4; if Google does not sue, he argues, that signals that large-scale copyright violations for training will effectively be tolerated.
- •They predict messy, multi-year litigation across circuits, ending in a Supreme Court decision followed by more tailored legislation.
- 1:37:00 – 1:54:00
What Counts As Infringement In AI? Training vs. Output
Friedberg and Chamath clash over whether copyright enforcement in AI should focus on training data or on generated outputs. Using Ed Sheeran’s court victory and Oasis–Beatles analogies, Friedberg argues that learning from copyrighted works should be permitted, while Chamath insists that without enforceable rights upstream, creators will lose all economic leverage.
- •Friedberg explains that models encode statistical patterns, not raw media; like musicians studying others, they synthesize influences into new works.
- •He proposes that infringement should be judged by measurable similarity of outputs to protected works, not by which inputs were seen in training.
- •Chamath counters this creates an impossible standard: how much similarity (5%, 14%, 78%) is too much? And no individual artist can afford to litigate against trillion‑dollar firms.
- •He argues the decisive signal will be whether giants like Google aggressively defend their own IP (e.g., YouTube) against other AI giants; silence effectively devalues copyright for everyone.
- •Sacks outlines three probable outcomes: (1) broad fair use for training, with only outputs policed; (2) harsh restrictions that entrench incumbents and crush startups; or (3) a rights clearinghouse where data licenses are standardized and monetized.
- 1:54:00 – 2:06:00
AI Music Breakthroughs: Udio And The Coming Content Flood
Friedberg showcases Udio, a state-of-the-art AI music generator capable of producing rich, emotionally compelling songs from simple prompts. The demo prompts—from ‘Dune: The Broadway Musical’ to Irish folk and dance pop—illustrate how far AI composition has come and hint at how quickly user time might shift from traditional media to interactive, personalized generation.
- •Udio, founded by ex-DeepMind researchers, can generate full songs with vocals, harmonies, instrumentation, and genre style from short text prompts.
- •Examples like a fictional Dune musical song show multi-layered arrangements, coherent lyrics, and genre-faithful production quality.
- •Early adopters are already spending hours iterating on prompts and listening to their own AI-created tracks, effectively becoming co-creators rather than passive consumers.
- •Such tools will intensify legal and economic pressure on the music industry as they blur lines between derivative works, parody, and original compositions.
- •The hosts caution legislators not to preemptively clamp down on these tools before their societal and creative potential is better understood.
- 2:06:00 – 2:18:00
Drones, Asymmetric Warfare, And The Autonomy Arms Race
The discussion pivots to modern warfare, where cheap drones are reshaping battlefields in Ukraine and the Red Sea. Sacks and Chamath describe how ubiquitous FPV drones, EW, and cost asymmetries are eroding the dominance of legacy platforms like aircraft carriers, while opening a huge space for innovation in autonomous systems, surveillance, and counter-drone defenses.
- •On the Ukraine front, both sides deploy swarms of FPV drones, making it dangerous to leave trenches; Russia’s scale and superior EW capabilities currently give it an advantage.
- •Houthis in the Red Sea have used inexpensive drones and missiles to threaten commercial shipping, forcing the U.S. to respond with extremely costly interceptors, a lopsided economic equation.
- •A senior U.S. military view (relayed secondhand) suggests aircraft carriers are becoming obsolete in the face of long-range, precision, and swarming threats.
- •Chamath has long invested in non-weaponized autonomous systems like Saildrone, which provide persistent maritime surveillance and data collection, already engaged by the U.S. Navy.
- •He argues unmanned systems will eventually lower costs, reduce human casualties, and yield richer situational awareness, but they require long-term capital and deep integration with the Pentagon.
- 2:18:00
Industrial Dependencies, Counter-Drone Tech, And Silicon Valley’s Moral Dilemma
Friedberg highlights America’s dependence on China for lithium-ion batteries as a strategic risk in a drone-dominated future. Sacks describes his investment in an automated gun turret for drone defense, and the trio wrestle with the moral tension of funding weapons-adjacent startups while wanting to avoid becoming part of the war lobby.
- •Around 79% of global lithium-ion battery production is in China; the U.S. has only ~6%, creating a bottleneck for scaling drone fleets if warfare becomes highly drone-centric.
- •Beyond drones themselves, a new category of defense tech is emerging: EMP systems, EW, and kinetic counter-drone solutions to neutralize swarms.
- •Sacks invested in Allen Control Systems’ “Bullfrog,” an automated M240 gun turret that uses computer vision to detect and shoot down drones, envisioning use on convoys and critical assets.
- •Both Sacks and Chamath are wary of investing in offensive weapons platforms; they draw a distinction between defensive systems and becoming stakeholders who profit from war frequency.
- •Chamath presents a DoD chart showing U.S. military enlistment at multi-decade lows, arguing this demographic reality forces a shift toward automation and unmanned systems regardless of moral discomfort.
- •They close by noting that many in Silicon Valley are ideologically opposed to defense work, but that posture may leave the U.S. technologically exposed while adversaries scale up their own drone and autonomy programs.
