All-In Podcast

SpaceX’s $2T Case, Nvidia’s Shock Selloff, America Turns on AI, Trump Pulls AI Order, Bond Crisis?

Key Themes

AI talent shiftrecursive self-improvementAI public backlashAI regulationSpaceX valuationNvidia earningsmacro stressU.S.-China tech race

1h 42mMay 22, 2026

Summary

A wide-ranging AI, policy, and macro debate tying together Anthropic, SpaceX, Nvidia, and U.S.-China tech competition

This episode moves from the latest AI talent and product developments to a broader argument about AI’s economic impact, public backlash, regulation, and geopolitical stakes. The panel makes a bullish case for recursive AI improvement and practical AI utility, then pivots to American skepticism toward AI, the politics of regulation, and workforce disruption. Later sections unpack a highly ambitious SpaceX valuation framework, debate Nvidia’s earnings and the chip cycle, and close with a discussion of China-related diplomacy and chip exports. Across the full conversation, the recurring thread is that AI is no longer just a model story; it is now a policy, infrastructure, labor, and national-security story too.

AI is increasingly a systems story, not just a model story

The discussion repeatedly shifts from raw benchmark wins to practical deployment, product integration, and the possibility that models will help build better models. That framing matters because it suggests progress will come from end-to-end workflows, tooling, and iteration speed rather than headline releases alone.

Public resistance to new technology often comes from uneven early benefits

The panel’s explanation for AI backlash is that gains appear concentrated among a small group first, while the wider public sees disruption, uncertainty, and headlines about layoffs. That dynamic is useful beyond AI because it helps explain why transformative technologies often become politically unpopular before they feel broadly helpful.

Regulation debates are really debates about control, speed, and permanence

The AI policy section shows how quickly a technical safety conversation becomes a broader struggle over who sets the rules and whether emergency powers linger after the moment has passed. That matters because it highlights the tradeoff between targeted safeguards and creating lasting institutions that may outlive their original justification.

Infrastructure ownership can be a powerful strategic moat

The SpaceX section argues that owning the physical layer—launch, connectivity, and possibly compute—creates leverage that software-only players cannot easily copy. More generally, it shows how control over scarce infrastructure can compound into pricing power, speed, and optionality.

Big technological opportunities often arrive as combinations of software and tooling

The conversation around Cursor, Grok Build, and orbital compute emphasizes that model quality alone is not enough. Adoption and performance improve when models are paired with strong harnesses, domain data, and the right operational layer, which is a useful lens for evaluating many modern platforms.

Macro stress often shows up first in rates, energy, and leverage

The market segment treats rising oil, inflation expectations, and bond yields as warning lights that can foreshadow broader stress. Even without a crisis, these indicators are useful because they often reveal where financing conditions, carry trades, or speculative excess may be most vulnerable.

Geopolitical dialogue can matter even when public outcomes look modest

The China discussion suggests that the visible headline result of a diplomatic visit may understate the value of quieter negotiations. That is a useful reminder that in complex international settings, stability can come from incremental coordination rather than dramatic announcements.

Select any chapter text to Deep Dive with AI

01Karpathy joins Anthropic and the case for recursive AI self-improvement

The episode opens with introductions and quickly turns to Andrej Karpathy joining Anthropic. The speakers frame him as a rare technical talent whose move matters for the next phase of AI, including systems that improve themselves. The discussion broadens from model hype to practical user-facing value, with examples ranging from coding tools to browser-integrated AI.

Gavin Baker is introduced as the guest and judge for the show.

Karpathy’s background at Tesla and OpenAI is used to underscore his technical credibility.

Anthropic is presented as a serious AI company with strong growth and economics.

The panel argues AI progress may increasingly come from systems helping build better systems.

Smaller, specialized, and vertically integrated AI models are discussed as a likely direction.

Google’s Gemini Nano in Chrome is cited as an example of everyday AI utility.

The conversation emphasizes concrete user value over benchmark chatter.

02Why Americans Have Turned Against AI

The panel explores why AI is becoming politically and culturally unpopular in the U.S. They argue the backlash comes from fears of job loss, concentrated power, and a technology that feels hard to understand or trust. The discussion also contrasts fear-based messaging with more optimistic examples of AI improving real lives.

The speakers call for stronger advocacy around AI’s positive possibilities.

AI backlash is linked to concerns about inequality and power concentration.

A personal story about a child’s treatment illustrates AI’s life-saving potential.

AI is described as unusually difficult for ordinary people to understand or predict.

The panel argues global competition makes slowing AI unrealistic.

Job displacement debates are framed as needing input from actual workers.

AI is characterized as challenging human-centered worldviews.

03Trump scraps AI executive order amid U.S.-China AI race and tech layoff fears

The conversation shifts to an AI executive order that was reportedly pulled at the last minute, then expands into debate over frontier-model testing and international safeguards. The panel also examines layoffs and company messaging, arguing that public framing of AI-driven cuts is intensifying fear and making the transition feel dystopian.

The executive order was announced and then removed at the last minute.

The speakers want stronger safeguards around dangerous frontier models.

They discuss KYC-like controls and possible U.S.-China coordination.

There is skepticism about giving regulators broad, permanent AI powers.

Layoffs at major companies are used to illustrate AI labor anxiety.

Corporate messaging around AI-related cuts is criticized as counterproductive.

04SpaceX S-1 teardown: businesses, valuation, and orbital compute

The panel breaks down SpaceX as a three-part business spanning Starlink, launch, and AI/compute. Starlink is portrayed as the profit engine, while the broader pitch rests on infrastructure control, AI customer demand, and longer-term optionality around orbital compute. The chapter builds toward a very bullish valuation framework and even considers a possible Tesla/SpaceX combination.

SpaceX is described as a three-business platform.

Starlink is the main cash generator, while launch and AI/compute are growth engines.

A major Anthropic compute deal is highlighted as evidence of demand.

Cursor and Grok are used to show that tooling and harnesses matter alongside models.

Orbital compute is treated as plausible, not science fiction.

The speakers argue SpaceX’s speed and cost structure create a durable moat.

Rapid Starship reusability is framed as the key technical unlock.

05Nvidia earnings, chip valuation debate, and market macro warning signs

Nvidia’s earnings are presented as exceptional, but the panel questions how to interpret valuations across the AI stack. They argue that chip-demand narratives, useful life assumptions, and benchmarking gaps are often misread. The back half turns into a macro warning about oil, inflation, rates, and speculative excess, paired with a preference for concentration in a small number of long-term winners.

Nvidia’s quarter is framed as extraordinarily strong.

The panel debates whether the market is properly pricing AI infrastructure winners.

Benchmarking and competitive comparisons are treated as incomplete.

GPU amortization and financing assumptions are questioned and defended.

Oil, inflation, and bond yields are all flagged as rising risks.

Global rates and Japan are discussed as possible sources of stress.

The speakers advocate concentration over speculation.

06China visit: optics vs. behind-the-scenes dealmaking

The final chapter questions whether the China trip produced real policy progress or mostly symbolic optics. The panel suggests the more meaningful work may have happened privately, with implications for chips, trade, and geopolitical stability. They end by arguing that continued U.S.-China dialogue may help reduce the risk of a broader conflict.

The public outcome of the China trip appears limited.

A few concrete commercial items were mentioned, including chips and aircraft.

The speakers suspect the real value may have been behind closed doors.

Nvidia chip exports to China are presented as potentially stabilizing.

U.S.-China dialogue is framed as useful for reducing conflict risk.

Broader geopolitical bargaining is discussed as part of the background.