All-In PodcastOpenAI vs Anthropic IPOs, Anthropic $3T, Zuck's Price War, China Ends Open Source?, Trump Accounts
CHAPTERS
- 0:00 – 2:54
Brad Gerstner joins as co-host; quick catch-up from Paris, Geneva, and DC
Jason opens the episode by introducing Brad Gerstner filling in for Friedberg, with light banter about travel and summer routines. Chamath shares a recent Geneva trip and participation in a UN AI commission, setting up later debates about open source and national AI strategies.
- •Brad fills in for Friedberg; casual opening banter and travel updates
- •Jason records interviews in Paris for the All-In feed
- •Chamath describes Geneva dinner and UN AI commission involvement
- •Early tease: open source and sovereign AI will be discussed later
- 2:54 – 4:26
Why trillion-dollar IPOs matter: SpaceX as the new playbook
The group frames a new era of massive IPOs, using SpaceX’s recent listing as a reference point for how mega-offerings can be structured and priced. They discuss float, lockups, volatility, and how the public markets absorb issues at unprecedented scale.
- •SpaceX IPO recap: pricing, post-IPO trading range, and market cap context
- •Mega-IPO mechanics: float size, volatility expectations, and lockup design
- •Why the market hasn’t seen offerings of this magnitude before
- •SpaceX seen as a blueprint for future trillion-dollar AI IPOs
- 4:26 – 8:43
Will Anthropic and OpenAI go public soon? Appetite, burn rates, and timing
Jason presses for probability and timelines on Anthropic and OpenAI IPOs, while Chamath and Brad explain what ultimately drives the decision: public market demand, profitability, and operational complexity. They contrast OpenAI’s higher burn and restructuring complexity with Anthropic’s perceived profitability and momentum.
- •Market-clearing price depends on public market absorption and scale
- •OpenAI: higher burn, consumer-heavy mix, and structural complexity
- •Anthropic: rumored profitability and strong enterprise traction
- •Brad expects high likelihood of near-term IPOs; Anthropic may go first
- 8:43 – 13:21
SpaceX index-inclusion controversy and what it teaches future IPOs
Brad explains why early index inclusion created concern and how rules were adjusted to mitigate forcing buyers into peak volatility. The discussion outlines the tension between protecting passive investors and acknowledging that certain companies are too large to exclude.
- •Concern: IPOs often see large drawdowns shortly after listing
- •Risk of jamming a volatile new stock into passive indices at the top
- •Counterargument: a company can be ‘too important’ to exclude from indices
- •Outcome: modified approach created a workable precedent
- 13:21 – 15:26
How OpenAI vs Anthropic may be positioned: consumer brand vs enterprise ROI
Jason frames OpenAI as the dominant consumer brand (ChatGPT) and Anthropic as an enterprise favorite (Claude), asking whether OpenAI misallocated focus. Chamath argues enterprise spend will eventually demand hard ROI, while consumer revenue can be more resilient due to breadth of buyers and lower scrutiny per user.
- •Brand positioning: consumer (ChatGPT) vs enterprise (Claude)
- •Enterprise spend eventually faces ROI/CFO scrutiny
- •Consumer scale can ‘inoculate’ revenue from ROI debates
- •Risk: token costs rising faster than measured productivity gains
- 15:26 – 17:31
Token spend shock: costs doubling, productivity flattening, and the ROI reckoning
Chamath shares a concrete internal data point: token costs doubling every 45 days with only marginal productivity lift, suggesting a coming enterprise ‘reckoning.’ The group debates whether the industry can outrun this with new breakthroughs or whether CFO-led cost control will reshape demand.
- •Anecdote: token costs doubling rapidly while gains look capped
- •Why incremental improvements may require disproportionately more tokens
- •Prediction: companies will be forced to justify spend within a few years
- •Strategic implication: ‘get out now’ logic for IPO timing
- 17:31 – 25:02
CTO playbooks emerge: Uber’s agentic pods and DoorDash’s model routing
Jason points to public CTO posts showing enterprises getting sophisticated about deployment: embedding engineers into functions, creating agentic pods, and routing tasks across models. The conversation highlights a shift from ad-hoc ‘slot machine’ usage to measurable programs targeting real operational savings.
- •Uber: broad AI tool adoption and structured expansion beyond engineering
- •Agentic pods: forward-deployed engineers partnering with operations teams
- •DoorDash: routing hard tasks to frontier models, easier work to cheaper models
- •Benchmarks and governance become central as CFOs demand accountability
- 25:02 – 36:47
Jevons paradox in practice: cheaper inference drives more usage, not less
Jason describes personal experimentation where token costs dropped ~95%, causing him to run more jobs more frequently and decompose tasks into multiple agents. Brad ties this to historical token price declines and the core debate: whether open/cheap models will commoditize frontier labs or expand overall usage.
- •Lower inference prices can increase total consumption (Jevons paradox)
- •Agent workflows scale with price drops: daily → hourly → multi-agent pipelines
- •Token prices have seen repeated ~90% declines over multiple years
- •Central question: commodity tokens rise, but frontier revenue share persists
- 36:47 – 42:35
Frontier vs open source: why premium models still capture wallet share
Brad argues that while cheap models can handle many tasks, high-stakes, long-running agentic work favors reliability and capability, making price differences less relevant versus labor costs. Sacks adds that most enterprises lack the technical ability to build routing/middleware and portable memory/context layers, so convenience keeps frontier labs dominant.
- •Premium workloads: failures are costly, so ‘95% as good’ may not be enough
- •Inference cost vs labor replacement: token price gaps can be immaterial
- •Enterprise limitation: few can implement routing, memory portability, and harnesses
- •Observed outcome: closed/frontier models’ share of spend remains very high
- 42:35 – 48:22
Zuckerberg’s price-war move: Meta’s low-cost agentic model and API ambitions
Chamath and Jason interpret Meta’s new release as a strategic pivot toward competing on cost rather than purely open-source strategy. They discuss the practicality barriers (enterprise distribution, past misfires) but note the competitive pressure a credible ‘same quality at a fraction of cost’ claim can create.
- •Meta releases a ‘strong agentic encoding model’ at low price
- •Shift in competitive vector: cost-driven price war
- •Meta’s goal expands into token/API distribution, not just model release
- •Skepticism remains: enterprise adoption and execution challenges
- 48:22 – 54:10
Sovereign AI expands: UAE, Saudi, Japan—and why countries don’t want closed US models
The panel discusses sovereign AI strategies emerging worldwide and why many governments prefer owning a full stack rather than depending on closed American frontier labs. Examples include Gulf state models and Japan’s investment pushing toward robotics/physical AI.
- •Countries pursue ‘sovereign stacks’ to reduce technical and political dependence
- •UAE (Falcon), Saudi initiatives, and Japan’s consortium investment
- •Motivation: control, security, and local optimization (language/regulation)
- •Coexistence thesis: open source, sovereign models, and frontier labs all persist
- 54:10 – 1:01:44
China may restrict overseas access to top models: export controls and security fears
Reuters reports prompt a discussion that Chinese regulators may limit foreign access to leading Chinese models and treat AI leaks as national security offenses. The group analyzes motives: preventing exploitation, limiting foreign funding, and the strategic logic of being open to catch up and closing once competitive.
- •China considering restrictions on overseas access to top models
- •Concerns: vulnerabilities, foreign exploitation, and national security leaks
- •Strategy: go open while behind; go closed once near the frontier
- •US view: staying ahead of China is a bipartisan priority; distillation crackdown expected
- 1:01:44 – 1:03:01
The real bottleneck: energy and infrastructure constraints on AI progress
Chamath argues that the limiting factor may become electricity generation rather than software or chips, citing large projected load shortfalls through 2050. They connect energy vulnerability to geopolitics (Taiwan/LNG) and the urgency of scaling nuclear, solar, storage, and permitting reform.
- •Projected US load growth implies massive energy shortfall by 2050
- •AI inference and data centers intensify demand for electrons
- •Taiwan blockade risk: limited energy reserves amplify geopolitical fragility
- •Policy implication: accelerate nuclear, renewables, storage, and grid buildout
- 1:03:01 – 1:09:20
Trump Accounts launch: what they are, how they work, and why Brad went to DC
Brad recounts the DC event and rollout of ‘Trump Accounts’ (Invest America Act), including app launch momentum and high-profile participation. He explains the core design: a free, lifetime account invested in the S&P 500 with contribution pathways for families, employers, states, and philanthropists.
- •Accounts: $1,000 at birth concept; S&P 500 investment; free lifetime structure
- •Launch results: rapid account creation, significant deposits, viral sharing via QR/Apple Pay
- •DC milestone: joint NYSE/Nasdaq bell event from the Oval Office
- •Distribution mechanics: targeting by zip code/age; employer and philanthropic contributions
- 1:09:20 – 1:13:48
Scaling participation and politics: auto-creating accounts, bipartisan support, and ‘TDS’ backlash
The group debates the political branding (‘Trump Accounts’) and whether opposition could reduce adoption, while Brad claims parents are prioritizing children’s financial futures. They discuss the push to auto-create accounts at scale, bipartisan endorsements, and contrasting worldviews on state control vs private ownership.
- •Goal: expand to 50–70M accounts rapidly, potentially via auto-creation
- •Backlash risk: some oppose due to the name; Brad says adoption is broad-based
- •Bipartisan support cited alongside criticism from anti-capitalist factions
- •Core ideological divide: government redistribution vs private ownership and compounding
- 1:13:48 – 1:42:04
Why it’s powerful: tax advantages, employer contributions, philanthropy at scale, and compounding
Sacks walks through the planning/CPA perspective: tax-free compounding, annual contribution limits, employer contributions, and rollover strategies that could create substantial retirement security. Brad positions the platform as a new national compounding engine that bypasses NGO overhead and reconnects families to equity ownership.
- •Tax-free compounding until 18; partial access; rollover to IRA/Roth strategies
- •Employer contributions (up to $2,500) framed as a major no-brainer benefit
- •Philanthropy becomes direct, scalable, and ‘middleman-free’ via mass accounts
- •Compounding advantage: capturing the ‘first third of the hill’ (ages 0–25)
