Can the AI Industry Regulate Itself? Stripe Wants PayPal, China Catches Up, NY Bans Datacenters

Summary

Theses (episode spine)

• Demis Hassabis’s “FINRA for AI” self-regulatory organization (SRO) proposal is broadly popular with the hosts; Sacks conditionally backs it only if it includes five safeguards (broad industry representation to prevent regulatory capture, frontier-models-only scope, catastrophic-risk-only focus, voluntary-first, and it must substitute rather than add to regulation), warning it could otherwise become “the opening bid” for heavier regulation pushed by Anthropic.
• Chamath argues Anthropic is running a “sophisticated regulatory capture strategy” via a deliberate state-by-state push to ratchet up AI guardrails (citing a Politico report), using fear to lock out smaller competitors and open-source models; Sacks agrees and separately flags Anthropic funding anti-data-center advocacy groups.
• Stripe, Advent, and Block are jointly bidding roughly $53B ($60/share) for PayPal; Chamath sees this as a path to building a real challenger to Visa/Mastercard by combining Stripe’s merchant rails, Block’s point-of-sale/Cash App, and PayPal’s 400M+ consumer accounts, while Sacks calls PayPal’s core product a stagnant “legacy” asset.
• Apple has sued OpenAI alleging theft of trade secrets tied to its hardware device (involving former Apple hardware chief Tang Tan, now OpenAI’s chief hardware officer, and claims of 400+ poached Apple employees); Chamath and Sacks decline to judge the facts but agree departing employees should take nothing but what’s in their heads.
• xAI’s Grok Build coding tool was found uploading developers’ full codebases (including secrets) to xAI servers despite privacy claims otherwise; Chamath argues this proves “zero data retention” promises from AI labs can’t be fully trusted and enterprises need an independent third-party layer to manage exposure.
• Token pricing is wildly divergent across providers (Chamath cites ~$56/million tokens for Claude vs. ~$0.50-$1.50 for Grok/open/Chinese models); with RAMP reporting 21x growth in customer token spend, Chamath warns unmanaged AI opex could eventually cause a public company to miss earnings.
• Jason is bullish on Apple stock, betting that the upcoming M7 Ultra chip’s support for up to 1.5TB of RAM will let frontier-level AI models run locally on Mac Studios, undercutting cloud AI providers.
• Chamath details a severe projected US energy deficit (2.5x California’s total consumption by 2050) and describes Elon Musk’s “behind the meter” strategy of using mobile generators to bypass grid bottlenecks and clean-air permitting for data centers like Colossus in Memphis.
• Sacks rebuts NY Governor Kathy Hochul’s statewide hyperscale data center moratorium point-by-point (power, land, water, noise), calling the water-usage concern “largely a hoax” and warning a moratorium effectively costs ~5 years of buildout even if later reversed; Freeberg and Sacks cite evidence of alleged Chinese/Russian-linked influence campaigns manufacturing anti-data-center sentiment, drawing a parallel to past anti-GMO disinformation.
• Sacks and Jason argue the current AI “moral panic” is premature, pointing to Dario Amodei’s unfulfilled 2025 prediction that 50% of entry-level knowledge-worker jobs would vanish within 1-5 years, and warn heavy regulation risks ceding the AI race to China given Kimi K3’s rapid progress.

$PYPL (PayPal)

Convergence / divergence: All three treat the Stripe/Advent (show also names Block) takeover bid as the dominant near-term catalyst, but they diverge sharply on what it means: Jason reports the deal mechanics neutrally, Chamath is bullish that the price goes higher (possibly via a competing bid), while Sacks is skeptical of the underlying business even as he concedes its assets have strategic value to an acquirer.

Speaker calls:

• Jason (neutral, unspecified timeframe): Reports that Stripe, Advent, and Block are jointly bidding about $60 per share (roughly $53B) for PayPal, a modest premium that the market expects could rise toward $70 per share.
• Chamath (bullish, unspecified timeframe, target 10-15% above current offer): Believes the current bid is not the final clearing price and expects it to go another 10-15% higher, and speculates Elon Musk could submit a competing bid given his PayPal co-founder history and available capital.
• Sacks (bearish): Says PayPal’s core product is a stagnant, “long in the tooth” legacy product growing only about 7% a year because founders were pushed out after the 2002 eBay acquisition, though its 439 million consumer accounts and Braintree business remain valuable assets for a buyer.

Cross-check:

• Price: $56.40 as of 2026-07-18 (P/E 10.64, mkt cap $49.9B). Next earnings: 2026-07-28.
• Recent headlines worth knowing: Stripe and Advent International made a joint $53B ($60.50/share, 28% premium) takeover offer on July 15, 2026, sending shares up 17% on record volume; PayPal board reportedly meeting as soon as July 20 to discuss the offer, working with Goldman Sachs and Evercore as advisers.
• ⚠️ Inconsistencies: Cross-checked spot price ($56.40) sits below the reported $60.50/share offer price cited on the show — the quote may lag the deal news or reflect a pullback since the July 15 spike; treat the exact current price with caution. Show also referenced Block as a joint bidder alongside Stripe/Advent, which is not corroborated by mainstream coverage (only Stripe and Advent International are named).

$AAPL (Apple)

Convergence / divergence: Only Jason stakes out an explicit call on Apple; no other host offers a counter-view in the transcript, so there is nothing to converge or diverge against.

Speaker calls:

• Jason (bullish, unspecified timeframe): Says Apple is “a screaming buy right now” (not financial advice) because the upcoming M7 Ultra chip’s support for up to 1.5TB of RAM will let frontier-level AI models run locally on Mac Studios, undercutting cloud AI subscription costs.

Cross-check:

• Price: $333.69 as of 2026-07-18 (P/E 37.99 trailing / 34.39 forward, mkt cap $4.901T). Next earnings: 2026-07-30.
• Recent headlines worth knowing: Apple Intelligence approved for launch in China via Alibaba Qwen integration (July 15-16); Apple sued OpenAI for trade secret theft (July 11); new Siri confirmed to use Google Gemini instead of ChatGPT; shares up ~16% since bottoming June 25, adding ~$650B market value.
• ⚠️ Inconsistencies: None flagged — no public confirmation yet of an M7 Ultra chip supporting 1.5TB RAM; that detail is a forward-looking claim from the show, not something to verify against current specs.

Topics discussed

AI Self-Regulation: Demis Hassabis’s “FINRA for AI” (SRO) Proposal

Summary: Demis Hassabis proposed a self-regulatory organization (SRO) modeled on FINRA for the AI industry: voluntary at first then mandatory, federally overseen but industry-funded, with frontier models submitted for review 30 days before release. The proposal drew public support from Elon Musk, Sam Altman, Jack Clark, Sundar Pichai and others.

Speaker views:

• Jason: Claims he previously called for industry self-certification (comparing it to the MPAA and video game ratings systems) and endorses the concept.
• Sacks: Says he could get behind an SRO as far better than a new government agency, which he compares to a “DMV for AI” or Dario Amodei’s proposed “FAA for AI” (warning FAA-style certification takes years). Lists five required conditions: broad industry representation to avoid regulatory capture, frontier-models-only scope, catastrophic-risk-only focus (cyber/CBRN, not disinformation), voluntary-first, and it must substitute for, not add to, other regulation.
• Chamath: Supports the proposal, saying it heads off “a torrent of money” trying to create regulatory capture from both sides of the political aisle, while preserving DOJ/Commerce oversight.

Potential impact: Sacks warns that without his five conditions, the SRO could become “the opening bid” for Anthropic-driven regulatory capture and progressively heavier AI regulation.

Anthropic’s State-by-State Regulatory Strategy

Summary: Chamath revisits his year-old claim that Anthropic runs a “sophisticated regulatory capture strategy based on fear-mongering,” citing a Politico article describing Anthropic’s push to ratchet up AI guardrails state by state (building on California’s SB53) rather than settle on one national framework.

Speaker views:

• Chamath: Says evidence now supports his year-old claim; Anthropic, now a trillion-dollar company, is deliberately fragmenting state regulation to keep raising the bar, pulling up the ladder behind it, and argues this is damaging the startup ecosystem.
• Sacks: Agrees, adding that once companies ask government for regulation “there’s hardly anyone in government who will say no,” and warns that conceding ground now just invites government to “ratchet up the pressure” for more control later. Also flags Anthropic funding anti-data-center advocacy groups like “Public First,” to which Dario Amodei gave a “first seven figure contribution.”

Potential impact: Chamath argues this state-by-state regulatory frenzy is damaging the US startup ecosystem broadly.

Stripe/Block/Advent Bid for PayPal and the Visa/Mastercard Threat

Summary: Stripe and private equity firm Advent, joined by Block (contributing $17B in equity), are jointly bidding about $53B ($60/share) for PayPal. Hosts frame it as part of a broader trend of operators buying undermanaged “digital native” businesses to modernize them, with a potential path to a combined Visa/Mastercard challenger.

Speaker views:

• Sacks: Says antitrust regulators would likely have blocked this deal two years ago, but reframing the relevant market (the Visa/Mastercard duopoly rather than merchant payment APIs) makes it pro-competitive and consumer-friendly.
• Freeberg: Sees this as part of a broader wave of operators (citing Ryan Cohen’s eBay bid and Bending Spoons’ rollups) acquiring undermanaged legacy digital businesses and reviving them with better operations and AI; predicts eBay/PayPal are “the beginning of a wave of mega deals.”
• Chamath: Argues combining Stripe’s merchant rails, Block’s point-of-sale/Cash App, and PayPal’s consumer accounts could enable low-cost “on-us” transactions that bypass card networks, creating a real challenger to Visa and Mastercard.

Potential impact: Hosts argue the combined entity would likely lower merchant fees and consumer prices, which paradoxically makes the deal more defensible on antitrust grounds now than it would have been two years ago.

Apple v. OpenAI Trade Secrets Lawsuit

Summary: Apple filed a 41-page lawsuit against OpenAI on July 10, alleging OpenAI’s hardware team (including former Apple VP of iPhone design Tang Tan, now OpenAI’s chief hardware officer) stole Apple trade secrets for its consumer hardware device, and that OpenAI poached 400+ Apple employees, with alleged messages referencing accessing Apple’s network storage after leaving.

Speaker views:

• Chamath: Notes Apple hasn’t been litigious in 25 years, so something must have “really really upset them”; declines to speculate on guilt but says “nobody should be stealing things from their former employer.”
• Sacks: Declines to opine on the specific facts since the matter is alleged and unadjudicated, but states the simple rule that departing employees should take nothing with them except what’s in their head.

xAI Grok Build Data Leak

Summary: xAI’s Grok Build (a coding tool inside Cursor, powered by Grok 4.5) was found to be uploading developers’ entire codebases — including passwords and API keys — to xAI servers despite a privacy setting meant to prevent this. xAI disabled the upload on July 13, Elon Musk said all uploaded data was deleted, and xAI then open-sourced the Grok Build harness.

Speaker views:

• Chamath: Connects this to his recent CNBC comments and Alex Karp’s remarks that privacy in AI is “very fragile and brittle” even at trustworthy companies; argues “zero data retention” promises can’t guarantee safety since labs can have trap doors they don’t even know about, and enterprises need an independent third-party layer (referencing Sacks’ “reverse information paradox” post) to manage exposure.

Potential impact: Chamath argues this pushes toward a “stratified ecosystem” with third-party interfaces to AI models rather than direct enterprise reliance on model providers’ data policies.

AI Token Pricing and Enterprise Cost Risk

Summary: Hosts discuss wide divergence in per-million-token pricing (Chamath cites roughly $56 for Claude vs. about $1.50 for Grok and around $0.50 for Chinese/open models) and RAMP CEO Eric Glyman’s new token-spend-management product, noting RAMP customer token spend has grown 21x over the past year.

Speaker views:

• Chamath: Warns that engineers who aren’t cost-accountable will keep using the most expensive frontier models even when cheaper models suffice, that unmanaged token spend could eventually cause a public company to miss earnings, and that enterprises pay a “double whammy” by overpaying while also feeding proprietary data to closed frontier labs.
• Jason: Shares that building a personal app on the newest Grok model cost about $11 versus much higher costs using other models (he ran out of Claude/“Fable” credits), illustrating the price gap.

Potential impact: Chamath says unmanaged AI opex could eventually cause a public company to miss an earnings quarter.

US Energy Deficit and Data Center Power

Summary: Chamath describes a projected US energy deficit equal to 2.5x California’s total energy consumption by 2050, cites a recent PGM utility auction that received only about 156 megawatts against a roughly 7-8 gigawatt need, and explains Elon Musk’s “behind the meter” strategy (using mobile generator units) to bypass grid bottlenecks and clean-air permitting for Colossus in Memphis.

Speaker views:

• Chamath: Frames this as a severe structural bottleneck for AI buildout, noting roughly 40% of data center projects are currently getting mothballed due to lack of available power, and describes newer compliant solutions like Bloom Energy.

Potential impact: Chamath says the power shortfall could limit AI-driven advances in drug discovery, cancer diagnosis, and healthcare because demand for AI compute may outstrip available energy.

New York’s Data Center Moratorium and Alleged Foreign Influence Campaigns

Summary: NY Governor Kathy Hochul signed the nation’s first statewide moratorium on hyperscale data centers, citing power costs, land use, water depletion, and noise. Sacks rebuts each claim, and Freeberg and Sacks separately cite an OpenAI blog post and Politico coverage on alleged “PRC-linked influence operations” targeting US AI debates, drawing a parallel to past Russian media influence on US anti-GMO sentiment.

Speaker views:

• Sacks: Says utility price increases only occur if data centers draw from the grid without adding power (solvable via behind-the-meter generation); calls the land-use concern overstated given data centers’ economic ROI per acre; calls the water-consumption claim “largely a hoax” given closed-loop recirculation (comparing usage to “two and a half in-n-out burgers”); notes natural gas, the primary power source, is one of the cleanest-burning fuels; and separately argues China has an incentive to slow US data center buildout both domestically and via chip export controls to allied countries.
• Jason: Adds that New York State is roughly 70-80% undeveloped land outside NYC, undercutting the land-use argument, and notes data centers generate significant tax revenue and construction jobs, citing teacher bonuses funded by data center tax revenue in North Dakota.
• Freeberg: Draws a parallel between current anti-data-center sentiment and past anti-GMO sentiment, tying the latter to Russia Today media influence (via Google Trends data) and noting a poll showing over 50% of Americans believe data centers raise water/electricity costs even when centers recycle water and self-generate power.

Potential impact: Sacks warns a moratorium effectively costs about 5 years of lost buildout even if later reversed, and that if China can incentivize US labs to support restrictive regulation, it directly helps China win the AI race given the large cost gap between US and Chinese model pricing.

AI “Moral Panic” and Dario Amodei’s Job-Loss Predictions

Summary: Sacks and Jason argue current AI regulatory anxiety is a premature “moral panic” based on hypothetical future catastrophes rather than observed harms, criticizing Dario Amodei’s year-old prediction that 50% of entry-level knowledge-worker jobs would disappear within 1-5 years, which they say has not materialized.

Speaker views:

• Sacks: Says nothing resembling catastrophic AI harm (cyber risk, job loss) has occurred yet and warns of throwing away America’s free-market innovation advantage over speculative risks, especially as China’s Kimi K3 model closes the capability gap.
• Jason: Jokes that Dario’s alarm might stem from a “panic attack” or could be a savvy strategy (“brand yourself as a safe AI company, ban unsafe AI, profit”), and invites Dario onto the podcast.

Potential impact: Sacks warns overregulation driven by unrealized fears could cause the US to lose its AI lead to China.

Science Corner: AlphaFold-Designed Anti-Aging Enzyme

Summary: Freeberg covers a new paper from Calico (Google’s aging-research arm) and Revel Pharma, which used AlphaFold and directed evolution to design a novel enzyme that breaks down CML, an advanced glycation end-product that drives extracellular matrix aging. Applied to donated elderly human skin, the enzyme eliminated 55% of CML, reversing skin age to roughly that of a 31-year-old.

Speaker views:

• Freeberg: Calls it a “groundbreaking demonstration” combining AlphaFold-based protein discovery with directed evolution and high-throughput screening; says open questions remain about delivery method (cream, injection, or RNA therapy that produces the protein in the body).
• Jason: Predicts the first commercial market will be a cosmetic skin cream, calling it a potential “$2 trillion” market opportunity.

Potential impact: Jason and Freeberg say this could open a large cosmetic/therapeutic market and both frame it as a clearly beneficial, uncontroversial application of AI (AlphaFold).