Podcast, Roundtable
Can the AI Industry Regulate Itself? Stripe Wants PayPal, China Catches Up, NY Bans Datacenters
- Demis Hassabis of DeepMind proposed a US-led international self-regulatory organization (SRO) for AI, modeled after FINRA, featuring:
- Federal oversight but industry funding and management by independent technological experts.
- A requirement for frontier labs to submit models 30 days prior to release.
- Initial voluntary adoption transitioning to mandatory compliance.
- Assessment criteria covering cybersecurity, national security, and biological threats with quarterly benchmark updates.
- Authority to coordinate global development slowdowns if specific risks emerge.
- Major industry leaders including Elon Musk, Sam Altman (OpenAI), Jack Clark (Anthropic), Sundar Pichai (Google), Satya Nadella (Microsoft), Jack Dorsey (Block), and the Carlson brothers have publicly supported the proposal.
- David Freiberg views the SRO model as superior to creating a new government agency, predicting that a regulatory body would devolve into a "DMV for AI" with excessive delays.
- Freiberg outlined five critical conditions for the SRO to be effective:
- Broad Representation: The body must include startups and open-source contributors (e.g., Jensen Huang, Elon Musk, Mark Zuckerberg, Mira Murati) to prevent regulatory capture by the top three labs.
- Scope Limitation: Regulation should apply only to frontier models representing a step-change in state-of-the-art intelligence, not incremental updates.
- Risk Focus: The body must address catastrophic risks (cyber, CBRN) only, explicitly excluding speech regulation or disinformation concerns.
- Voluntary First: The organization must prove efficacy before becoming legally mandatory.
- Substitution: The SRO must replace, not add to, existing government regulatory frameworks.
- Freiberg expressed skepticism regarding Anthropic's motives, suggesting their support is an "opening bid" intended to pave the way for a more stringent "FAA for AI" model.
- Unlike the SRO, an FAA model would require type certification for new aircraft designs, taking 5–9 years, which would stall AI development and cede the race to China.
- Freiberg notes Anthropic is actively funding a strategy to ratchet up state-level regulations in a patchwork manner to create a "ladder-pulling" effect against open source.
- Stripe, Advent Global, and Block have submitted a joint offer to acquire PayPal at approximately $60 per share, representing a small premium and a market cap reduction from its peak of $200 billion.
- Block is contributing $17 billion in equity to the combined deal.
- The potential merger aims to create a vertically integrated payment infrastructure competing directly with Visa and Mastercard by combining Stripe's merchant network (approx. $2T annual volume) and PayPal's consumer base (439 million accounts).
- Zachary Kaufman predicts this deal is the precursor to a wave of acquisitions targeting mature, founder-less digital assets (like eBay) that can be revitalized using AI-driven operational efficiencies.
- Robert argued that private equity and AI-native operators are uniquely positioned to exploit undervalued digital assets that failed to integrate AI, citing similar "roll-up" strategies by Ryan Cohen (eBay) and Bending Spoons.
- Zachary noted that the proposed synergies between Stripe, PayPal, and Block could bypass traditional credit card rails, potentially lowering consumer costs but triggering significant antitrust scrutiny regarding market definition.
- Apple filed a 41-page lawsuit against OpenAI alleging the theft of trade secrets for a new consumer hardware device, led by former Apple VP Tang Tan who is now OpenAI's Chief Hardware Officer.
- Apple claims OpenAI poached over 400 employees and induced the transfer of IP, including instances where candidates brought hardware parts to interviews or accessed network storage.
- Jason Calacanis emphasized that while employment at will allows job mobility, taking proprietary data from a former employer is strictly prohibited regardless of the source of the knowledge.
- SpaceX experienced a data privacy incident with its Grok Build tool, where users' entire codebases (including passwords and API keys) were inadvertently uploaded to servers despite "zero data retention" settings.
- SpaceX addressed the leak by immediately disabling the upload feature and subsequently open-sourcing the Grok Build harness.
- Chamath Palihapitiya highlighted that such "trap doors" necessitate a stratified AI ecosystem where enterprises use independent third-party layers to manage data exposure and ensure sovereignty.
- A critical economic gap in AI costs has emerged, with enterprise models (e.g., Grok 4.5) costing significantly less ($1.50 per million tokens) than frontier competitors like GPT-5 ($56 per million tokens), creating incentives for CFOs to enforce spend controls.
- Eric Blyman of Ramp announced new features to manage token spending, responding to a 21x year-over-year increase in enterprise token consumption.
- Mark German predicted that Apple's upcoming M7 Ultra chip will support up to 1.5 terabytes of RAM, enabling the local execution of frontier models and disrupting the reliance on cloud-based inference.
- Governor Kathy Hochul signed a moratorium on hyperscale data centers in New York, citing concerns over energy usage, land displacement, and pollution.
- Critics argued the moratorium relies on false premises, noting that modern data centers use closed-loop water systems, have high land-use efficiency, and generate significant tax revenue.
- Chamath Palihapitiya and David Freiberg attributed the regulatory push to "foreign influence operations" (potentially from China or Russia) designed to halt US AI infrastructure development and cede the AI race.
- Anthropic is accused of funding groups like "Public First" to oppose data center construction, despite the resulting compute scarcity threatening the company's own revenue growth.
- The US faces a projected energy deficit equivalent to 2.5 California's worth of consumption by 2050, driving a shift toward "behind-the-meter" power generation where data centers build their own supply.
- Elon Musk bypassed clean air permitting hurdles for data centers by utilizing mobile turbines classified as personal use equipment.
- Global interest in data centers is shifting toward energy-rich regions like the Middle East and Australia, though US allies are also facing export control friction on chip technology.
- Science Corner Update: A breakthrough study by Calico and Revel Pharma, utilizing AlphaFold and directed evolution, created a novel enzyme capable of degrading Advanced Glycation End-products (CML).
- The enzyme reduced CML accumulation in human skin samples by 55%, effectively reversing skin age from 70+ to 31 years old in vitro.
- The discovery validates AI-driven protein design for therapeutic applications, with initial commercial markets expected to be cosmetic and dermatological.