Can the AI Industry Regulate Itself? Stripe Wants PayPal, China Catches Up, NY Bans Datacenters
Saturday, 18 July 2026 · 4 min read · Listen to the episode ↗
This episode opens with a debate over whether the AI industry can police itself through a FINRA-style self-regulatory organization proposed by Demis Hassabis, backed by Elon Musk, Sam Altman, and others, under which frontier labs would submit models 30 days before release for assessment on catastrophic risks, with speakers warning the body must exclude content moderation and avoid regulatory capture by the three largest labs.
The central debate of the episode is whether the AI industry can regulate itself through a FINRA-style self-regulatory organization, as proposed by Demis Hassabis. Under the proposal, frontier labs would submit models 30 days before release for assessment on cybersecurity, national security, and biological risks, with benchmarks updated quarterly and the body empowered to coordinate a development slowdown if needed. Supporters include Elon Musk, Sam Altman, Sundar Pichai, Satya Nadella, Jack Dorsey, and Anthropic leadership, though one speaker characterized Anthropic's endorsement as an opening bid toward an FAA-style mandatory certification regime. Speakers outlined five conditions for viability: broad industry representation including startups and open source, review of frontier models only, focus on catastrophic risks only, voluntary before mandatory, and structured as a substitute for new regulatory agencies rather than an addition. Regulatory capture was flagged as a serious risk if only the three largest labs are represented, and speakers warned the body must not become a speech regulator covering disinformation or content moderation.
Anthropic was described as principally responsible for a state-by-state regulatory frenzy damaging the startup ecosystem, citing a Politico article detailing its strategy of encouraging states to impose increasingly strict AI guardrails using California's SB 53 as a model, with each subsequent state receiving more expansive rules. Anthropic has already secured regulatory victories in California, Illinois, and New York. Separately, Anthropic and Dario Amodei made a first seven-figure contribution to Public First, a group working to slow AI development and data center construction, which speakers called contradictory given that compute availability, not demand, is Anthropic's primary revenue constraint. Gavin Baker predicted Anthropic will reach a three trillion dollar valuation after its IPO, up from approximately one trillion dollars today.
Stripe and private equity firm Advent are jointly offering to acquire PayPal at approximately 60 dollars per share, with speakers expecting the final clearing price near 70 dollars per share. Block, led by Jack Dorsey, is contributing 17 billion dollars in equity to the combined offer. PayPal has 439 million consumer accounts and grows at roughly 7 percent per year but was described as a legacy business that peaked at 322 billion dollars in market cap and was trading at 30 to 40 billion before the offer. Stripe processes approximately 2 trillion dollars per year in transaction volume against PayPal's 1.7 trillion, but PayPal's consumer account base is the critical strategic asset. A combined entity of Stripe, Advent, and Block could reach 600 to 700 million accounts and, combined with Stripe's Bridge stablecoin infrastructure and PayPal's existing PYUSD stablecoin, could enable on-us transactions that bypass Visa and Mastercard entirely. Speakers predicted the deal could offer merchants a 3 to 5 percent discount to switch to proprietary payment rails. Antitrust scrutiny is possible if the market is defined narrowly as merchant payment APIs, though speakers argued the deal is pro-competitive if the relevant market is defined as Visa and Mastercard. Sacks argued PayPal stagnated after eBay acquired it in 2002 and removed all founding DNA, and speakers predicted this deal is the beginning of a wave in which AI-native companies acquire first-generation digital businesses that have become stale.
New York Governor Kathy Hochul signed the nation's first statewide moratorium on hyperscale data centers, citing fossil fuel use, land displacement, utility costs, water supply, and noise. Speakers disputed each justification, noting modern data centers use closed loop water recirculation, that New York is 70 to 80 percent undeveloped land, and that a typical data center uses the same water as two and a half In-N-Out Burger locations. One speaker predicted the moratorium will result in at least five years before a new data center can be switched on in New York, and noted that approximately 40 percent of all data center projects are currently being mothballed, creating extreme pricing pressure at the front end of the data center asset curve.
Apple filed a 41-page lawsuit against OpenAI on July 10 alleging stolen trade secrets after OpenAI poached over 400 Apple employees. Tang Tan, Apple's former VP of iPhone design and now OpenAI's chief hardware officer, allegedly directed job candidates to bring actual Apple parts to interviews. Chamath noted Apple is rarely litigious and its decision to sue signals something egregious occurred, though he does not believe OpenAI leadership deliberately induced the theft.
Token pricing varies from approximately 56 dollars per million input tokens on Claude to 50 cents on Chinese models, and token spend among Ramp customers grew 21 times over the past year. One speaker estimated 98 percent of Claude 5 prompts could be handled by cheaper lower-tier models. Apple's forthcoming M7 Ultra chip supporting up to 1.5 terabytes of RAM could allow employers to run 90 to 99 percent of AI workloads locally, putting significant downward pricing pressure on closed frontier models.
Researchers at Calico and Revel Pharma used AlphaFold and five cycles of directed evolution to produce an enzyme that eliminates 52 to 97 percent of advanced glycation end products, the predominant driver of extracellular aging, on proteins including collagen and hemoglobin. Applied topically to skin from elderly human donors, the enzyme eliminated 55 percent of the target compound, reverting biological skin age to approximately that of a 31-year-old. Delivery mechanisms remain an open question, and the first commercial application was predicted to be cosmetic skincare, with one speaker estimating a potential two trillion dollar market.
This summary was generated from the episode transcript and can contain mistakes.