All-In Podcast
1 hr 30 min video
3 min read
AI Regulation, PayPal Acquisition, Data Center Wars
You just saved 1 hr 27 min.
The big takeaway
DeepMind proposes self-regulatory AI standards body modeled on FINRA; Stripe, Block, and Advent bid for PayPal to create payments competitor; Apple sues OpenAI for alleged IP theft; Grok data leak exposes privacy risks; NY bans hyperscale data centers amid energy concerns; aging reversal breakthrough uses AI-designed enzymes.
AI Self-Regulation Proposal
DeepMind's SRO Framework
Demis Habis proposes a self-regulatory organization (SRO) for AI modeled after FINRA, with federal oversight but industry funding and independent expert management. Frontier labs submit models 30 days before release for assessment on cyber, national security, and biological risks, with quarterly benchmark updates and optional coordination of development slowdowns.
1
Frontier labs submit models 30 days pre-release
2
Independent experts assess cyber, national security, biological risks
3
Benchmarks updated quarterly
4
Body can coordinate development slowdown if needed
5
Voluntary initially, then mandatory
DeepMind SRO proposal workflow
Industry Support Coalition
Elon Musk, Sam Altman (OpenAI), Jack Clark (Anthropic), Sundar Pichai (Google), Jack Dorsey (Block), and others endorse the proposal as preferable to government-created regulatory agencies that would lack technical expertise and slow innovation.
1
Elon Musk
2
Sam Altman (OpenAI)
3
Jack Clark (Anthropic)
4
Sundar Pichai (Google)
5
Jack Dorsey (Block)
Key industry leaders supporting SRO proposal
Five Conditions for Success
Saxby Chambliss outlines critical conditions: broad industry representation including startups and open source to prevent regulatory capture; review only frontier models to avoid blocking incremental progress; focus exclusively on catastrophic risks (cyber, CBRN) not speech regulation; remain voluntary first to prove effectiveness; substitute for new government agencies rather than adding layers.
1
Broad representation (startups, open source)
2
Review frontier models only
3
Catastrophic risk focus only
4
Voluntary first, then mandatory
5
Substitute for new agencies, not addition
Chambliss's five conditions for SRO viability
FAA for AI as Cautionary Tale
FAA type certification for new aircraft takes 5-9 years; amendments take 3-5 years. An 'FAA for AI' would shift from monthly model releases to years-long approval timelines, potentially ceding AI leadership to China. This permission-based regulation prioritizes safety over innovation speed.
Current AI model release cycle
2 months
FAA aircraft type certification
7 years
FAA amendment timeline
4 years
Timeline comparison: current AI vs. FAA-style regulation
Anthropic's Regulatory Capture Strategy
Anthropic pursues state-by-state regulatory escalation rather than a single national framework, progressively tightening rules across California, Illinois, New York, and other blue states. This strategy creates regulatory ratcheting that favors incumbents with resources to comply while disadvantaging startups and open source.
Oct 2024
Saxby tweets Anthropic regulatory capture strategy
Recent
Politico: Anthropic state-by-state plan to ratchet up rules
Current
Regulations passed in CA, IL, NY; more states follow
Timeline of Anthropic's regulatory escalation strategy
Stripe-Block-Advent PayPal Bid
Acquisition Structure
Stripe and private equity firm Advent jointly bid ~$60/share for PayPal; Block (Jack Dorsey's company) contributes $17 billion in equity. The deal combines Stripe's merchant relationships, Block's point-of-sale infrastructure and Cash App, and PayPal's 439 million consumer accounts.
PayPal consumer accounts
439 million
Stripe annual transaction volume
2 trillion
PayPal annual transaction volume
1.7 trillion
Scale of combined payment ecosystem
Visa-Mastercard Disruption Play
Combined entity creates 600-700 million accounts with massive stable coin infrastructure (Bridge, SIUSD) and risk management systems. Vertical integration enables direct payment rails bypassing traditional credit card networks, potentially undercutting Visa and Mastercard's duopoly.
PayPal's Stagnation
PayPal grows only 7% annually despite 439 million accounts. Product is 25 years old and legacy; eBay's 2002 acquisition removed founding DNA and replaced it with consulting-type management, creating the 'PayPal diaspora' that founded SpaceX, Tesla, LinkedIn, and other companies.
7%
PayPal annual growth rate
PayPal's sluggish growth despite massive user base
Broader M&A Trend
PayPal and eBay bids exemplify a wave of AI-native operators acquiring stale first-generation digital businesses. Ryan Cohen (Chewy, GameStop), Bending Spoons (Evernote, AOL, Vimeo), and others diagnose underutilization, overspending, and lack of AI integration in legacy platforms, then revitalize them.
1
Bending Spoons: Evernote, AOL, Vimeo, Eventbrite
2
Ryan Cohen: GameStop, eBay bid
3
Stripe/Block/Advent: PayPal bid
4
Josh Kushner: Accounting firm rollups
AI-era operators acquiring legacy digital businesses
Deal Clearing Price
Initial $60/share bid likely undervalues PayPal; market expects $70+ per share. Elon Musk, with $4-5 trillion market cap and recent $60B acquisition appetite, could make competitive bid. Deal signals M&A resurgence post-Trump election as corporate development teams see regulatory environment shift.
Initial bid
60 $/share
Market expectation
70 $/share
PayPal market cap (pre-bid)
35 billion
PayPal valuation trajectory
OpenAI Legal Troubles
Apple IP Theft Lawsuit
Apple files 41-page lawsuit alleging OpenAI stole trade secrets for consumer hardware. Tang Tan (Apple's former VP of iPhone design, now OpenAI chief hardware officer) allegedly directed Apple job candidates to bring actual parts to interviews. OpenAI poached 400+ Apple employees; Apple rarely litigates, signaling egregious conduct.
400+
Apple employees poached by OpenAI
Scale of OpenAI's Apple talent acquisition
iPhone AI Default Relationship Breakdown
Sam Altman secured ChatGPT as default AI on iPhone—the most important platform for AI adoption. Lawsuit suggests Altman leveraged this relationship to gain access to Apple IP and talent, then burned the relationship. Pattern of burning benefactors (Elon, now Apple) emerges.
Employment Law Principle
California allows at-will employment and no non-competes, but employees cannot take employer IP, documents, code, or physical assets. Only knowledge in one's head is portable. OpenAI employees allegedly violated this by accessing Apple network storage and bringing parts to interviews.
Grok Data Leak and Privacy Fragility
Grok Build Privacy Breach
xAI's Grok Build (launched May 2024) promised no codebase transmission to servers, but actually uploaded entire codebases including passwords, API keys, and changelogs without user consent. Privacy setting failed to prevent upload. xAI disabled upload July 13 via server switch; Elon promised deletion and open-sourced Grok Build.
May 2024
Grok Build launched with privacy promise
July 13, 2024
Upload disabled via server switch
Post-incident
Grok Build open-sourced
Grok Build privacy incident timeline
Privacy Fragility in AI
Despite best efforts by trustworthy operators like Elon, non-obvious data leak vectors lurk in AI systems. Zero data retention (ZDR) policies cannot guarantee safety; trap doors exist that companies discover only after external exposure. Enterprises must assume information leakage despite vendor assurances.
Stratified Ecosystem Solution
Independent third-party layers (like 8090's software factory) interface with models to manage data exposure. Enterprises need explicit trust boundaries, private evaluation loops, decoupled orchestration, and rights to fine-tune outputs—operational control over compute, models, weights, data, and alpha.
AI Model Pricing Wars
Token Cost Disparity
Input token pricing varies wildly: Claude (Anthropic) $56/million, OpenAI $26-48, Grok $1, Chinese models $0.50. Enterprises paying 100x+ premium for closed models while open alternatives deliver comparable results at 1/100th cost. RAMP CEO reports token spend among customers grew 21x in one year.
Claude (Anthropic)
56 $/M tokens
OpenAI
37 $/M tokens
Grok
1 $/M tokens
Chinese models
0.5 $/M tokens
Input token pricing across AI providers
CFO Spend Control Crisis
Engineers use frontier models for tasks requiring only mid-tier models, unaware of cost. CFOs lack visibility into token spend until bills arrive. RAMP's token spend management feature addresses misaligned incentives: engineers optimize for capability, not cost; CFOs optimize for budget.
21x
Token spend growth among RAMP customers (1 year)
Explosive AI token consumption growth
Open Model Acceleration
Grok 4.5 (open-sourced), GLM-52, Llama variants, and Thinking Machines offer frontier-adjacent performance at 1/100th cost. Enterprises can fine-tune open models for specific tasks. Anthropic's regulatory push targets open source to protect pricing power and market share.
Apple Silicon Opportunity
M7 Ultra chip supports 1.5TB RAM (vs. M5's current capacity), enabling frontier-level models (Opus-class) to run locally on Mac Studio. Employers can run 90-99% of workloads on-device, eliminating token costs. Apple positioned to undercut Claude and OpenAI pricing by orders of magnitude.
M5 chip RAM
512GB
M7 Ultra RAM
1.5TB
Apple Silicon memory expansion enabling local AI
Data Center Energy Crisis and NY Ban
NY Hyperscale Data Center Moratorium
Governor Kathy Hochul signs nation's first statewide moratorium on hyperscale data centers, citing power consumption, land use, water use, and noise. Claims are largely false: modern data centers recirculate water (closed-loop), produce own power (behind-the-meter), use minimal land relative to economic value, and use natural gas (clean-burning).
1
Hochul claim: Power consumption
False—behind-meter solves
2
Hochul claim: Land use
False—best ROI per acre
3
Hochul claim: Water use
False—closed-loop recycling
4
Hochul claim: Noise pollution
False—manageable with distance
Hochul's data center claims vs. reality
US Energy Deficit Crisis
By 2050, US will be 2.5x California's energy consumption in deficit. PGE auction needed 7-8 GW; only 156 MW bid. 40% of data center projects mothballed due to power scarcity. Compute will chase energy; GPUs relocate to Middle East, Australia, Asia where energy abundant and regulations permissive.
2.5x
US energy deficit vs. California by 2050
Projected US electricity shortage
Behind-the-Meter Innovation
Elon's Colossus data center uses behind-the-meter power: mobile natural gas engines, solar, batteries. Bloom Energy provides large installations under personal-use clean air permits. Starlink enterprise versions enable 10-20 Gbps connectivity anywhere. Distributed edge compute becomes viable alternative to grid-dependent data centers.
Regulatory Capture and Foreign Influence
OpenAI blog post (July 2024) documents PRC-linked influence operations targeting US AI debates. China benefits from US data center moratoriums and export controls on chips—slowing US AI progress while China builds unconstrained. Anti-data-center activists funded by same groups that opposed GMOs and fracking (traced to Russia Today influence campaigns).
2010-2022
Russia Today anti-GMO media campaign in US
Post-2022
Anti-GMO sentiment declines after RT exit
2024
Similar pattern: anti-data-center activism
Pattern of foreign-influenced activist campaigns
Data Center Tax Revenue Benefits
Data centers generate substantial tax revenue for states and counties. Teachers in North Dakota received $30-40K bonuses from data center tax revenue. Construction jobs created during build phase; ongoing operations jobs persist. NY's moratorium sacrifices these economic benefits.
Global Data Center Shift
Middle East, Australia, and Asia accelerating data center builds. UAE now imports leading-edge chips. Starlink enterprise versions enable remote deployment. If US restricts data centers via regulation and export controls, compute relocates to allies and adversaries, ceding AI leadership and economic value.
AI-Driven Aging Reversal Breakthrough
Glycation and Extracellular Matrix Aging
Aging occurs partly through glycation: sugars and fats bind to proteins in the extracellular matrix (space between cells), accumulating and blocking repair. Advanced glycation end products (AGEs), particularly CML, cause stickiness, reduced mobility, wrinkles, joint stiffness, and inflammation.
AlphaFold-Designed Enzyme Breakthrough
Calico (Google) and Revel Pharma use AlphaFold to design novel enzyme breaking down CML. Directed evolution cycles test thousands of protein variants. Enzyme degrades 52-97% of CML in vitro; on human skin from 70+ year-old donors, eliminates 55% of CML, reversing skin age to 31-year-old equivalent.
CML degradation in vitro (best sites)
90 %
CML elimination on elderly human skin
55 %
Skin age reversal (donors >70)
31 years old
Enzyme efficacy in aging reversal
AI-Protein Synthesis Pipeline
Modern protein engineering: AlphaFold predicts protein structure from amino acid sequence; DNA encodes amino acids (3 letters per amino acid); bacteria print proteins from DNA; high-throughput screening tests variants; directed evolution iteratively improves function. Creates novel proteins not found in nature.
1
AlphaFold predicts protein structure
2
DNA encodes amino acid sequence
3
Bacteria synthesize protein variants
4
High-throughput screening measures activity
5
Directed evolution refines function (5+ cycles)
6
Test on human tissue
AI-driven protein engineering workflow
Delivery and Market Implications
Next challenge: delivering enzyme into extracellular matrix. Candidates include topical cream, injection, or mRNA therapy enabling in-vivo protein production. Cosmetic skin market likely first application (potential $2 trillion market); joint pain and systemic aging follow.
Worth quoting
"The industry needs to regulate themselves. We need self-certify each model before asking government to certify them."
— David Freeberg, at [3:35]
"If my choices are between FAA for AI or DMV for AI, I'd much rather go for Demis' SRO for AI."
— Saxby Chambliss, at [13:14]
"We are so massively short electrons. By 2050, the US will be 2.5x California's energy in deficit."
— Chimath Palihapitiya, at [56:43]
Made with Glimpse by Wozart
glimpse.wozart.com/v/aixdfaqv
Share this infographic
Read this infographic as text

AI Regulation, PayPal Acquisition, Data Center Wars

Summary of the video “Can the AI Industry Regulate Itself? Stripe Wants PayPal, China Catches Up, NY Bans Datacenters by All-In Podcast.

DeepMind proposes self-regulatory AI standards body modeled on FINRA; Stripe, Block, and Advent bid for PayPal to create payments competitor; Apple sues OpenAI for alleged IP theft; Grok data leak exposes privacy risks; NY bans hyperscale data centers amid energy concerns; aging reversal breakthrough uses AI-designed enzymes.

AI Self-Regulation Proposal

DeepMind's SRO Framework

Demis Habis proposes a self-regulatory organization (SRO) for AI modeled after FINRA, with federal oversight but industry funding and independent expert management. Frontier labs submit models 30 days before release for assessment on cyber, national security, and biological risks, with quarterly benchmark updates and optional coordination of development slowdowns.

Industry Support Coalition

Elon Musk, Sam Altman (OpenAI), Jack Clark (Anthropic), Sundar Pichai (Google), Jack Dorsey (Block), and others endorse the proposal as preferable to government-created regulatory agencies that would lack technical expertise and slow innovation.

Five Conditions for Success

Saxby Chambliss outlines critical conditions: broad industry representation including startups and open source to prevent regulatory capture; review only frontier models to avoid blocking incremental progress; focus exclusively on catastrophic risks (cyber, CBRN) not speech regulation; remain voluntary first to prove effectiveness; substitute for new government agencies rather than adding layers.

FAA for AI as Cautionary Tale

FAA type certification for new aircraft takes 5-9 years; amendments take 3-5 years. An 'FAA for AI' would shift from monthly model releases to years-long approval timelines, potentially ceding AI leadership to China. This permission-based regulation prioritizes safety over innovation speed.

Anthropic's Regulatory Capture Strategy

Anthropic pursues state-by-state regulatory escalation rather than a single national framework, progressively tightening rules across California, Illinois, New York, and other blue states. This strategy creates regulatory ratcheting that favors incumbents with resources to comply while disadvantaging startups and open source.

Stripe-Block-Advent PayPal Bid

Acquisition Structure

Stripe and private equity firm Advent jointly bid ~$60/share for PayPal; Block (Jack Dorsey's company) contributes $17 billion in equity. The deal combines Stripe's merchant relationships, Block's point-of-sale infrastructure and Cash App, and PayPal's 439 million consumer accounts.

Visa-Mastercard Disruption Play

Combined entity creates 600-700 million accounts with massive stable coin infrastructure (Bridge, SIUSD) and risk management systems. Vertical integration enables direct payment rails bypassing traditional credit card networks, potentially undercutting Visa and Mastercard's duopoly.

PayPal's Stagnation

PayPal grows only 7% annually despite 439 million accounts. Product is 25 years old and legacy; eBay's 2002 acquisition removed founding DNA and replaced it with consulting-type management, creating the 'PayPal diaspora' that founded SpaceX, Tesla, LinkedIn, and other companies.

Broader M&A Trend

PayPal and eBay bids exemplify a wave of AI-native operators acquiring stale first-generation digital businesses. Ryan Cohen (Chewy, GameStop), Bending Spoons (Evernote, AOL, Vimeo), and others diagnose underutilization, overspending, and lack of AI integration in legacy platforms, then revitalize them.

Deal Clearing Price

Initial $60/share bid likely undervalues PayPal; market expects $70+ per share. Elon Musk, with $4-5 trillion market cap and recent $60B acquisition appetite, could make competitive bid. Deal signals M&A resurgence post-Trump election as corporate development teams see regulatory environment shift.

OpenAI Legal Troubles

Apple IP Theft Lawsuit

Apple files 41-page lawsuit alleging OpenAI stole trade secrets for consumer hardware. Tang Tan (Apple's former VP of iPhone design, now OpenAI chief hardware officer) allegedly directed Apple job candidates to bring actual parts to interviews. OpenAI poached 400+ Apple employees; Apple rarely litigates, signaling egregious conduct.

iPhone AI Default Relationship Breakdown

Sam Altman secured ChatGPT as default AI on iPhone—the most important platform for AI adoption. Lawsuit suggests Altman leveraged this relationship to gain access to Apple IP and talent, then burned the relationship. Pattern of burning benefactors (Elon, now Apple) emerges.

Employment Law Principle

California allows at-will employment and no non-competes, but employees cannot take employer IP, documents, code, or physical assets. Only knowledge in one's head is portable. OpenAI employees allegedly violated this by accessing Apple network storage and bringing parts to interviews.

Grok Data Leak and Privacy Fragility

Grok Build Privacy Breach

xAI's Grok Build (launched May 2024) promised no codebase transmission to servers, but actually uploaded entire codebases including passwords, API keys, and changelogs without user consent. Privacy setting failed to prevent upload. xAI disabled upload July 13 via server switch; Elon promised deletion and open-sourced Grok Build.

Privacy Fragility in AI

Despite best efforts by trustworthy operators like Elon, non-obvious data leak vectors lurk in AI systems. Zero data retention (ZDR) policies cannot guarantee safety; trap doors exist that companies discover only after external exposure. Enterprises must assume information leakage despite vendor assurances.

Stratified Ecosystem Solution

Independent third-party layers (like 8090's software factory) interface with models to manage data exposure. Enterprises need explicit trust boundaries, private evaluation loops, decoupled orchestration, and rights to fine-tune outputs—operational control over compute, models, weights, data, and alpha.

AI Model Pricing Wars

Token Cost Disparity

Input token pricing varies wildly: Claude (Anthropic) $56/million, OpenAI $26-48, Grok $1, Chinese models $0.50. Enterprises paying 100x+ premium for closed models while open alternatives deliver comparable results at 1/100th cost. RAMP CEO reports token spend among customers grew 21x in one year.

CFO Spend Control Crisis

Engineers use frontier models for tasks requiring only mid-tier models, unaware of cost. CFOs lack visibility into token spend until bills arrive. RAMP's token spend management feature addresses misaligned incentives: engineers optimize for capability, not cost; CFOs optimize for budget.

Open Model Acceleration

Grok 4.5 (open-sourced), GLM-52, Llama variants, and Thinking Machines offer frontier-adjacent performance at 1/100th cost. Enterprises can fine-tune open models for specific tasks. Anthropic's regulatory push targets open source to protect pricing power and market share.

Apple Silicon Opportunity

M7 Ultra chip supports 1.5TB RAM (vs. M5's current capacity), enabling frontier-level models (Opus-class) to run locally on Mac Studio. Employers can run 90-99% of workloads on-device, eliminating token costs. Apple positioned to undercut Claude and OpenAI pricing by orders of magnitude.

Data Center Energy Crisis and NY Ban

NY Hyperscale Data Center Moratorium

Governor Kathy Hochul signs nation's first statewide moratorium on hyperscale data centers, citing power consumption, land use, water use, and noise. Claims are largely false: modern data centers recirculate water (closed-loop), produce own power (behind-the-meter), use minimal land relative to economic value, and use natural gas (clean-burning).

US Energy Deficit Crisis

By 2050, US will be 2.5x California's energy consumption in deficit. PGE auction needed 7-8 GW; only 156 MW bid. 40% of data center projects mothballed due to power scarcity. Compute will chase energy; GPUs relocate to Middle East, Australia, Asia where energy abundant and regulations permissive.

Behind-the-Meter Innovation

Elon's Colossus data center uses behind-the-meter power: mobile natural gas engines, solar, batteries. Bloom Energy provides large installations under personal-use clean air permits. Starlink enterprise versions enable 10-20 Gbps connectivity anywhere. Distributed edge compute becomes viable alternative to grid-dependent data centers.

Regulatory Capture and Foreign Influence

OpenAI blog post (July 2024) documents PRC-linked influence operations targeting US AI debates. China benefits from US data center moratoriums and export controls on chips—slowing US AI progress while China builds unconstrained. Anti-data-center activists funded by same groups that opposed GMOs and fracking (traced to Russia Today influence campaigns).

Data Center Tax Revenue Benefits

Data centers generate substantial tax revenue for states and counties. Teachers in North Dakota received $30-40K bonuses from data center tax revenue. Construction jobs created during build phase; ongoing operations jobs persist. NY's moratorium sacrifices these economic benefits.

Global Data Center Shift

Middle East, Australia, and Asia accelerating data center builds. UAE now imports leading-edge chips. Starlink enterprise versions enable remote deployment. If US restricts data centers via regulation and export controls, compute relocates to allies and adversaries, ceding AI leadership and economic value.

AI-Driven Aging Reversal Breakthrough

Glycation and Extracellular Matrix Aging

Aging occurs partly through glycation: sugars and fats bind to proteins in the extracellular matrix (space between cells), accumulating and blocking repair. Advanced glycation end products (AGEs), particularly CML, cause stickiness, reduced mobility, wrinkles, joint stiffness, and inflammation.

AlphaFold-Designed Enzyme Breakthrough

Calico (Google) and Revel Pharma use AlphaFold to design novel enzyme breaking down CML. Directed evolution cycles test thousands of protein variants. Enzyme degrades 52-97% of CML in vitro; on human skin from 70+ year-old donors, eliminates 55% of CML, reversing skin age to 31-year-old equivalent.

AI-Protein Synthesis Pipeline

Modern protein engineering: AlphaFold predicts protein structure from amino acid sequence; DNA encodes amino acids (3 letters per amino acid); bacteria print proteins from DNA; high-throughput screening tests variants; directed evolution iteratively improves function. Creates novel proteins not found in nature.

Delivery and Market Implications

Next challenge: delivering enzyme into extracellular matrix. Candidates include topical cream, injection, or mRNA therapy enabling in-vivo protein production. Cosmetic skin market likely first application (potential $2 trillion market); joint pain and systemic aging follow.

Notable quotes

The industry needs to regulate themselves. We need self-certify each model before asking government to certify them. — David Freeberg
If my choices are between FAA for AI or DMV for AI, I'd much rather go for Demis' SRO for AI. — Saxby Chambliss
We are so massively short electrons. By 2050, the US will be 2.5x California's energy in deficit. — Chimath Palihapitiya

More like this