Welcome to Nofinity

This transcript of a conversation between Ilia (Bankless) and an unnamed speaker (presumably a representative of Near Protocol) analyzes current developments in AI, particularly government regulation and the resulting opportunities for decentralized, user-owned AI solutions.

Key Points:

• Entropic's Fable-5 Export Ban: The US government imposed an export ban on Entropic's Fable 5 AI model, citing security concerns (potential jailbreaks). This sets a dangerous precedent for internet control, as it could potentially affect not just AI models but all digital services.
• Paradox of AI Regulation: The discussion highlights the paradox that while governments deem AI models dangerous (like Fable), they also seek to control and restrict their use. This is seen as a violation of fundamental freedoms (policing thoughts vs. actions).
• KYC and Surveillance: The demand for "Know Your Customer" (KYC) regulations for accessing powerful AI models is criticized. This would enable widespread surveillance of all users, resembling the surveillance mechanisms of 1984.
• The "Pandora's Box" is Open: The speaker argues that the development of powerful AI models is unstoppable. Open-source models (like JLM52) and decentralized computing infrastructure (e.g., Near) will soon offer an alternative to centralized, state-controlled models.
• The Role of Near/Web3: Near Protocol (with Ironclaw, Near AI, and Near Intents) is presented as a decentralized, user-owned alternative. It offers:
  • Confidential Inference: AI models running on user data without being sent to central providers.
  • User Ownership and Control: Data and AI model usage remain the property of the user.
  • Decentralization and Resilience: Networks are resistant to government censorship and outages.
  • Cost-Efficiency: Decentralized AI model usage is often cheaper than centralized options.
• The Marketplace for AI Agents: The speaker describes a marketplace for specialized AI agents (e.g., finance, law) on a decentralized, trustworthy infrastructure (Ironclaw).

Conclusion:

The discussion concludes with a call that the future of AI lies not in state regulation and surveillance, but in decentralized, user-owned technologies like Near. The speaker emphasizes that Near (and Web3) was built precisely for the challenges emerging today (state censorship, monopolization) and offers the solution for open, free, and private AI usage.

Key Takeaways:

• State control of AI models is a dangerous precedent.
• Decentralization and user control are the only way to preserve freedom and privacy.
• Near Protocol offers a practical alternative to centralized AI providers.
• KYC and surveillance measures are unacceptable and contradict the core principles of the internet.
• The future of AI lies in decentralized, open systems.

Summary follows...

In this episode of the Bankless Weekly Rollup, David and Ryan discuss several key topics:

1. Stock Market Recovery: The S&P 500 hit a new all-time high after a V-shaped recovery, driven by the liquidation of a major fund (Ashen Brener) which removed forced selling pressure.

2. MicroStrategy Selling Bitcoin: Michael Saylor sold 1,638 BTC ($105M) but Bitcoin barely reacted and even rose – a sign that Saylor is no longer the main marginal pricer of Bitcoin.

3. Ethereum Staking Yield Debate: A proposal from EF researchers (including Justin Drake) suggests burning staking rewards once 50% of ETH is staked, effectively driving yields to zero. Proponents argue Ethereum overpays for security and that unlimited staking harms ETH's monetary properties. Opponents (DeFi builders, Lido, Etherfi) warn it would crush solo stakers, scare institutions, and undermine credibility. The hosts believe the proposal is dead in the water ( <5% chance) given strong industry pushback.

4. Cold Card Exploit: A vulnerability in Coldcard's random number generator allowed hackers to brute-force seed phrases, draining over $130M in BTC from offline wallets. Victims did everything right but still lost funds – raising serious questions about self-custody in the age of AI-powered attacks.

5. Other News: Uniswap launches its own token launchpad (pools), the Clarity Act's chances drop to 15%, Cloudflare introduces wallets to protect content from AI bots, Near enables staking for AI inference, and Polymarket seeks a $20B+ valuation.

Overall Market Sentiment: The hosts feel crypto is in a waiting game – meme coins dominate, DeFi growth is slow, and fundamental progress (RWA tokenization, scaling) is incremental. Until the AI trade cools, crypto may remain in the background.

Summary follows...

This video discusses the massive opportunities and changes driven by AI, AGI, and chips. The speaker analyzes where the money is flowing in the age of AI and how to position oneself.

Computing Costs Skyrocket

• In 2025, a gigawatt of compute costs $50 billion, rising to $60 billion in 2026 and $73 billion in 2027.
• Growth rate of about 20–22% YoY.
• Bottlenecks include not just GPUs and servers, but also power supply, cooling, and maintenance.

GPU Rental Market Booms

• H100 GPUs are not obsolete; they are generating more revenue than ever.
• Prices for Blackwell clusters doubled from $2 to $4 per GPU hour in seven months.

SpaceX and Tesla's Terra Fab

• Elon Musk and Tesla plan the world's largest factory: 100 million sq ft (approx. 9.3 million m²) in Texas.
• Goal: Terawatt of compute, split into two product lines:
  • 25% for Tesla Optimus robots
  • 75% for Starship space missions
• Starting 2026, space-based data centers using Nvidia Vera Rubin chips are planned.

High Growth Rates at SpaceX

• SpaceX projects $100 billion in annual revenue by year-end.
• Three scenarios for monthly compute leasing revenue:
  • Conservative: $5.8–7.1 billion
  • Base: $7.5–9.2 billion
  • Optimistic: $10–11.7 billion
• CAGR of 170.4% to reach $1 trillion in revenue by 2029.

Starlink and Telecom Market

• Starlink has over 12 million active subscribers and is growing rapidly.
• CEO Gwynne Shotwell targets $600 billion through partnerships with Verizon, AT&T, and T-Mobile to eliminate "dead zones."

Company Valuations and Stock Trends

• Nvidia's P/E ratio is at its lowest (around 20), despite critics calling AI a "bubble."
• Tesla is seen as a "co-option on SpaceX" as the latter currently steals the spotlight.
• Google loses four top AI scientists but invests in companies like Anthropic and SpaceX.

AI Agents and Crypto

• AI agents increasingly use cryptocurrencies (e.g., Base from Coinbase) since banks are unsuitable.
• Base chain overtakes Solana and Polygon in cumulative transactions (over 150 million AI agent transactions).

Regulation and Risks

• The White House requires OpenAI, Anthropic, Google, and Meta to review AI models 30 days before release.
• 1,130 AI researchers call for slowing AI development due to loss of control.
• The speaker warns that slowing down could give China an advantage.

Outlook: End of Source Code?

• AI could soon convert legacy code (like COBOL or Pascal) directly into binaries, marking the end of traditional software development.

The chip and AI market offers enormous opportunities, but also risks from high costs, regulation, and geopolitical tensions. To position yourself, focus on Nvidia, SpaceX, and AI agent platforms.

• The ECM Aesthetika is a dual boiler with volumetric dosing, priced around €4,000.
• It sets benchmarks in many areas and can keep up with more expensive top machines (Sanremo, La Marzocco) – often even outperforming them.

• Only 6.5 to 7 minutes heat-up time thanks to a self-developed saturated brew group with its own heating element.
• Achieved by overheating and then letting the user cool the machine down.

• Very stable temperatures in KM and WBC protocols (5 and 14 consecutive shots).
• Excellent standard deviation – among the best machines tested in the last 10 years.

• Three programmable buttons for individual water amounts.
• Very good consistency even under load and after recalibration (7/10 points).
• Many other dual boilers (e.g., La Marzocco Linea Micra) lack volumetric dosing – only the more expensive GS3 offers it.

• High-quality, robust, solid build – typical ECM, including accessories.
• Espresso: solid 9-bar brew profile, no complex flow profiling, but can brew almost all coffees very well.

• Very powerful steam reaches milk temperature in just 24 seconds, with a nice rolling phase for creamy foam.
• 2-liter boiler maintains performance through many steaming cycles – ideal for small cafés or events.
• Easy connection to drainage and water supply (rotary pump) for continuous operation.

• Overall a top espresso machine with an excellent balance of temperature, volumetric dosing, and steam – a clear recommendation for serious espresso enthusiasts.
• Further details in the long video and test article.

• The S&P 500 hit a new all-time high (~5,700 points), driven by strong tech stocks (Nasdaq +4% for the week).
• Tech and Consumer Cyclicals dominate; Energy declines due to falling oil prices (hopes for an Iran deal).
• Earnings Season Highlights:
  • Visa, Mastercard – solid results
  • Microsoft (+10%), Amazon (+16%) – strong AI-driven earnings
  • Palantir (+29% in one day) – surprise rally
  • Eli Lilly – continues to be a pharma favorite
  • Apple, Tesla, Coinbase, Strategy – disappointed or were sold off
• Michael Burry shorted Palantir and Tesla – with mixed success.

• Leobold Aschenbrenner (former OpenAI employee) suffered massive losses in July due to 4-5x leverage. Ken Griffin (Citadel) bought the portfolio.
• Despite the setback, the fund is said to be still up 80% (thanks to an Anthropic bet).
• Lesson: Leverage is extremely risky. Successful investing requires unpopular decisions – the opposite of being an influencer.

• Bitcoin has underperformed the Nasdaq by 55% since January 2025, despite seemingly favorable conditions (pro-crypto administration, Fed, SEC).
• Cold Card Hack: Over $100M stolen due to a weak random number generator. Debate about Bitcoin's overall security.
• Outlook: Missing narrative and use case – dependency on Michael Saylor (MicroStrategy) and potential new catalysts.

• Critique of an article calling compound interest a fairy tale – from the author's perspective, untenable.
• Own strategy:
  • Fiat as a 'melting ice cube' – necessary safety buffer (1-3 months of expenses)
  • Base: broad ETFs (MSCI World) + sector weighting (e.g., cyclical vs. defensive)
  • Individual stocks only after pullbacks – current opportunities in:
    1. AI chips (after Q2 correction)
    2. Robotics (megatrend)
    3. Pharma (M&A, e.g., Eli Lilly)
  • Risk management: Don't jump between asset classes; reallocate within the portfolio.

• Upcoming events: Berkshire Hathaway, Nvidia earnings.
• Feedback on format welcome – weekly recap or prefer shorts?