20VC: Meta's Muse Hits No. 1. ChatGPT Finally Has a Rival | Menlo Sounds the AI Bubble Alarm | Factory Triples Its Valuation to $5 Billion | Keith Rabois vs Airwallex: Who is Right? | Crusoe's $3.9 Billion Round. Is the Data Centre Trade Overheating?

Summary of 20VC: Meta's Muse Hits No. 1. ChatGPT Finally Has a Rival | Menlo Sounds the AI Bubble Alarm | Factory Triples Its Valuation to $5 Billion | Keith Rabois vs Airwallex: Who is Right? | Crusoe's $3.9 Billion Round. Is the Data Centre Trade Overheating?

by Harry Stebbings

1h 20m•September 24, 2026
Listen to Episode

Overview of 20VC: Meta's Muse Hits No. 1. ChatGPT Finally Has a Rival | Menlo Sounds the AI Bubble Alarm | Factory Triples Its Valuation to $5 Billion | Keith Rabois vs Airwallex: Who is Right? | Crusoe's $3.9 Billion Round. Is the Data Centre Trade Overheating?

This episode is a fast-moving AI and venture capital roundup focused on three big themes: the rapid productization of AI, the rising capital intensity of the AI stack, and the increasingly heated debate over valuation discipline in venture. The hosts argue that AI is creating both massive new opportunities and new risks, with clear winners emerging in consumer AI, developer tools, and infrastructure — while also warning that some parts of the market may be getting overheated.

Major AI and Market Headlines

Anthropic / IPO timing

  • The hosts discuss Anthropic reportedly pushing its IPO timeline from October to November.
  • They frame the delay as a timing and storytelling decision, not necessarily a sign of weakness.
  • The key logic: waiting allows a cleaner Q4 narrative and the inclusion of stronger numbers after a pivotal quarter.

OpenAI burn and AI capital intensity

  • The discussion highlights just how capital-intensive frontier AI has become.
  • One estimate cited: OpenAI could burn $278B by 2030, with total required CapEx far higher.
  • Core takeaway: AI is not “software” economics anymore; it looks more like a massive industrial infrastructure business.

Meta’s Muse: The First Real ChatGPT Competitor?

Why Muse matters

  • Muse is described as the first real ChatGPT competitor in consumer AI.
  • The hosts are highly positive on the product:
    • It works well
    • It has strong consumer utility
    • It is free
    • It combines agents + LLM capabilities

Why it could matter strategically

  • Muse is framed as a Trojan horse: not just an assistant, but a way for Meta to fight ChatGPT with distribution.
  • The key advantage is Meta’s reach across Instagram and Facebook.
  • The launch is seen as a major validation of Meta’s AI strategy and a meaningful catalyst for the stock.

Amazon vs Shopify response to agentic commerce

  • The episode also discusses how Muse-like agentic shopping could affect commerce platforms:
    • Amazon is seen as defensive because it loses ad revenue and upsell opportunities if agents bypass its marketplace.
    • Shopify is seen as more aligned, since agentic demand can still benefit its merchants and payment rails.
  • Broadly, the hosts believe AI agents will compress middlemen and search-based businesses.

JEV: A New AI Model for Simple, Fast Decisions

What JEV is

  • JEV is described as a classifier-backed model, not a full reasoning LLM.
  • It is optimized for:
    • fast decisions
    • low cost
    • simple classification
    • routing / ranking / yes-no decisions

Why it matters

  • The hosts argue that a large share of LLM spending is wasted on tasks that don’t need full-model reasoning.
  • JEV represents the unbundling of AI into:
    • expensive “thinking” models
    • cheap “system one” models for quick decisions

Developer implications

  • This is framed as a developer product, not a consumer product.
  • Important implication: AI stacks will become more complex, because teams must now choose between multiple models and routing layers.
  • The episode suggests this will create demand for:
    • model harnesses
    • eval pipelines
    • routing infrastructure
    • DevOps for AI

Venture, Valuation, and the “AI Bubble” Debate

Menlo’s warning

  • The hosts react to a Menlo piece warning about overheating and reminding investors that “the music can stop.”
  • They debate whether venture is entering a phase where:
    • FOMO is driving prices
    • late-stage rounds are too aggressive
    • LPs may be paying for future growth that may not materialize

Main tension in the discussion

  • One side argues:
    • you should stay disciplined
    • price matters, especially when rounds are 10x higher than before
    • not every company deserves a huge check
  • The other side argues:
    • the best deals are still the best deals
    • venture persistence is real
    • you should back the strongest teams in the strongest categories

LP advice

  • For seed and Series A, the advice is to back managers who consistently get into the best deals in the newest categories.
  • But the episode also stresses:
    • later-stage pricing matters much more
    • liquidity and return compression are real risks
    • investors need to avoid overexposure to highly levered AI infrastructure names

Factory: Coding Is the Mother Load

Why Factory is attractive

  • Factory is described as an enterprise coding-agent business with very strong traction.
  • The hosts are bullish because:
    • coding is the largest value-creation category in AI
    • enterprises care about data sovereignty
    • companies want coding tools that don’t train on or expose proprietary code

Core thesis

  • “Coding is the mother load.”
  • The belief is that coding, QA, testing, and code review are the biggest near-term AI markets.
  • The hosts think enterprise buyers increasingly want:
    • on-prem or private deployments
    • control over their data
    • model choice
    • separation from the frontier-model providers

Why the valuation works

  • Even at a higher valuation, they believe the category is strong enough to justify continued investment.
  • Factory is seen as a platform that combines:
    • coding intelligence
    • enterprise trust
    • deployment flexibility

Legora: Strong Category, But Valuation and Margins Matter

What came up

  • Legora is discussed as a strong legal AI company with meaningful revenue traction.
  • But there is concern around margins, with one comment noting negative gross margins.

Takeaway

  • The category is attractive, but the hosts are more cautious than with coding.
  • Their view:
    • legal AI is real
    • it will matter
    • but margin structure and valuation should determine how aggressive investors can be

Crusoe: Data Center Trade Is Hot, But Risky

Why Crusoe is interesting

  • Crusoe is presented as a full-stack data center and GPU infrastructure business.
  • It has:
    • a large order book
    • strong demand
    • exposure to the AI infrastructure buildout

Why there is caution

  • The hosts repeatedly emphasize that data center businesses are levered to AI demand.
  • That means:
    • they can work extremely well in a boom
    • they can also get hit hard if AI spending slows

Main investing point

  • They like the category, but warn against being overconcentrated in the trade.
  • They prefer deals with:
    • long-term customer commitments
    • strong financing visibility
    • enough runway to survive a slowdown

Airwallex vs Keith Rabois: Politics, China Sensitivity, and Social Media Warfare

What’s going on

  • Keith Rabois and Joe Lonsdale are criticized for going hard at Airwallex on social media.
  • The hosts debate whether the criticism is fair, overly aggressive, or politically motivated.

Main themes

  • Chinese ownership / China exposure is a real issue in enterprise and M&A, even if the exact allegations are disputed.
  • The hosts distinguish between:
    • what is true
    • what is socially amplified
    • what real-world transaction friction looks like

Founders need an advocacy army

  • One key lesson: if you are a founder under attack, you need:
    • a strong comms strategy
    • public backers
    • defenders who can push back quickly
  • They suggest Airwallex has not built enough of that social/media defense layer.

Big Takeaways

  • Muse is a major Meta win and may be the first serious consumer challenger to ChatGPT.
  • AI is becoming more capital intensive, especially in frontier models and data centers.
  • JEV-style cheap classification models could meaningfully reduce wasted LLM spend.
  • Coding remains the biggest AI market and enterprise trust/data sovereignty are becoming decisive.
  • Valuations are heating up, but the hosts are split between discipline and “back the winners.”
  • Infrastructure trades like Crusoe are powerful but levered, so concentration risk matters.
  • Public narrative matters more than ever: founders need a real comms and advocacy strategy.

Notable Insights

  • “Coding is the mother load.”
  • “This is the first real ChatGPT competitor.”
  • “AI will maim the middlemen.”
  • “The real issue isn’t just safety — it’s data and trust.”
  • “The music can stop, but you still have to play the game.”

Practical Watchlist

For investors

  • Watch:
    • AI model burn rates
    • data center financing structures
    • enterprise demand for private / sovereign AI
    • margin profiles in legal AI and coding AI
    • whether consumer AI distribution shifts toward Meta

For founders

  • Build:
    • a public defender network
    • a comms strategy
    • data trust and sovereignty into the product story
    • model routing and evaluation infrastructure early