20VC: Andrej Karpathy Joins Anthropic & Anthropic Raises $30BN at $900BN Price | SpaceX Files S1: How Does it Trade | Cerebras Smashes Day 1: What it Means for IPOs | Why Mass Layoffs Are More Worrying Than Anyone Sees

Summary of 20VC: Andrej Karpathy Joins Anthropic & Anthropic Raises $30BN at $900BN Price | SpaceX Files S1: How Does it Trade | Cerebras Smashes Day 1: What it Means for IPOs | Why Mass Layoffs Are More Worrying Than Anyone Sees

by Harry Stebbings

1h 21m•May 21, 2026
Listen to Episode

Overview of 20VC: Andrej Karpathy Joins Anthropic, Massive AI Fundraises, IPOs, and the Politics of AI

Harry Stebbings, Rory O’Driscoll, and Jason Lemkin spend this episode unpacking the latest wave of AI mega-financings and what they signal about the market. The conversation ranges from Anthropic’s rumored $30B raise at a $900B+ valuation and Andrej Karpathy joining the company, to the state of public software stocks, the success of Cerebras’ IPO, the likely reception to a future SpaceX listing, and the broader political backlash building around AI-led layoffs.

Anthropic: Karpathy Joins and the $900B Valuation Debate

Why the round makes sense

  • The hosts argue that Anthropic’s valuation is less about classic venture logic and more about capital intensity and compute warfare.
  • Their view: if Anthropic can raise enormous sums at scale, it’s because the company needs massive infrastructure spending and wants hyperscalers to fund that growth indirectly.
  • They repeatedly frame the deal as a balance-sheet war, where the key constraint is access to capital and compute, not near-term profitability.

The valuation argument

  • They debate whether ARR multiples still matter for a company like Anthropic.
  • Their conclusion: if the business is growing extremely fast and likely to IPO, then the current price may still be attractive relative to many private-market software deals.
  • They suggest that for investors with broad mandates, Anthropic may simply be the best risk-adjusted growth asset available.

Karpathy’s hire

  • Andrej Karpathy joining Anthropic is treated as a major signal of momentum and technical seriousness.
  • It reinforces the idea that the frontier-model race is still attracting elite talent and that Anthropic is building for the long term.

AI Token Economics: What Companies Are Really Spending

Salesforce’s Anthropic token bill

  • Benioff reportedly said Salesforce spent $300M on Anthropic tokens this year, mostly for coding.
  • The hosts break that down and conclude it is surprisingly normal given Salesforce’s engineering scale.
  • Their broader takeaway: token spend is becoming one of the largest new line items in software budgets.

The big market thesis

  • The episode’s core economic question is whether AI vendors can capture enough of:
    • engineering spend,
    • knowledge-worker wages,
    • and broader R&D budgets to justify the enormous capex being deployed.
  • Their rough estimate is that the winners may need to capture something like:
    • 5%–7% of knowledge-worker salary spend
    • ~20% of engineering spend to make the current AI economics work at trillion-dollar scale.

Bear case: tokens may be cheaper than expected

  • Jason argues that many companies may use far fewer tokens than people assume.
  • Real-world examples from Klaviyo and their own usage suggest token costs may be manageable if workflows are designed well.
  • If this holds, some of the biggest AI revenue projections could prove too optimistic.

Public Software Stocks: Re-rating Winners and Losers

The “AI is like being 21” framework

  • The hosts describe AI as the current category that gets valued on future potential rather than current fundamentals.
  • They contrast this with mature SaaS, which now tends to be valued on revenue growth, margins, and cash flow.
  • Once a company loses that “21-year-old” premium, it rarely gets it back.

Datadog, Figma, Atlassian, Twilio

  • They point to recent strength in names like Datadog, Figma, Atlassian, and Twilio as evidence that some public software stocks may have more time than many expected.
  • The key signal is re-acceleration in growth, which could support rerating even if the stocks never return to prior highs.
  • For these companies, the goal is not necessarily to regain peak multiples, but to become credible, durable, profitable growth businesses again.

Figma’s opportunity and threat

  • Figma’s AI story is two-sided:
    • It benefits from a broader software-building boom.
    • But it also risks being bypassed by tools like Lovable and Replit, which can take users from mockup to prototype faster.
  • The hosts think Figma needs to own the design-to-production workflow or risk leakage.

Wix, Squarespace, and the “terminal” software debate

  • They are much more bearish on Wix and Squarespace, especially if the core businesses are stagnant.
  • Their argument:
    • low-end website building is being eaten by vibe coding,
    • and ecommerce growth has been undermined by Shopify.
  • They discuss whether these businesses are effectively managing decline unless they can successfully migrate customers to new AI products.

Cerebras IPO: A Signal for the Market

Why it worked

  • Cerebras’ IPO is described as a huge success, with strong day-one performance and a major pricing expansion.
  • The hosts think this was possible because Cerebras is a very specific, high-quality bet:
    • a highly differentiated semiconductor business,
    • tied directly to inference demand,
    • with marquee AI customers.

What it means for IPOs

  • Their view is that the success helps the market’s appetite for elite AI infrastructure names.
  • It does not necessarily reopen the IPO window for lower-quality companies.
  • Their general rule: if you’re not “better than Figma” or clearly category-defining, the public markets may still be hard.

SpaceX, OpenAI, and the Future of Mega-IPOs

SpaceX IPO discussion

  • The hosts talk through a hypothetical massive SpaceX listing and what the S1 would actually reveal.
  • Their key point: a filing would likely show a historical snapshot, not the full AI-related transformation taking place now.
  • They joke that the real story would be underwritten by bankers and roadshow storytelling, not the S1 itself.

Retail mania and float dynamics

  • They debate whether retail investors could drive a huge post-IPO pop.
  • Jason thinks a meme-like, retail-driven rerating is possible.
  • Rory is more skeptical on base rates and warns that buying on the first-day pop is usually a poor long-term strategy.

OpenAI filing rumors

  • The episode ends with a rumor that OpenAI may file soon, interpreted as a sign that “money station” is open and the last trains are leaving.
  • The hosts treat it as a major signal that frontier AI companies may soon be forced into more public-market-like transparency.

OpenAI, Elon Musk, and the Political Backlash Against AI

The Musk lawsuit

  • They discuss the dismissal of Elon Musk’s case against OpenAI, framing it as a technical/legal loss rather than a broader vindication.
  • Jason argues Musk is likely pursuing a strategy of maximum nuisance and pressure, not just a legal win.

Sam Altman’s position

  • The hosts are sympathetic to the fact that Sam Altman has tried to position OpenAI as mission-driven rather than personally monetized.
  • At the same time, they note that complex side structures and affiliated funds create reputational and governance risk.
  • Their conclusion: the more complicated the structure, the easier it is for critics to attack.

The politics of layoffs

  • A major theme is that AI companies and public tech firms are underestimating political backlash.
  • They point to layoffs at Meta, Cisco, LinkedIn, Intuit, and others as evidence that AI is becoming a real labor issue.
  • Their warning:
    • if companies keep laying people off while touting AI gains,
    • the political response will eventually be harsh.
  • They predict a coming world where tech firms may need to reinflate headcount or otherwise address social unrest.

Compute, Infrastructure, and the Bubble Question

Nebius, CoreWeave, and compute scarcity

  • The hosts see companies like Nebius and CoreWeave as beneficiaries of a compute-scarce world.
  • If compute stays scarce, these businesses are valuable.
  • If supply catches up, they become much more commoditized.

The capex flywheel

  • They repeatedly emphasize that hyperscalers and model companies are spending at an enormous pace.
  • The spending supports:
    • NVIDIA,
    • networking vendors,
    • power infrastructure,
    • and data-center providers.
  • Their concern is whether the ecosystem is building too much capacity too quickly relative to real token demand.

The big uncertainty

  • Their bottom line: AI may be transformational, but the timing of demand vs. infrastructure buildout is the key risk.
  • If software adoption lags while capex keeps exploding, the market could face a severe correction.

Key Takeaways

  • Anthropic is now framed as a capital-scale company, not just a model lab.
  • Token spend is becoming a real software budget line item, but the exact magnitude is still unclear.
  • Public software is bifurcating: some names are re-accelerating, while others may be in structural decline.
  • Cerebras’ IPO is a positive signal for top-tier AI infrastructure names, but not necessarily for everyone.
  • SpaceX and OpenAI represent the next phase of AI-era public-market excitement, with huge implications for retail, institutions, and valuation discipline.
  • The political consequences of AI-driven layoffs may become one of the biggest risks in the cycle.