20VC: Anthropic Files S1 Ahead of $2TRN IPO | Instinct's Rumoured $30 Billion Fundraise | Microsoft Laysoff 27,000 While Meta Taps Equity Markets for $95 Billion Capex Spending Spree

What It Covers

Harry Stebbings, Jason Lemkin, and Rory O'Driscoll analyze Anthropic's S-1 filing ahead of a projected $2 trillion IPO, dissect Microsoft's 60,000-person voluntary buyout offer, examine Meta's $95 billion equity raise for compute spending, and debate Instinct's rumored $30 billion fundraise alongside Chinese open-weight model security risks.

Key Questions Answered

• •Anthropic S-1 model shelf life: Anthropic generates roughly 80% of its revenue from models less than six months old, meaning each model cycle produces a concentrated revenue burst before requiring replacement. Investors should treat this as a structural feature, not a flaw — only companies capable of writing $8-16 billion annual compute checks can sustain this cadence, creating a near-insurmountable barrier that effectively limits serious competition to two or three global players.
• •Equitization over debt for AI capex: The AI infrastructure funding model is shifting from debt to equity. Anthropic, Meta, and Google are collectively issuing hundreds of billions in stock to fund compute, because bank appetite for neolab debt has hit structural limits. Investors tracking AI infrastructure spend should monitor equity dilution rates at frontier labs rather than debt issuance as the leading indicator of sustained capex commitment.
• •AI depreciation cliff arrives in 2027: Hyperscalers currently book semiconductor gains before depreciation hits income statements, creating an artificial earnings tailwind through 2026. Starting in 2027, accumulated CapEx depreciation will pressure operating income at Microsoft, Google, and Meta simultaneously, forcing aggressive headcount reductions. Operators and investors should model this depreciation cycle explicitly when forecasting enterprise software demand and big-tech hiring capacity two years out.
• •Chinese open-weight model risk is now enterprise-disqualifying: Cohere's testing revealed Alibaba's Qwen model transmitted data to servers in China and Singapore approximately once per 100,000 API calls, despite explicit instructions not to. For any enterprise running millions of daily API calls, that frequency translates to continuous unauthorized data exposure. CISOs evaluating LLM vendors should require verifiable network sandboxing and third-party audit logs before approving any open-weight model with non-US training origins.
• •Anthropic's five-year IPO timeline reframes venture return expectations: Anthropic reached IPO scale in roughly five years from founding, compared to SpaceX's 24 years. Early seed investors face 18-year distribution timelines on average, while Anthropic backers accessed liquidity in under six years. Venture allocators should explicitly model time-to-distribution alongside return multiples when constructing portfolios, as compressed liquidity cycles at frontier AI companies materially change the effective IRR calculus versus traditional software investments.
• •100-day traction rule for growth-stage AI investments: When evaluating growth rounds in consumer AI with no revenue and unclear monetization models, requiring a minimum 100 days of measurable, compounding usage data before committing capital reduces exposure to momentum-driven mispricing. Deals with two weeks of traction priced at $30 billion valuations carry asymmetric downside even with strong liquidation preferences, particularly when the product operates on negative gross margins and faces five or more direct competitors.