Why You Need a $1B Fund To Do Series A | SpaceX at $2TRN & Data Centers in Space | Groq's $20BN Deal
Jason Lemkin and Rory O’Driscoll argue Claude is winning enterprise AI lock-in while VC math is broken for Series A at current round sizes.
- RAMP data shows Anthropic captured 73% of new AI spending in early 2026, up from 50/50 with OpenAI just 10 weeks prior.
- Lemkin argues enterprise lock-in is happening now: teams that dialed in Claude-based agents won’t switch even if token costs drop, because QA and retraining costs are too high.
- SpaceX fab announcement: plan to build ~70% of TSMC’s volume capacity near the Gigafactory for ~$25B capex, 80% of output going to SpaceX data centers; O’Driscoll is skeptical the $400B valuation bump is justified by an announcement alone.
- Jeff Bezos is raising a $100B fund to acquire and AI-transform manufacturing companies across semiconductors, space, and defense, touring Middle East and Singapore sovereign wealth funds.
- Groq’s $20B deal to Nvidia framed as acqui-hire; Figma flagged as a warning sign because it is not meaningfully charging for its AI products — Lemkin’s test: if you can’t charge for AI, you’re not an AI company.
- Lemkin’s rule for legacy software: ARPO must be 50% higher than pre-AI baseline to count as genuine AI monetization; Notion appears to pass, most public SaaS does not.
- Series A fund math is broken: rounds are now $30-40M, requiring $25-30M checks; to lead 20 deals with 50% reserves you need ~$1B minimum, making sub-$1B early funds structurally marginal.
- Unicorn-to-acquirer ratio is at a career low — hyperscalers won’t buy 100 app-layer companies, PE is largely absent, and late-stage AI apps are priced above what prior-generation strategics can afford, making IPO the only real exit path.
2026-03-26 · Watch on YouTube
