20VC: NVIDIA Bonanza: Buys Poolside & Invests in Mercor and Perplexity | Anthropic's $30TRN Revenue Assumption & OpenAI Confirms IPO | Why Customer Service, Defence and Robotics are Overinflated

What It Covers

Harry Stebbings, Jason Lemkin, and Rory O'Driscoll analyze NVIDIA's $12B Poolside acquisition, $20B Mercor investment, and Perplexity stake, alongside OpenAI's confirmed 2027 IPO plans, Anthropic's $30 trillion TAM claim, and which AI categories — customer service, defense, robotics — are attracting overvalued venture capital in 2025.

Key Questions Answered

• •Frontier Model Capital Reality: Only four or five companies globally can finance frontier AI models — Microsoft, Google, Amazon, and now NVIDIA. Poolside's failure to raise $2B despite strong execution proves no traditional VC fund has sufficient capital for next-generation model training. Seed investors who entered Poolside at ~$600M effective pre-money received only 15x returns, well below the 50x minimum needed to return a seed fund.
• •NVIDIA's Ecosystem Financing Strategy: NVIDIA deploys its estimated $70-75B annual free cash flow by funding customers and ecosystem partners rather than accumulating cash. Open-source models are economic complements to NVIDIA chips — more open-weight adoption means more GPU revenue relative to closed frontier model builders. Investors should recognize this vendor-financing pattern as deliberate market expansion, not speculative dealmaking.
• •OpenAI vs. Anthropic Competitive Shift: Anthropic's revenue run rate reached $60B mid-year while OpenAI grew only 18% quarter-on-quarter in H1, implying under $30B annualized. This ranking reversal forces OpenAI into a weaker negotiating position for its 2027 IPO — it will price relative to Anthropic's public valuation rather than independently, with open-weight competitors further pressuring its number-two market position.
• •Token Addiction as Enterprise Budget Crisis: Enterprise CFOs systematically under-budgeted AI token spend in 2025. High-performing employees now require continuous multi-agent workflows and will leave organizations that restrict access — creating a retention-versus-cost tension. The Stripe model of treating intelligence allocation like capital budgeting — with spending controls, ROI tracking, and per-employee limits — represents the emerging framework for managing this in 2026-2027.
• •Coding as the Only Sentence That Matters: Anthropic's enterprise revenue dominance over OpenAI traces directly to prioritizing coding use cases over consumer applications. Coding delivers the highest ROI, fastest adoption, and strongest willingness to pay among all AI applications. Investors and founders should evaluate AI product positioning by asking whether the primary use case is coding-adjacent — that single variable predicts revenue trajectory more reliably than any other metric.
• •Overvalued AI Categories to Avoid: Three categories are absorbing disproportionate venture capital relative to likely returns: customer service software (commoditizing rapidly, sophisticated companies building proprietary systems), defense tech (requires portfolio-of-products account control that consolidates to two or three players like Anduril), and humanoid robotics (dexterity and real-world deployment timelines remain unsolved, with specific-purpose robots like Locus outperforming general humanoids commercially).