20VC: Deepseek Raises $50BN | Wall St's $725BN AI Question | The Rise of Open Source & How it Threatens OpenAI & Anthropic | OpenAI Builds it's Own Chip: Jalapeno | The Death of Moats & The New AI Software Winners

What It Covers

Harry Stebbings, Jason Lemkin, and Rory O'Driscoll analyze DeepSeek's $50B Series A with Chinese government retaining sole voting rights, Wall Street's $725B question about AI ROI, OpenAI's custom Jalapeno inference chip, open source's threat to closed-source models, and why the "flabby middle" of AI pricing creates existential risk for Anthropic and OpenAI.

Key Questions Answered

• •AI CapEx math: Hyperscalers spend roughly $700B annually on AI infrastructure, requiring approximately $1.5T in revenue to justify returns. That implies replacing around 8% of the entire US labor force with AI-generated tokens just to break even. By 2027, CIOs will shift from token-maxing experimentation to demanding hard ROI proof before approving further AI budget allocations.
• •Open source pricing threat: Chinese government subsidies effectively fund DeepSeek and five comparable open-source models, making "open source" a misnomer — it's state-sponsored competition. This creates a pricing ceiling on Anthropic and OpenAI's mid-tier offerings. Anthropic is already emailing customers about prompt caching discounts specifically to undercut open-source cost comparisons and retain enterprise workloads.
• •Closed-source market structure: The foundation model market is consolidating into a two-player oligopoly. A third closed-source competitor faces simultaneous pressure from above (OpenAI/Anthropic on quality) and below (Chinese open-source on price). Without a parent company's balance sheet — as Google Cloud has — a standalone number-three closed-source model cannot survive the margin compression.
• •LLM moat destruction: Any competitive advantage built on data lock-in, switching costs, or workflow integration is now vulnerable to LLM-powered migration. Databricks claims a full data lift to their platform in 30 days versus the traditional five-year Accenture-led migration. Founders pitching moats without acknowledging this risk signal a fundamental misunderstanding of the current competitive environment.
• •Agentic finance workflow: A fully functional AI VP of Finance can be built in hours using current models connected to Salesforce, Bill.com, Brex, and QuickBooks — handling quoting, contracting, invoicing, follow-up, payment reconciliation, and bookkeeping without human intervention. Variable contractor costs drop 50–60% incidentally, not through deliberate headcount reduction, as agents absorb tasks humans routinely skip or delay.
• •Startup team structure shift: Winning startups in 2025 run smaller teams at top-of-market compensation with higher equity per person, operating six-plus days per week in-office. The investment calculus has shifted: a company with 40 high-output employees outcompetes one with 100 average performers. Founders still building around remote, part-time work cultures face structural disadvantage against AI-native competitors running continuous sprint cycles.