20VC: Anthropic Buys Compute From Elon & Commits $200BN to Google | Cerebras IPO: The Breakdown | Ramp's $40BN Latest Valuation | Hubspot Tanks, Monday Rockets: WTF is Happening in Public Markets

What It Covers

Harry Stebbings, Rory O'Driscoll, and Jason Lemkin analyze five major developments: Anthropic's secondary share restrictions, its compute deals with SpaceX and Google ($200B over five years), the Cerebras IPO priced at $48B fully diluted, Ramp's $40B valuation, and diverging public market reactions to HubSpot, Monday.com, AppLovin, and Cloudflare earnings.

Key Questions Answered

• •Anthropic Secondary Restrictions: Anthropic's board approval requirement for all secondary transfers isn't new policy — most venture-backed companies have had identical charter provisions for years. The real risk lies in synthetic economic transfer agreements between individuals, where one party contracts to pass along share proceeds without Anthropic's involvement. These bilateral contracts remain legally enforceable between parties but leave buyers with zero recourse against Anthropic at IPO, creating significant loss exposure for retail SPV participants.
• •AI Market Consolidation Signal: SpaceX's Colossus data center running at roughly 11% utilization signals that xAI has effectively exited the frontier model race. Anthropic paying an estimated $3–5B annually for that capacity converts a money-losing asset into roughly 15–20% of SpaceX's total revenue run rate. Competitors with excess compute capacity should proactively engage rivals — the Samsung-Apple component supply dynamic applies directly, and refusing to meet with competitors forfeits significant revenue opportunities.
• •Parallel Agents Underestimate Token Demand: Goldman Sachs projects 24x token consumption growth by 2030, but parallel agent workflows — where models simultaneously run 10+ versions of a task and select the best output — suggest that figure is conservative by an order of magnitude. The cost per token drops roughly 10x every two years through chip improvements and optimization, while usage per workflow increases 10x per capability tier, making net demand forecasting extremely difficult but directionally massive.
• •Public Market Valuation Framework: Stock price at earnings time determines reaction more than quarter quality. Monday.com traded near 1.5x cash and raised forward guidance, producing a 20% stock gain despite deceleration. HubSpot lowered guidance and fell 18%. The actionable rule: decelerating SaaS companies trading at depressed multiples need only demonstrate 30% growth, profitability, and a credible AI roadmap to recover to 5–6x revenue — not the 20x+ multiples of 2021.
• •ZoomInfo as AI Disruption Case Study: ZoomInfo's revenue growth collapsed to approximately 1% as Clay and similar tools commoditized B2B data by building waterfall systems that compare multiple data providers simultaneously. Clay's core innovation predates LLMs — it aggregated competing data sources, eliminating ZoomInfo's monopoly pricing power, then layered AI on top. At roughly 1x revenue with 35% adjusted operating income, the company fits a classic private equity take-private profile if it cannot reestablish growth.
• •Cerebras IPO Risk-Reward Structure: Cerebras priced at $48B fully diluted, 20x oversubscribed, with historical revenue concentrated in UAE customers and forward revenue dependent on OpenAI and Amazon commitments not yet fully contracted. The investment thesis rests on one data point: NVIDIA's market cap exceeds $5.5T, and Cerebras needs only 1% of that addressable market to justify current valuation. Investors should expect a strong first-day pop but recognize the two-year business trajectory carries substantial customer concentration risk.