Anthropic Raises $45B but Falls Short on Compute & Thoma Bravo Hand Back Medallia

Key points

• $45B fresh into Anthropic from hyperscalers (cumulative compute-for-equity package). The pod's read: even with this they remain demand-bound on compute. Direct continuation of Patel's '72% margin floor' thesis from Issue 02 — Anthropic's bottleneck is supply, not customers.
• OpenAI missed user-growth AND revenue numbers. CoreWeave and Oracle dropped 5-7% in sympathy. Lemkin frames it as a *late-dropping shoe* on facts that were already visible: H2 2025 OpenAI didn't ship great models and lost market share to Anthropic.
• Lemkin's reversal — 'back to Team Sam.' Reasoning: 'I have to back my team of agents. They love OpenAI's API.' GPT-5.5 reviews-on-coding match Anthropic. The agent era reshuffles the deck.
• **Thoma Bravo hands Medallia to creditors.** $5.1B equity wiped out. Pre-AI low-growth SaaS that took on $2B+ debt couldn't service it. Direct evidence for our [Issue 02 prediction 2026-04-26-009](data/evals/predictions/2026-04-26.json) — though Medallia is a wind-down for capital-stack reasons more than an explicit LLM-displacement story. Partial validation, not clean.
• Three-bucket framework for SaaS valuations going forward: (1) **melting iceberg** — eroding terminal value, leveraged → dead. (2) **system of record** — sticky but no agent activity → bounded cash flow, deep-value play. (3) **agent-using** — increasing returns from AI traffic → growth re-acceleration possible.
• Agent-as-buyer thesis (Lemkin's strongest new framing): future demand routes through agents picking models and tools, not humans. Implication: Atlassian/Monday/Marketo are 'worthless to agents'; Stripe/Cloudflare/Twilio retain agent demand. Public markets seem to be pricing this distinction better than analyst narratives are.
• ServiceNow grew 20%+ but stock dropped because analyst call was *grilling* on agent-revenue authenticity — 'is it real, or just bundling?' The market wants to see agent acceleration as proof of crossing into bucket 3.
• China blocked Manus's $2B acquisition. Signal: cross-border AI M&A is now a regulated dimension. Adds friction to the 'every $2T+ company can buy something' thesis from Issue 02.
• Canva 2.0 will IPO well, but: 'agents won't use Canva.' SMB design seat is durable; agent-driven design will route around the canvas UI. Same logic Levie made on the migration from buttons to APIs.
• JIRA / Confluence call-out: 'an agent has no need for these products.' If you're a workflow vendor with thin business logic, the human ICP is shrinking faster than your contract length protects you.
• On private credit and PE wind-downs: 'You can service $2B of debt on a $1B low-growth pre-AI company that has to transform to AI' — meaning you can't. Expect more PE-fund-handovers in the next 12 months.

Notable quotes

Themes

• Anthropic still compute-bound after $45B raise
• OpenAI's late-dropping shoe and Lemkin's flip to Team Sam
• Agents pick the LLM — the new buyer
• Three-bucket SaaS framework for the agent era
• Medallia as the first PE wind-down of the cycle