20VC: Anthropic Raises $45BN but Falls Short on Compute | OpenAI Crushes with GPT5.5 and Codex: Back in the Game? | China Blocks Manus $2BN Deal to Meta | Thoma Bravo Hand Back Medallia Keys to Creditors | Why Google is a Bigger Buy Than Ever Before

What It Covers

Harry Stebbings, Rory O'Driscoll, and Jason Lemkin analyze five major stories: Anthropic's $45B hyperscaler fundraise amid compute shortages, OpenAI's revenue miss versus GPT-4.5 and Codex comeback, China blocking Meta's $2B Manus acquisition, Thoma Bravo's $5.1B Medallia equity wipeout, and the structural collapse of PE as a venture exit route.

Key Questions Answered

• •Agent-driven model selection: AI agents, not humans, will increasingly choose which LLMs and SaaS vendors get used. Lemkin reports his marketing and customer success agents prefer OpenAI over Claude for most workflows. This shifts competitive advantage away from human UX toward API quality and agent compatibility — founders should test their products against agent use cases, not just human ones.
• •Compute forecasting risk: Foundation model companies must commit 4-5x their current run-rate revenue in CapEx two years before that revenue materializes. At a $10B run rate growing 10x, that requires roughly $300B in forward infrastructure bets split between the company and hyperscaler partners. Getting this wrong in either direction — over- or under-building — creates either stranded assets or catastrophic capacity shortfalls.
• •Three-bucket SaaS valuation framework: Enterprise software now falls into three categories: melting icebergs (agents bypass entirely, terminal value near zero), systems of record (retained but no growth, calculable but modest value), and agent-accelerated platforms (increasing returns as AI leverages the product). Investors and founders should explicitly identify which bucket their company occupies before making capital allocation or exit decisions.
• •PE buyout model structural breakdown: Thoma Bravo's Medallia loss — $5.1B equity wiped on a company with ~$200M EBITDA and only ~$2B debt — shows the failure mode is overpaying, not over-leveraging. A $1B low-growth pre-AI business cannot service $2B+ in debt while simultaneously funding an AI transformation. Other at-risk names include Coupa, New Relic, Anaplan, Zendesk, Avalara, and Smartsheet.
• •Venture exit funnel contraction: The three traditional exit routes — strategic acquisition, IPO, and PE buyout — have narrowed to effectively one viable path: large IPOs requiring $400M+ revenue growing 40%+. Companies at $100M ARR growing 10-20% now lack credible exit options. Portfolio construction should shift toward fewer but larger positions, accepting that most companies will not reach exit scale rather than planning for mid-market outcomes.
• •Google as multi-vector AI winner: Google benefits regardless of whether Gemini or Anthropic wins the foundation model race, since it holds equity in Anthropic, supplies compute via TPUs, and generates cash flow from search. Unlike Nvidia's single-threaded CapEx demand bet, Google has multiple winning scenarios — AI adoption fast or slow — with the sole existential risk being ChatGPT materially eroding Google Search revenue.