The Twenty Minute VCAnthropic Raises $45B but Falls Short on Compute & Thoma Bravo Hand Back Medallia Keys to Creditors
CHAPTERS
OpenAI miss: market reaction vs underlying fundamentals
The group kicks off with OpenAI reportedly missing user growth and revenue targets, and how that rippled into public-market proxies like CoreWeave and Oracle. They argue the reaction is partly a “late shoe drop,” reflecting model-quality slippage in late 2024 more than a new 2025 shock.
Agents change the battlefield: humans stop picking models
Jason lays out a thesis that autonomous agents—not humans—will increasingly choose tools, vendors, and even which foundation model to call. That shifts advantage away from human preference (“Claude feels better”) toward what agents optimize for, potentially reshuffling winners.
What agents will recommend: momentum, leaders, and API ‘winners’
They explore whether only frontier models matter in an agent-picked world, using Jason’s experiment where multiple LLMs graded top APIs and tools. The takeaway: models tend to recommend market leaders with innovation momentum while dismissing legacy tooling as obsolete for agents.
Owning the agent layer: the real lock-in war
The conversation shifts from ‘best model’ to ‘best agent harness.’ If agents become the interface, whoever controls the agentic layer can influence model choice and create lock-in, echoing platform dynamics in prior software eras.
Enterprise contracts and “deferred churn”: masking vs solving decay
They debate whether multi-year enterprise deals still matter if agents erode the need for incumbent systems. Jason argues long contracts can postpone churn but don’t change terminal value; Rory offers a moderated view where systems of record may persist but growth can stall without agent-driven activity.
Software futures: melting icebergs, systems of record, or agent-powered platforms
Rory proposes a three-bucket framework for software companies facing agents: (1) eroding terminal value (“melting iceberg”), (2) stable system of record but low growth, and (3) platforms that gain compounding value from agent activity. They apply the lens to prosumer tools like Canva and why consumer vs enterprise timelines may differ.
Anthropic’s $45B hyperscaler financing: the compute constraint story
The panel unpacks Google and Amazon’s massive commitments to Anthropic as fundamentally about compute scarcity and capacity planning. Rory emphasizes the second job of foundation-model companies—securing compute—is capital intensive and requires high-stakes forecasting years ahead.
‘Compute ≠ revenue’: first cracks in the AI business model
They challenge the idea that more compute automatically translates to revenue, noting you need both adequate compute and a competitive model to generate demand. A model slip can break the perceived compute-to-revenue linkage and increase business risk.
Can excess compute be resold? The emerging ‘sublet’ market
Harry asks whether it’s easier to deal with surplus compute or shortages. The group suggests a market-like dynamic where hyperscalers reallocate capacity between model providers, creating de facto short-term “sublets” of massive compute blocks.
Why Google may be the biggest winner in AI infrastructure
Jason argues Google benefits regardless of whether customers use Gemini or Anthropic if it supplies capital and infrastructure. Rory adds the chip angle: Google/Amazon can bundle TPU/Trainium to capture margin that would otherwise go to NVIDIA.
China blocks Meta’s $2B Manus deal: leverage, people risk, and precedent
They discuss China’s move to unwind/block the acquisition, noting investors who already received distributions are unlikely to return capital. The deeper issue is geopolitical leverage, tech/control concerns, and the human impact on team members who may be stuck in China.
Medallia handed to creditors: not over-levered—overpaid and outpaced
The conversation turns to Thoma Bravo’s Medallia loss, framing it as a rare large-scale equity wipeout in PE. Key insight: the deal wasn’t necessarily extreme leverage at entry, but price paid plus weakening growth/terminal value made the debt unserviceable.
Private equity exit routes collapsing—and what that means for venture
They broaden from Medallia to systemic implications: if PE becomes a weaker buyer of last resort and IPOs require larger scale, exits narrow dramatically. Rory outlines portfolio construction shifts (“fewer but bigger winners”) and Jason suggests more founder-to-founder/peer mergers as an alternative path.
Venture ‘plumbing’ topics: revenue integrity, evergreen products, and retail VC access
The episode closes with venture-market mechanics: YC’s crackdown on “bullshit ARR,” Thrive’s ‘Eternal’ product (compared/contrasted with Sequoia’s evergreen move), and retail-style access vehicles (AngelList/Robinhood) with debate over high management fees versus access value.
