20VC: Anthropic Raises $45B but Falls Short on Compute

Category: Expert Interviews · Duration: 88 min · ▶ Watch

Speakers: Harry Stebbings · Jason Lemkin, Rory O’Driscoll

Switch language → 中文

Segments (10)

  • 00:00 · Intro & Headlines
    • Harry introduces the topics: Meta’s blocked acquisition of Manus, Anthropic’s funding, and Medallia’s PE wipeout.
  • 01:09 · OpenAI’s Missed Numbers & Market Share
    • The panel discusses OpenAI missing revenue targets and how Anthropic captured market share by building better models late last year.
  • 03:49 · The Agent War & Who Chooses the LLM
    • Jason argues that AI agents, not humans, will soon dictate which LLMs and software tools are used for enterprise workflows.
  • 10:06 · The Value of Multi-Year SaaS Deals in the AI Era
    • The discussion shifts to whether long-term enterprise SaaS contracts hold value if the underlying software is rendered obsolete by AI.
  • 18:46 · Anthropic’s Mega-Funding & The Compute Capital Expenditure Trap
    • Rory breaks down the massive capital intensity required to scale foundation models, noting the extreme ratio of CapEx to revenue.
  • 29:00 · Google and Nvidia’s Strategic Positions
    • The panel evaluates Google’s advantage with infinite compute capacity and Nvidia’s dominance in the hardware layer.
  • 34:59 · China Blocks Meta’s Acquisition of Manus
    • A brief discussion on the geopolitical implications of China blocking Meta’s $2 billion acquisition of the AI startup Manus.
  • 42:28 · Thoma Bravo, Medallia, and the Breaking of PE Software Models
    • Rory explains why the traditional private equity playbook of highly leveraging low-growth SaaS companies is failing in the AI transition.
  • 56:29 · The IPO Market and Exit Environment for Venture
    • The panel laments the closed IPO window and the lack of viable exit paths for late-stage venture-backed companies.
  • 01:06:29 · Rapid Fire Topics & Selling Figma/Duolingo
    • Harry runs through rapid-fire questions on various tech personalities and Jason discusses selling his positions in Figma and Duolingo.

Specific Prices (6)

Timestamp Item Value Context
00:10 Meta’s acquisition of Manus $2 billion The proposed acquisition price for the AI startup Manus, which was blocked by Chinese regulators.
00:24 Medallia equity wipeout $5.1 billion The amount of equity wiped out when Thoma Bravo handed Medallia over to creditors.
14:41 Canva subscription $18/month Jason mentions the monthly cost of a Canva subscription while debating if AI agents will render the human UI obsolete.
22:33 Microsoft PC software license $20/PC Rory uses Microsoft’s historical low-friction revenue model as a contrast to the high capital intensity of modern AI.
25:56 Salesforce annual bill $12,000 to $22,000 Jason notes his company’s Salesforce bill increased despite reducing human seats, due to increased API/token usage.
01:19:48 AngelList fund minimum investment $500 Jason jokes about the low minimum investment threshold for an AngelList syndicate.

Bottleneck Claims (1)

  • [20:33] Capital intensity is the primary bottleneck for scaling foundation models.
    • Evidence: Rory explains that to support $1 of run-rate revenue, an AI company must spend $4 to $5 in CapEx (chips, data centers, power) years in advance, creating massive financial risk.

Predictions (4)

  • [05:18, Late 2026 to 2027] Most enterprise workflows will be managed autonomously by AI agents.
  • [12:40, Within the next 1 to 2 quarters] Wall Street analysts will begin demanding metrics on ‘headless API’ or agent-driven revenue from SaaS companies.
  • [26:50, Medium-term (next few years)] There will be ‘air pockets’ in the AI market where compute supply temporarily outstrips demand, causing panic.
  • [54:00, Ongoing over the next few years] Many highly leveraged private equity software deals from the 2021 era will fail or require massive restructuring.

Key Technologies (3)

  • Large Language Models (LLMs): Foundation models like GPT-4 and Claude that power generative AI applications.
  • AI Agents: Autonomous software systems that can execute multi-step workflows, interact with APIs, and make decisions without human intervention.
  • GPUs / TPUs: The specialized hardware compute required to train and run inference for massive AI models.

Companies Mentioned (7)

OpenAI · Anthropic · Atlassian · Canva · Google · Medallia · Thoma Bravo

Notable Quotes (3)

The dirty little secret of venture again is how much of your money you make in that one year in 10 when everybody buys the dream. — Rory O’Driscoll @ 00:33

It’s possible you look back and see that as the first disconnect from compute equals revenue. — Jason Lemkin @ 24:15

You can’t service two billion plus of debt on a one billion low growth company with a pre-AI story that has to transform to AI. — Rory O’Driscoll @ 00:49

Key Topics

AI Agents · Foundation Models · Compute Economics · SaaS Evolution · Private Equity · Venture Capital Exits

Takeaways

  • AI agents will fundamentally change how software is purchased and used, shifting power from human users to autonomous systems.
  • The capital requirements for scaling foundation models are staggering, creating a high-risk environment even for market leaders.
  • Traditional SaaS and PE models relying on high debt and low growth are highly vulnerable to AI disruption.