Skip to content

Anthropic Opens the Kimono, Bubble Questions and Frontier Costs, Q&A on Meta, Starlink, Brewers

Sharp Tech with Ben Thompson

Hosted by Andrew Sharp and Ben Thompson

1h 38m
Transcript

Original podcast

Where to listen

Summary

Written by a model from the transcript; not yet reviewed.

This episode focuses on Anthropic's public S1 filing, analyzing their underlying business health, heavy R and D and training costs, and creative accounting related to hyperscaler cloud revenue recognition. The discussion also covers Meta's soaring capital expenditures for AI and advertising-driven growth, alongside the broader macroeconomic realities of venture debt, share dilution, and the power dynamics between private tech labs and nation states. Finally, the episode touches on personal reflections regarding sports fandom, movies, and modern tech infrastructure.

  • Anthropic's largest portion of costs is out of their control due to competitive training expenses. 9:15
  • A significant portion of Anthropic's revenue involves platform providers through cloud services. 12:20
  • Meta's capital expenditures on AI are massive, yet their ads business remains exceptionally strong. 56:25
  • The long-term success of AI labs depends on their ability to move up the software stack for lock-in. 28:00

Chapters

  1. 0:04
    Opening and Brewers Banter

    Andrew and Ben kick off the episode with lighthearted talk about the Brewers and a playful jersey moment.

  2. 1:35
    Anthropic S1 Overview

    Ben and Andrew dive into Anthropic’s newly filed S1, noting its 400-page legal document and the company’s favorable market window.

  3. 2:58
    Risk Factors and Satire

    The hosts discuss the 93 pages of risk factors, the Rocco Basilisk satire tweet, and the broader conversation around Anthropic’s risk narrative.

  4. 5:05
    IPO Bubble and Market Dynamics

    Ben weighs in on whether Anthropic’s IPO will be a bubble similar to Coinbase, and the hosts explore the inevitable market volatility and funding challenges.

  5. 6:56
    Business Model and Costs

    The conversation turns to Anthropic’s profitable subscription model, high R&D spend, and the cost of training versus inference, highlighting the need to pace the frontier for safety.

  6. 11:00
    Governance and Share Classes

    The hosts unpack Anthropic’s seven share classes, the benefit corporation structure, and the implications for investors and profit motives.

  7. 15:00
    Anthropic vs. OpenAI Debate

    The episode examines the rivalry between Anthropic and OpenAI, touching on alignment, governance, and the broader AI ecosystem dynamics.

  8. 15:43
    Amazon Deal and Capital Structure

    Ben explains Amazon’s loan and equity deal with Anthropic, the resulting dilution, and how hyperscalers influence the company’s financials and governance.

  9. 17:00
    Future Outlook and Funding

    Ben and Andrew look ahead to Anthropic’s November IPO, potential valuation, and how hyperscalers’ capital will shape the AI industry’s future.

  10. 33:51
    Stripe and Platform Abstraction

    Andrew and Ben discuss Stripe’s role as a financial OS, its importance for Substack and other platforms, and why companies outsource payments.

  11. 48:32
    Funding Challenges for AI Startups

    The hosts discuss the difficulty of securing capital for AI companies and the impact of Amazon’s investment on Anthropic’s competitiveness.

  12. 52:53
    Meta’s AI Spending and Share Performance

    The conversation turns to Meta’s recent AI investment, its impact on share price, and the broader implications for the tech industry.

Highlights

  • 5:17

    Invevitable AI Bubbleinsight

    The discussion explores whether Anthropic's IPO mirrors Coinbase as the top of the AI market cycle, acknowledging an inevitable bubble.

  • 12:20

    Anthropic Revenue Accountingdata

    The host and guest unpack how Anthropic records platform service fees as revenue and why those margins are compressed.

  • 16:09

    Conversion of Amazon Loansstrategy

    An analysis of how Amazon's initial financing and computing loans convert into equity shares, resulting in large accounting losses.

  • 18:49

    Share Dilution Realityinsight

    The hosts explain how moving from debt and cash flow to equity financing means existing shareholders bear the brunt of dilution.

  • 32:11

    Stripe Infrastructure Abstractionstrategy

    The discussion highlights how Stripe acts as an essential financial operating system, allowing internet businesses to scale easily.

  • 43:20

    Capex and Revenue Growthdebate

    The hosts debate Craig Moffet's core question on whether end-demand revenues will eventually exceed the massive capex value chain.

  • 53:10

    Advertising and Human Desireinsight

    The guest argues that anti-ad sentiment in Silicon Valley misses the fact that consumerism and advertising drive the modern economy.

Quotes

  • “The largest portion of their cost is this training cost.”
    Ben Thompson · 9:05
  • “In that half year, the cost of their R and D spend for training and all that sort of thing went up more than the cost to serve.”
    Ben Thompson · 8:35
  • “And this money was actually given in the form of a loan.”
    Ben Thompson · 15:05
  • “The subscription for sure is profitable. It's less profitable than the API.”
    Ben Thompson · 6:56
  • “And when you're multiplying humans, you're basically creating entirely new things to do, new workflows, new opportunities.”
    Ben Thompson · 50:06
  • “There's a lag between the new technology and understanding the core.”
    Ben Thompson · 52:24
  • “They have all the dials turned to 11 of making you watch more reels, making you never look away, see more ads.”
    Ben Thompson · 1:08:38

About this episode

Ben and Andrew begin with a salute to the world champion Milwaukee Brewers before turning to Anthropic’s S-1 as the company prepares for its public offering in November. Topics: Anthropic’s impressive margins, questions about the second half of 2026, how the company “lost” $80 billion in 2026, how compute will be paid for going forward, and an SEC disclosure that finally shows us the breakdown of training vs. inference costs. From there: Anthropic’s obsession with OpenAI, reactions to Craig Moffet’s questions about the AI bubble, understanding Meta’s AI spending as an echo of Anthropic’s S-1, a brief 1Password history, and Muse as the ultimate founder bet. At the end: An emailer wonders about Elon’s role in the Ukraine War, Ben sees You Can See Everything, and a few emotional thoughts on the Brewers World Series win.

Transcript

0:00
1:38:03

17,719 words · 2 speakers

Speaker names marked are inferred from speaking time and the show’s known hosts — they are not voice-verified.

Topics

  • Anthropic
  • Amazon
  • OpenAI
  • Google
  • Brewers
  • Stripe
  • World Series
  • Microsoft
  • Nvidia
  • Neo Clouds

Page updated · transcript and summary are machine-generated