MurmurCast
NewsDiscussion

20VC: Apple Sues OpenAI | Zuckerberg Back on X and Challenging Codex and Claude Code | SK Hynix's $26BN IPO | Is Seed Investing Dead: Jason Calacanis Departs Seed for Growth | Greylock Raises New $1.5BN Fund

The Twenty Minute VC (20VC): Venture Capital | Startup Funding | The Pitch1h 22m

A weekly tech news roundup covering Apple's lawsuit against OpenAI for trade secret theft, Meta's aggressive pricing for Llama 3.1 models, SK Hynix's $26.5B NASDAQ IPO, the shift toward growth-stage investing, and analysis of AI token consumption, pricing dynamics, and the sustainability of AI spending growth.

Summary

The episode opens with Apple suing OpenAI for trade secret theft, involving a 24-year Apple veteran (Tang Tang) and a 6-year employee (Cheng Liu) who allegedly brought hardware components and confidential information to OpenAI. The hosts discuss how individuals who engage in this behavior face severe legal consequences and will be abandoned by their employers, while higher-level executives face deposition risks. They conclude Apple likely uses this leverage to halt OpenAI's hardware initiatives, which may have been a distraction for a company focused on frontier LLMs and enterprise coding solutions.

Meta launches Llama 3.1 and Spark 1.1 with aggressive API pricing compared to OpenAI and Anthropic, marking Meta's shift to a paid API model. Mark Zuckerberg breaks a three-year silence on X to announce the product, highlighting Meta's struggle with Threads' lack of developer engagement. The hosts note Meta is competing in the 'cheap seats' of model pricing while operating at massive scale.

The Databricks paper on cost-per-task versus cost-per-token receives significant discussion. The hosts emphasize that different models excel at different tasks (Pareto curve), and true cost assessment requires measuring completed tasks rather than raw token pricing. This framework explains why companies will adopt tiered models internally—using expensive frontier models for complex work and cheaper models for routine tasks.

Token consumption is analyzed extensively. The hosts discuss how top-tier developers using AI tools like Claude Design can easily consume 100x more tokens than current usage, and that even CFOs are now budgeting AI spend as a percentage of revenue (approximately 10% of software spend is emerging as a ceiling). The total addressable market for AI tokens is calculated at roughly $250-350 billion when including software engineering salaries, agentic software taxes, and knowledge worker displacement scenarios.

SK Hynix's NASDAQ IPO is noted as successful but volatile, with discussion of memory companies' current profitability (70% net margins at Samsung) being cyclical. IBM's stock crashed 20% after missing guidance, partly attributable to customers redirecting budgets toward memory purchases ahead of price increases.

Jason Calacanis's shift from seed investing to growth-stage opportunities prompts discussion about whether seed investing is dead. The hosts distinguish between structural trends (late-stage companies now reaching scale faster) and cyclical dynamics. They note that late-stage venture is increasingly replacing what public equity markets once provided, and that YC's continued focus on early stage reflects institutional advantage in that market niche.

The conversation covers high-valuation seed rounds ($200M+), which concentrate in capital-intensive areas like chip design and AI infrastructure. The hosts argue this reflects capital needs and fund ownership targets rather than a fundamental change in how venture works. They also discuss how large late-stage funds must invest in competing companies (unlike early-stage boards), as they're effectively replacing public market diversity.

Ethical gray areas in startups are discussed using Phoebe Gates' Fear and affiliate marketing cookie-stuffing as examples. The hosts debate whether widespread industry practices should be adopted despite ethical concerns, noting that Uber and Airbnb succeeded despite regulatory violations. They conclude that while cutting corners happens, legal liability and long-term consequences typically emerge.

Touch Bistro's acquisition by Constellation at 1x revenue (following a misaligned cap table with senior debt) serves as a case study in how venture debt combined with slow growth creates catastrophic outcomes. The hosts predict most pre-AI SaaS businesses face terminal decay accelerating faster than anticipated, potentially undermining roll-up valuations trading at 3x revenues.

Greylock's $1.5B fund size is discussed as a disciplined choice reflecting the firm's strategic focus on early-to-mid stage ventures rather than mega-funds pursuing late-stage platforms. The hosts distinguish between optimization for carry timing and genuine strategic discipline.

About this episode

<p>AGENDA:</p> <ul> <li>00:00 – Apple SUES OpenAI: Did They Steal Apple's Biggest Secrets?</li> <li>05:10 – Is OpenAI's $6BN Hardware Bet Already Dead?</li> <li>12:50 – Zuckerberg Is Back: Meta Finally Takes On OpenAI</li> <li>18:05 – The AI Spending Bubble Nobody Is Talking About</li> <li>23:45 – Claude Is Coming for Designers, Product Managers & Figma</li> <li>27:15 – Anthropic's $50BN Explosion: Have We Already Hit AI's TAM?</li> <li>36:00 – The $26BN AI IPO Powering the Entire Industry</li> <li>40:00 – Seed Investing Is Dead? Jason Calacanis Changes Strategy</li> <li>57:00 – SaaS Is in Trouble: AI Is Accelerating Terminal Decay</li> <li> 01:15:00 – Why Greylock Said No to Billions of Extra Dollars</li> </ul> <p> </p>

Key Insights

  • Individuals who steal trade secrets for personal career advancement face severe legal consequences and are typically abandoned by both former and new employers once litigation begins.
  • OpenAI's hardware initiative may have been a strategic distraction from its core mission in frontier LLMs and enterprise coding, making Apple's lawsuit potentially beneficial to OpenAI's focus despite the legal threat.
  • Meta's shift to paid API pricing for Llama models represents an acknowledgment that competing on cost at scale is viable only for companies with massive infrastructure, marking a fundamental business model change from open-weight models.
  • Cost-per-token is an incomplete metric for evaluating AI model economics; the true measure is cost-per-completed-task, which varies significantly by use case and creates opportunities for multi-tier model strategies within enterprises.
  • The total AI token market opportunity is constrained by finite software engineering budgets (approximately $250B annually in the US) and an emerging 10% organizational spending ceiling on AI, suggesting the market may be approaching saturation faster than commonly assumed.
  • Top-tier developers using agentic AI tools can increase token consumption by 100x without reaching practical limits, indicating that latency and infrastructure constraints (not pricing) may eventually govern adoption curves.
  • Memory chip makers (Samsung, SK Hynix, Micron) are experiencing cyclical peak profitability with 70% net margins that historically precede margin compression, suggesting current valuations at 5-8x PE may underestimate correction risk.
  • Late-stage venture has structurally replaced the role of public equity markets in providing diversified exposure to high-growth companies, requiring late-stage funds to invest in competing companies unlike early-stage boards.
  • Venture debt layered on slow-growth businesses creates catastrophic cap table misalignment where senior debt holders prioritize capital recovery over growth, effectively trapping equity holders in terminal decline scenarios.
  • Pre-AI SaaS products with sticky revenue (like POS systems) face accelerated obsolescence in the AI era, with renewal-cycle decay happening faster than historical precedent suggests.
  • Widespread ethical corner-cutting by startups (affiliate marketing, data scraping, regulatory arbitrage) does eventually incur legal and reputational costs, but successful companies often get those practices legalized retroactively.
  • Companies raising at high seed valuations ($200M+) concentrate in capital-intensive domains (chip design, AI inference) where billion-dollar capital requirements justify high pre-money valuations rather than representing a fundamental shift in venture economics.

Topics

Apple vs OpenAI trade secret lawsuit and hardware division impactMeta's Llama 3.1 aggressive pricing and developer platform strategyCost-per-task versus cost-per-token economic modelsAI token consumption growth and corporate budgeting constraintsSK Hynix IPO and memory company profitability cyclesSeed investing decline and growth-stage venture expansionHigh-valuation seed rounds in capital-intensive AI infrastructureLate-stage venture replacing public equity marketsEthical gray areas and startup corner-cutting practicesPre-AI SaaS terminal decay and roll-up valuationsVenture debt misalignment and failed acquisitionsFund size strategy and carry distribution timing

Transcript

To some extent, it may have been a distraction for OpenAI that they were so successful in consumer. If this was a mercy killing, Apple may have done them a favor. If there was action on threads, Harry, you'd be there. Action on Mars, Harry, you'd be there. Every company with a CIO who's half awake is going to have a cheap token model to hand to stop this madness. Shit, I'm putting everyone out of a job. The least I can do is keep them alive. Treason does not succeed, but what's the reason? If it does succeed, no one calls it treason. In the early stage when you're on the board, you can't invest in two competitors. In…

Full transcript available for MurmurCast members

Sign Up to Access
View original source →

More from The Twenty Minute VC (20VC): Venture Capital | Startup Funding | The Pitch

Get AI summaries like this delivered to your inbox daily

Get AI summaries delivered to your inbox

MurmurCast summarizes your YouTube channels, podcasts, and newsletters into one daily email digest.