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

Summary

The episode opens with Apple’s lawsuit against OpenAI, which the speakers frame as potentially evidence-based and strategically aimed at OpenAI’s hardware ambitions rather than just employment mobility. They walk through California’s weak non-compete regime, the difference between general expertise and actionable trade-secret transfer, and why discovery could reveal how far any alleged misconduct went. The discussion then broadens into AI model economics: Meta’s Spark 1.1, cheaper tokens, and the argument that cost per completed task matters more than cost per token because reasoning, harness design, and workflow complexity drive real spend. From there the show moves into the memory supercycle and SK Hynix IPO, with AI capex benefiting a three-company oligopoly and crowding out other infrastructure budgets. The latter half focuses on venture structure: Jason Calacanis’s shift from seed to growth, the rise of late-stage private markets and liquid secondaries, gray-area startup tactics, terminal SaaS decay under AI, and Greylock’s decision to raise a comparatively disciplined $1.5B fund.

Notes

• Apple’s lawsuit against OpenAI is framed as a trade-secret case, not just a hiring dispute.
• The alleged conduct centers on a six-year Apple employee, not only the named long-tenured veteran.
• The alleged transfer included physical parts and show-and-tell material.
• Rory argues the person who physically took materials is the one most exposed to litigation.
• Jason says the employee likely already lost his job and is not helping his company.
• Apple may be leveraging the lawsuit to pressure OpenAI and map the chain of responsibility upward.
• Jason draws a practical line between trade-secret theft and ordinary employment portability.
• California makes non-competes hard to enforce and uses inevitable-disclosure logic only within limits.
• OpenAI’s hardware effort is described as a likely distraction that could be paused under pressure.
• Rory links the lawsuit to OpenAI hardware strategy and competitive positioning.
• Meta’s Spark 1.1 is described as moving from open weight/free to API monetization.
• The competitive response is cheaper models everywhere, especially for token-heavy workflows.
• Rory expects every company to maintain a low-cost internal model tier to control spend.
• Cost per completed task is the preferred unit of AI economics, not cost per token.
• Reasoning tokens and task complexity can dominate headline token prices.
• The harness and surrounding infrastructure materially affect efficiency.
• Agentic workflows can burn through budgets very quickly.
• Top-tier developers can multiply token usage by an order of magnitude.
• The user-facing problem is that AI can make workers look better while the company absorbs the cost.
• Claude Design could let teams ship further with fewer designers.
• Losing the low end can still damage the funnel for products like Figma.
• Anthropic’s revenue may have jumped from roughly $9B to $50B, with coding a major driver.
• Rory estimates 1.8 million U.S. developers and about $250B of annual wage spend.
• There are physical budget ceilings even if AI is compelling.
• Agentic AI can map to a roughly 10% software spend tax, implying large addressable spend.
• Rory’s framework splits the market into coding, an agentic software tax, and eventual knowledge-worker replacement.
• There is no clear evidence yet that model providers are moving into legal or biosciences at scale.
• SK Hynix, Samsung, and Micron are framed as an AI-capex oligopoly.
• The SK Hynix NASDAQ listing was described as the largest foreign listing ever.
• The recent memory-stock weakness may be trading noise rather than a true cycle break.
• IBM’s miss is used as evidence that CIO budgets are being diverted to memory.
• Finite capex budgets can crowd out other SaaS and infrastructure purchases.
• Jason Calacanis’s move from seed toward growth is treated as a major signal about venture-market structure.
• Jason says the shift to growth reflects how large winners now make early-stage ownership less compelling.
• Late-stage private investing is described as a new, durable product category.
• Secondaries are now liquid enough to make late-stage exposure more attractive.
• These companies are seen as going from zero to billions in about five years, not simply staying private longer.
• Paul Graham and YC are described as a business model, not just a picking exercise.
• Jason argues Paul Graham has earned the right to be followed because of his record.
• YC’s edge is described as owning a platform and network, not just stock-picking skill.
• Jason says affiliate-marketing behavior is common enough that it changes the ethical frame.
• Widespread gray-area GTM tactics make founder culpability harder to judge.
• Rory says some ethical gray areas are different from direct misconduct, but success can later reframe them.
• Top seed rounds are being priced around $200M pre in a narrow set of consensus bets.
• The late-stage private market has become a new asset class with big-check ownership math.
• Tranche structures and YC-style ownership math are now normalized by huge outcomes.
• Capital-intensive AI and neo-lab deals are being done on founder-friendlier terms than old biotech or semiconductor financings.
• Greylock’s new fund is $1.5B, which the hosts describe as a disciplined size.
• Jason says this fund size could be disciplined if it helps move capital through faster.
• Rory says the key question is whether the GP is optimizing for franchise health and returns.