The Twenty Minute VCThe New VC Game | CoreWeave Sell-Off Creates Panic | Revenue Multiples Out of Control
At a glance
WHAT IT’S REALLY ABOUT
Mega-funds, AI volatility, and why “build-to-prod” becomes the prize
- The episode argues that venture is increasingly optimized for mega-funds that require billion-dollar positions, concentrating capital into a small set of AI-adjacent ‘A-tier’ outcomes and leaving many strong profitable companies unfundable.
- The hosts debate whether early-stage investing has become harder than growth due to inflated seed entry prices and allocation squeeze from multi-stage funds, which limits ownership and reduces fund-level impact.
- They interpret CoreWeave’s sell-off as a reminder that leveraged AI infrastructure businesses are hypersensitive to any perceived slowdown in AI demand, and may behave like high-beta ‘B-tier’ assets versus NVIDIA.
- They assess Anthropic’s slowing growth and IPO chatter as still compatible with a massive outcome, but emphasize that LLM revenues may be less predictable than SaaS because developer loyalty is low and price competition could emerge.
- They highlight ‘build-to-prod’ (code review, testing, deployment, sandboxing) as a major new software bottleneck created by rapid LLM-driven code creation, and view inference vendors like Modal as evidence of exploding demand for cheaper compute pathways.
IDEAS WORTH REMEMBERING
5 ideasVenture capital is being structurally pulled toward only trillion‑dollar narratives.
They argue that mega-funds mathematically need “billion-dollar positions,” pushing them toward a narrow set of perceived category winners (often AI) and away from merely excellent $100M–$2B outcome businesses. The result is a barbell market: a few hyperscaled bets get abundant capital while many strong, profitable companies struggle to raise.
Early-stage returns are being squeezed by high entry prices and shrinking ownership.
They contend seed has become harder to produce top-tier returns because entry prices are high (e.g., $50M–$100M “seeds”) while allocations and pro‑ratas get compressed by multi-stage funds. This combination makes it difficult for smaller early checks to “move the needle” even when companies become very large.
The AI infrastructure trade is fragile where leverage and CapEx are high.
CoreWeave’s volatility is framed as inherent to “neo-clouds” underwriting massive CapEx with heavy debt, making them hypersensitive to any perceived slowdown in AI demand. Lemkin further argues these are “B-tier” assets compared to NVIDIA—tradable, but fragile versus macro and sentiment shocks.
Foundation-model revenues may behave less like stable SaaS ARR and more like a competitive commodity market.
Anthropic’s reported growth deceleration is positioned as a narrative shock more than a collapse, but they stress that LLM revenue may be less predictable than SaaS due to rapid share shifts and potential price competition. They suggest IPO outcomes could be choppy even if absolute revenue remains huge.
Model routing is becoming a cost/quality arms race—and power users reveal the truth.
Replit switching its default model to OpenAI is interpreted as an optimization decision: similar quality at materially lower cost, not a “breakup” with Anthropic. They propose companies like Replit/Lovable act as real-time laboratories for routing decisions that reveal where model quality/price is genuinely competitive.
WORDS WORTH SAVING
5 quotesYou gotta have billion-dollar positions in everything in big boy venture today, I think. And, and I don't mean boy with a gender attribution. Big, big, big folk. You gotta have billion... It's billion-dollar positions or nothing. There's no point in investing otherwise with large funds.
— Jason Lemkin
Unless you can paint me the trillion-dollar market cap, I'm not interested.
— Harry Stebbings
When you think of the whole AI trade, these kind of neo clouds, they're always gonna be more risky than anything else in, in the system because they have, you know, they're underwriting high growth with large CapEx and lots of debt, right? So you try to thread the needle. If you grow too slowly, you don't make your c- customers happy. If you grow too quickly and demand slows even slightly, then you're screwed.
— Rory O’Driscoll
Compared to NVIDIA, we're all fucking B-tier assets. How about that?
— Rory O’Driscoll
I actually think code review is the single most important category today in software.
— Jason Lemkin
High quality AI-generated summary created from speaker-labeled transcript.
