Overview of 20VC: Trillion or Bust: The New VC Game...
Harry Stebbings, Rory O’Driscoll, and Jason Lemkin debate how AI is reshaping venture capital, public market valuations, and startup fundraising. The episode centers on a core thesis: venture is becoming more concentrated, with only a small set of companies and categories capable of attracting capital at today’s scale. They also discuss the recent CoreWeave sell-off, Anthropic’s latest numbers and rumored IPO pricing, the growth of inference infrastructure, why code review/build-to-prod may be the most important AI software category, and why deals like Modal, Town, and Anduril illustrate how disconnected private-market pricing has become.
Big Themes From the Conversation
Venture is becoming a “billion-dollar position” game
- The hosts argue that large venture funds increasingly need to make billion-dollar positions to move the needle.
- For mega-funds, the math pushes them toward:
- fewer bets,
- larger ownership targets,
- and only the most massive outcomes.
- As a result, many strong companies with good growth are becoming harder to fund, even when their fundamentals look excellent.
Only a narrow set of categories is truly “venture fundable”
- The conversation suggests that venture capital has become highly selective:
- AI infrastructure
- foundation-model-adjacent products
- deep tech / frontier science
- Meanwhile, many software businesses serving specific verticals or markets are seen as less compelling because investors fear AI will commoditize them.
Seed and growth are both under pressure
- Jason and Rory note that even at the seed stage, valuations are high enough that it’s hard to generate strong returns unless you get exceptional ownership and outcome size.
- At the same time, later-stage rounds are absorbing enormous amounts of capital, squeezing early-stage investors on price and allocation.
CoreWeave, AI Trade, and “B-Tier” Assets
CoreWeave’s sell-off reflects AI-market sensitivity
- CoreWeave’s weak week is framed as a reminder that AI infrastructure companies are highly levered to demand expectations.
- These businesses carry heavy capex and debt burdens, so any sign of slower AI demand can hit valuations quickly.
“B-tier” versus “A-tier”
- The hosts repeatedly contrast companies like CoreWeave and Oracle with NVIDIA:
- NVIDIA is the core IP / dominant platform.
- CoreWeave and Oracle are seen as more exposed, less essential, and therefore more vulnerable to shocks.
- The point is not that B-tier assets are uninvestable, but that they are more sensitive and arguably less attractive than the category leader.
Anthropic: Growth Slowing, But Still Massive
Revenue and IPO pricing
- Anthropic’s latest reported metrics are interpreted as strong, but with growth naturally slowing from extremely high levels.
- The rumored IPO valuation being revised from $2T to $1.5T is framed as more of a normalization than a crisis.
The key question: how fast does growth slow?
- The hosts emphasize that the real issue is not whether Anthropic is good, but whether its growth can remain steep enough to justify the multiple.
- They note that if growth continues in large quarterly increments, the forward multiple can still make sense.
- But they also caution that post-IPO volatility could be high because:
- developers switch quickly,
- market share can move rapidly,
- and pricing pressure could emerge if growth slows.
Inference Is Becoming a Huge Market
Modal’s $15.75B valuation
- Modal’s round is viewed as a signal that inference infrastructure is exploding.
- The hosts see this as a major opportunity and note that the total inference market may still be only a small fraction of the combined spend flowing through OpenAI and Anthropic.
The takeaway
- A major insight from the discussion: much of the AI value chain is outside the frontier model companies themselves.
- There is enormous demand for:
- inference platforms,
- routing layers,
- compute optimization,
- and cost-efficient model serving.
Code Review / Build-to-Prod: The Most Important AI Software Layer
Why this matters
- Jason argues that code review is the single most important category in software today.
- His broader framing is that the biggest challenge is no longer writing code—it’s getting AI-generated code into production safely.
Why this is a big opportunity
- AI has made it possible to generate far more software than before.
- But that creates a bottleneck in:
- testing,
- validation,
- review,
- deployment,
- and production readiness.
- The hosts believe that “build-to-prod” tooling could be one of the most valuable categories in the AI stack.
Why incumbents are vulnerable
- Anthropic’s code tools are powerful enough to make competitors nervous.
- But the discussion also suggests the market is still wide open for:
- code review startups,
- testing platforms,
- memory management tools,
- sandboxes,
- and deployment automation.
Town, Anduril, and Revenue Multiples
Town at $1B
- Town is treated as a classic venture-style bet:
- promising team,
- interesting product,
- but not yet something the hosts view as obviously deserving a huge premium.
- The broader point: a $1B valuation is no longer shocking, even for earlier-stage companies.
Anduril at $90B
- Anduril is described as a generational company, but the hosts question whether its valuation is too stretched relative to public comps.
- This leads to a wider worry:
- private markets may be assigning valuations that public markets would not support,
- and those multiples may not be sustainable unless growth remains exceptional.
The broader concern: multiple compression
- They repeatedly ask whether AI and defense companies are being priced as if growth will stay hyper-accelerated forever.
- If not, some of these valuations could prove too rich even if the companies themselves are excellent.
Bending Spoons and the Berkshire Analogy
Why the hosts like the model
- Bending Spoons is compared to “the Berkshire version of software”:
- buy durable assets,
- improve them operationally,
- hold them for the long term.
- The analogy is that the company can turn underappreciated software products into cash-generating assets, much like Berkshire did with consumer businesses.
Why it works
- The hosts see value in:
- disciplined capital allocation,
- long holding periods,
- and operational improvement rather than pure growth hype.
- They view it as one of the few models that can create durable value in a market dominated by growth-at-all-costs narratives.
General Atlantic IPO and the Future of Public Growth Firms
Why an IPO is plausible
- Rory explains that only firms with substantial fee income can really support an IPO.
- The financial logic is straightforward:
- public markets reward stable management fees,
- carry is less valued,
- and scale is required to matter.
Cultural tradeoff
- The harder issue is not just economics—it’s culture.
- Going public may make a firm more:
- compliance-heavy,
- growth-obsessed,
- and risk-averse.
- That makes an IPO possible, but also potentially changes what the firm is.
Other Notable Points
37 Signals / bootstrapped success
- The hosts admire how 37 Signals generated huge profit without needing traditional venture funding.
- It reinforces a broader point: not every great software business needs to be a venture outcome.
Harvard endowment
- Harvard’s strong return is tied to venture exposure, especially the large winners.
- The discussion highlights how concentrated venture outcomes can dramatically affect LP performance.
Tesla and Elon Musk
- Elon’s latest Tesla stunts are interpreted as brilliant marketing and brand management.
- The idea of distributing value directly to broad audiences—like giving shares to kids—is framed as a way to reduce backlash around extreme wealth creation.
Key Takeaways
- Venture is narrowing: capital is concentrating into fewer companies and categories.
- AI infrastructure is still huge, but highly sensitive to growth expectations and capex.
- Anthropic remains exceptional, even if its growth rate is naturally slowing.
- Inference and build-to-prod tooling are major opportunities outside the frontier model layer.
- Private-market valuations may be outrunning public comps, especially in hot AI and defense categories.
- Large funds now need massive ownership and outcomes, which makes investing harder, not easier.
Notable Phrases and Ideas
- “Billion dollar positions or nothing”
- “Power law on steroids”
- “Code review is the single most important category today”
- “B-tier assets”
- “Bending Spoons is the Berkshire version of software”
- “A billion is the new hundred million”