The Twenty Minute VCThe New VC Game | CoreWeave Sell-Off Creates Panic | Revenue Multiples Out of Control
CHAPTERS
- 0:00 – 3:51
Venture feels “broken”: strong ARR businesses can’t raise while moonshots do
Harry opens with a founder-centric frustration: profitable, fast-growing companies outside the hottest AI narratives are getting “crickets” from VCs. Rory argues funding isn’t only concentrating into a handful of startups; capital is simultaneously flowing to ultra-high-upside bets, while many “normal” vertical SaaS-like plays are feared to be disrupted by AI.
- •High-quality, profitable ARR companies struggling to raise despite strong metrics
- •Capital remains risk-on for extreme-upside science/infrastructure bets (quantum, space, etc.)
- •Perceived AI disruption creates an “inarticulate fear” for category-specific software
- •Power-law expectations have intensified: investors want a clear path to massive outcomes
- 3:51 – 8:08
Why mega-funds demand billion-dollar positions (and the market narrows)
Jason frames the modern large-fund reality: to matter, each investment must plausibly become a billion-dollar+ position, pushing funds toward only the biggest potential outcomes. He argues founders now must meet far more investors to find a fit, because top-tier firms have narrowed their “box” for what they’ll back.
- •Mega-funds optimize for billion-dollar positions, not just big exits
- •Top firms’ “acceptable” categories have narrowed (e.g., best-in-class AI-native winners)
- •Founders often must meet 100+ investors—more like “the old days”
- •Concentration of capital makes many good-but-not-hyped companies effectively unfundable
- 8:08 – 10:46
Seed vs growth flips: entry prices, allocation cuts, and why early-stage got harder
The group debates whether seed should be easier to generate returns than late-stage. Jason argues it’s now harder because seed rounds are priced high, ownership is lower, and multi-stage funds squeeze allocations—making it difficult for seed checks to ‘move the needle.’ Rory agrees the last five years likely inverted the traditional risk/return relationship.
- •Seed entry prices are up; ownership and pro-rata access often down
- •Multi-stage funds lead and compress earlier investors’ allocation
- •Early-stage “should” outperform, but recent dynamics invert the expectation
- •Late-stage value creation (SpaceX/OpenAI/Anthropic) has dominated returns
- 10:46 – 17:03
CoreWeave sell-off: AI trade sensitivity, debt, CapEx, and “B-tier” risk
Rory explains CoreWeave’s volatility as inherent to neo-cloud economics: massive CapEx financed with debt creates a tightrope where even slight demand slowdowns spook markets. Jason argues these names are structurally more fragile and ‘B-tier’ versus platform owners like NVIDIA, and therefore should be more sensitive to shocks.
- •Neo-clouds underwrite growth with heavy CapEx and debt; there’s little room to ‘hunker down’
- •Market reaction hinges on demand growth vs. spending trajectory, not demand in absolute terms
- •CoreWeave/Oracle seen as more exposed than A-tier platform owners
- •Debate: B-tier can outperform tactically, but carries higher risk and volatility
- 17:03 – 18:48
Anthropic’s IPO math: slowing growth, profitability optics, and a $1.5T narrative
They parse Anthropic’s reported profitability and decelerating growth, emphasizing that slowing from extreme growth is expected. Rory sketches how modest quarterly revenue additions can still justify a very large valuation on forward revenue, while Jason argues the company should ‘raise what you need’ and expects post-IPO chop due to share shifts.
- •Growth deceleration is normal after multi-year 10x compounding
- •Forward-revenue math can still support very high valuation if ramp continues
- •IPO pricing/round sizing framed as pragmatic: take capital, avoid over-optimization
- •Expect choppy public trading due to fast-moving competitive dynamics
- 18:48 – 21:49
Why foundation-model revenue may be less predictable than SaaS
Jason and Rory discuss why model-provider revenue behaves differently from classic SaaS ARR: developers and customers can switch quickly, routing can change week-to-week, and price competition could emerge to defend share. The key stabilizer, Rory argues, is sustained end-user demand at the top of the funnel; if that slows, competition becomes harsher.
- •Developer loyalty is low; market share can shift rapidly
- •Routing and model choice drive volatility in usage and revenue
- •Price competition becomes likely if growth slows and share battles intensify
- •Sustained end-user demand is the ‘beating heart’ that allows multiple winners
- 21:49 – 25:30
Replit switching default models: routing as the real story (quality/price wins)
Jason calls the Replit ‘switch’ a misframed media story: Replit already routes across models and simply changed the default based on cost-performance. The broader takeaway is that platforms like Replit/Lovable/Base44/V0 act as real-world laboratories for model economics, because they’re forced to optimize quality at scale under tight cost constraints.
- •Replit uses a router; default selection changed due to better cost-performance
- •For end-user products, routing optimization is existential (can’t fully pass costs on)
- •These platforms provide a clearer signal than hype: what’s actually best in production
- •Implicit critique: if alternative models were truly superior, routing would reflect it
- 25:30 – 28:47
Claude Code Review lands: competitive shockwaves for code-review startups
The release of Claude Code Review triggers discussion on which startups get pressured and how much bundling risk matters. Jason argues code review (broadly defined) is central because LLMs can generate code fast but making it safe and correct is the bottleneck; Rory notes these ‘close-to-the-sun’ markets grow fastest but face recurring platform encroachment risk.
- •Platform bundling risk is real when foundation models ship adjacent products
- •Pure-play tools still matter (multi-model checks, workflow depth), but must stay ahead
- •Fastest-growing AI app categories sit closest to foundation models—high upside, high risk
- •Code review emerges as critical as generation becomes cheap and abundant
- 28:47 – 31:35
“Build to prod” is the bottleneck: testing, deployment, memory, and the new SDLC
They broaden ‘code review’ into an end-to-end ‘build to production’ category: validating, testing, sandboxing, and shipping agentically-generated code. Both argue software output is exploding, forcing a redesign of the software development lifecycle and creating room for new tooling layers beyond code generation.
- •LLMs often produce 95% solutions; the remaining 5% costs days of verification
- •‘Build to prod’ includes testing, security, sandboxes, deployment and operational readiness
- •Software volume is increasing orders of magnitude; SDLC must evolve accordingly
- •Large opportunity for ancillary tooling even if outcomes aren’t $25B+ winners
- 31:35 – 35:42
Modal at $15.75B: inference is exploding and open-weights economics shift margins
Modal’s valuation and run-rate prompt a discussion about the inference layer’s rapid expansion. Rory contrasts inference providers’ current revenue with the much larger closed-model revenue base, arguing there’s massive headroom—especially as buyers prefer lower-margin token supply via open-weights when quality is ‘good enough.’
- •Modal’s acceleration signals explosive inference demand
- •Inference revenue today is still small relative to OpenAI+Anthropic—suggesting runway
- •Economic incentive: buy tokens at lower margin if quality parity is sufficient
- •Debate: hyperscaler-like winner-take-most vs. a broader field with many players
- 35:42 – 38:45
Investment Committee: Town at $1B—‘classic bet’ vs. capital intensity and uncertainty
In the show’s ‘IC’ segment, Jason is lukewarm on Town at a $1B valuation, framing it as a traditional Series A/B bet dressed up at today’s pricing. He focuses on limited data, unclear upside, and the looming need to fund inference—while joking that needing ‘only’ $500M is a relative positive versus more capital-hungry peers.
- •$1B is treated as the new ‘classic’ Series B entry point in today’s market
- •Town viewed as an early, uncertain bet with insufficient evidence to underwrite conviction
- •Capital intensity (inference spend) is now a central diligence dimension
- •Tension between optionality bets and the need for outsized outcomes
- 38:45 – 42:41
Private vs public valuation gap: revenue multiples, Anduril, and ‘priced wrong’ risk
Jason questions whether private AI valuations are drifting too far from public comps, using Anduril and other ‘great but expensive’ rounds as examples. Rory agrees the disconnect is likely real, noting that only extreme growth can justify such pricing—and multiple expansion in public markets may not be available as a bailout.
- •Concern: private revenue multiples far exceed public-market comparables
- •Anduril example: valuation depends on aggressive future growth assumptions
- •Public markets credit fee-like stability; multiple expansion may be limited from here
- •Even great companies can be ‘priced wrong,’ creating return compression risk
- 42:41 – 46:06
Bending Spoons as ‘Berkshire of software’: buying unloved assets and compounding
They analyze why markets reward Bending Spoons’ roll-up/holding-company strategy, especially after a high-profile buyer increases its stake. Jason argues the firm is improving asset quality over time and aims to hold and optimize durable software cashflows; Rory counters that software may not behave like timeless consumer brands, making the analogy imperfect.
- •Bending Spoons’ model: acquire, improve, and hold software assets long-term
- •Market premium reflects confidence in process discipline and deal selection
- •Debate over whether ‘Berkshire’ logic translates to fast-changing software products
- •Trendline: moving from challenged assets toward higher-quality platforms
- 46:06 – 50:24
37signals and the case for never raising VC: profits, buyouts, and ‘none and done’
The conversation turns to bootstrapping, with 37signals cited as an example of substantial profits without venture dependency. Jason argues the AI era may make ‘one and done’ venture rounds less attractive because dividends and buyouts get complicated; Rory emphasizes many businesses can be extraordinarily valuable to founders without fitting the venture power-law mold.
- •Bootstrapped outcomes can generate enormous founder value without IPO pressure
- •Buying out investors with cashflow can be hard and risk-laden at scale
- •Jason’s shift: ‘none and done’ may beat ‘one and done’ in the AI era
- •Not every great company needs to be framed as a $25B-or-bust venture outcome
- 50:24 – 55:59
VC firms going public + LP FOMO: fee scale, culture tradeoffs, and the next boom cycle
Rory explains what it takes for investment firms to IPO: substantial, predictable fee income and scale, while public investors discount carry. They close by discussing Harvard’s strong returns, LP FOMO around SpaceX/Anthropic/OpenAI exposure, and the likelihood that distributions will recycle capital back into venture—potentially reigniting fundraising momentum.
- •IPO feasibility hinges on large, stable fee income; public markets value fees over carry
- •Going public can pressure firms to grow AUM and change culture toward risk-aversion
- •Harvard’s venture-heavy returns highlight concentration in a few massive winners
- •LP FOMO and impending liquidity could heat venture fundraising again
- 55:59 – 59:40
Tesla Roadster, the ‘Elon premium,’ and marketing via shared upside
They end on Tesla’s Roadster spectacle and Jason’s thesis of consolidation/merger dynamics around Elon’s empire. The group frames Elon’s promotional stunts as a valuation-supporting ‘Elon premium,’ and discusses the idea of broadly distributing ownership (e.g., a share for kids) as both marketing and political/PR insulation.
- •Roadster as a ‘swan song’ narrative and part of a larger strategic arc
- •Elon’s presence sustains premium valuations via attention and belief
- •Ownership distribution (giving shares) reduces hostility and aligns stakeholders
- •Marketing and financial engineering blend to extend narrative runway
