The Twenty Minute VCSequoia Partner, David Cahn on Who Wins in AI, Defence & The New $0–$100M Playbook
CHAPTERS
- 0:00 – 1:09
AI bubble thesis in one minute: fragility, compute economics, and VC humility
A cold open that compresses Cahn’s core worldview: AI markets are bubbly, but the key question is who survives. He frames bubbles as potentially beneficial for compute buyers (lower COGS → higher margins) and reiterates a hard VC lesson: investors can’t force outcomes.
- •Cahn states plainly he believes we’re in an AI bubble
- •Overbuilt compute can advantage “consumers of compute” via falling prices
- •Bubble survivorship matters more than bubble-calling
- •Venture cannot “make” a company succeed—founder and PMF dominate
- 1:09 – 3:07
From “bits” to “atoms”: why physical data centers and power became the real moat
Cahn revisits last year’s thesis that AI progress is constrained by physical infrastructure—steel, servers, electricians, generators, and power. He argues 2025 validated the view: energy constraints moved to the center of the narrative and even into mainstream coverage and GDP statistics.
- •AI’s limiting factors shifted from models/data to power and construction capacity
- •Supply chain constraints: electricians, generators sold out, vendor bottlenecks
- •“AI power trade” as a defining market bet of 2025
- •Physical buildout increasingly shows up as measurable GDP contribution
- 3:07 – 9:19
The $600B → $840B question: can end-user revenue justify the buildout?
Cahn updates his famous payback math: massive chip and data center capex implies an enormous revenue requirement to earn back investment at reasonable margins. The unresolved issue isn’t whether hyperscalers can spend—it’s whether there’s real, durable end demand (“the customer’s customer”).
- •Original payback framing: chip spend implies far larger data center investment
- •Updated 2025 revenue requirement estimate rises to ~$840B
- •Core risk: unclear end-user monetization versus visible infrastructure spend
- •Construction delays and “shovel-ready” reality start to surface
- 9:19 – 10:36
Vertical integration pressure: model labs become power, servers, and chips companies
The discussion turns to why labs increasingly need tight coupling between model work and infrastructure. Cahn argues OpenAI/Anthropic are moving down the supply chain (power procurement, chip work), and competitive pressure will force more providers to do the same.
- •Vertical integration as an advantage when compute is the constraint
- •Big labs increasingly act like infrastructure companies
- •Power procurement and custom chips as strategic responses
- •Expectation that the integration trend is durable across model providers
- 10:36 – 14:42
Are we in an AI bubble—and why “timeline compression” is the real tension
Cahn says the bubble view went from contrarian to consensus, even among major AI bulls. He separates long-run inevitability (AI reshapes society) from short-run market cycle dynamics, arguing markets may be pricing too much transformation too quickly on very specific near-term assumptions.
- •Bubble conversation shifts from fringe to broad consensus
- •Dot-com analogy: bubbles can still produce enduring winners
- •Key mismatch: transformative 50-year arc vs near-term market expectations
- •Investing posture: seek companies resilient to volatility and funding cycles
- 14:42 – 16:17
Who wins and loses: consumers vs producers of compute
Cahn lays out a simple investment framework: compute buyers benefit when compute is overproduced, while compute producers face commodity economics and cyclicality. The enduring value is created by turning cheap power/compute into differentiated intelligence and products people love.
- •Consumers of compute gain from falling compute prices (COGS down, GM up)
- •Producers of compute resemble commodity businesses with less destiny control
- •Commodity dynamics drive cyclicality and lower multiples
- •Value accrues to differentiated applications built on top of compute
- 16:17 – 20:32
Big Tech, monopolies, and why AI may be more competitive than the last era
Pressed on AWS/Azure/Google, Cahn argues today’s market over-indexes on monopoly outcomes because we live in an anomalous “monopoly era.” He claims AI is different: everyone expects it to be massive, so competition is intense—making monopoly profits less likely (and socially better).
- •We’re in a historically unusual concentration of power (Mag 7 dominance)
- •Past monopolies formed when the opportunity wasn’t obvious at the start
- •AI is universally recognized as huge, inviting many entrants and competition
- •Fewer monopoly rents would benefit consumers and healthy AI diffusion
- 20:32 – 22:22
How the bubble ends: game theory, no coordination, and Taleb-style fragility
Cahn explains why bubble behavior isn’t centrally coordinated—spending continues until incentives change. Using Taleb’s “wobbly building” idea, he focuses less on timing the break and more on recognizing fragility in the system.
- •Bubble spending is incentive-driven, not coordinated
- •“Invisible hand” dynamics dominate capital allocation
- •Taleb framing: you can observe fragility without predicting the exact collapse
- •Key investor problem becomes positioning for survivability
- 22:22 – 24:48
Circular deals and the new risk absorbers: from hyperscalers to Oracle/CoreWeave to chip vendors
Cahn identifies circular deal structures as the most visible sign of fragility and the biggest shift in 12 months. He describes hyperscalers stepping back, smaller players stepping up with less capacity to absorb risk, and chip companies financing builds—sometimes with effectively “negative” cost of capital due to revenue recognition.
- •Circular deals as a primary driver of today’s bubble consensus
- •Microsoft/Amazon stepping back from absorbing ecosystem risk
- •Oracle/CoreWeave stepping in but with smaller balance sheets
- •Chip vendors financing buildout while also booking revenue (circularity)
- 24:48 – 34:27
Gigawatts not dollars: the capital intensity shock and what an unwind might look like
They unpack what it means to announce projects in gigawatts, translating into tens of billions per GW and trillions overall—often unfunded. Cahn argues the unwind is more likely to be an equity narrative reversal (given heavy equity/cash funding) rather than a classic 2008-style credit crisis, with broad impact via public portfolios and Mag 7 concentration.
- •Rough conversion: ~1 GW can imply ~$40–$60B of buildout
- •Talk of hundreds of GW implies multi-trillion-dollar requirements
- •Key risk: time-to-breakthrough vs stranded physical infrastructure
- •Unwind thesis: more equity repricing than bank-driven credit collapse
- 34:27 – 49:53
VC realities: kingmaking skepticism, why margins matter (but aren’t destiny), and the new growth bar
Cahn challenges the idea that VCs can “anoint” winners, while acknowledging brand and recruiting can shift probabilities. He discusses gross margins as a directional signal that can improve as compute costs fall, then connects the current AI market to the “0–$100M revenue club” where breakout PMF shows up faster and compresses fundraising cadence.
- •Kingmaking is overstated; founders and PMF are decisive
- •VC brands help most via talent/recruiting and some flywheel effects
- •Margins: useful indicator of product built on top of models, but can expand over time
- •“0–100” growth trajectory as a modern signal of exceptional PMF
- 49:53 – 58:30
Talent and careers in the AI era: why 23-year-olds matter and the “memetic algorithm” is breaking
Cahn argues AI has leveled experience advantages—nobody has decades of GenAI tenure—so slope and AI-native intuition matter more than traditional seniority. He advises young people to account for AI as new data that breaks recursive career mimicry, and distinguishes between career “extractors” and high-impact “builders.”
- •Companies underestimate AI-native 23–25-year-olds in a fast-changing field
- •Hiring trade-offs: prefer visible risks (youth/inexperience) over hidden ones
- •Career choice is often recursive imitation; AI is disruptive new information
- •Builders focus on contribution first, which often leads to greater long-term rewards
- 58:30 – 1:07:33
Defense as “the next AI”: national champions, buyer concentration, and why Sequoia was late
Cahn concedes Sequoia was late to defense but argues the sector is earlier in its transformation curve than most people think. He frames Ukraine as a “Transformer moment,” predicts a small number of national-champion winners due to buyer concentration, and cites Sequoia’s bets in Israel (Kella) and Europe (Stark).
- •Defense parallels AI’s early days: inflection identified, mass adoption not yet
- •Ukraine war as a catalytic proof point for modernization needs
- •Deterrence as the goal; defense cycles tied to shifting world order
- •Thesis: concentrated outcomes—few winners per geography (national champions)
- 1:07:33 – 1:13:28
Quick-fire close: life updates, biggest miss, undervalued voice interface, and long-term AI optimism
In rapid Q&A, Cahn shares personal changes (learning to drive, becoming a father), a formative investing miss (Datadog) and how it reshaped focus, and a technology bet: voice as a major AI interface (Sesame). He ends by reaffirming AI as the defining story of the era despite near-term volatility.
- •Personal: driver’s license, fatherhood, and shifting priorities
- •Investing lesson from Datadog miss: focus deeply on a small set of top opportunities
- •Undervalued tech: voice-first AI interaction and relationship-style interfaces
- •Closing thesis: AI’s long-run transformation outweighs short-run cycle noise
