CHAPTERS
- 0:03 – 1:33
Why tech dominates markets and why growth investing moved private
David George opens with the founding premise of the growth fund: technology’s share of global market cap keeps rising, and the best companies are staying private for much longer. That combination expands the investable universe—but also raises the bar on generating realized returns (DPI) in a world of longer hold periods.
- •US tech companies make up a large share of the most valuable global firms
- •Companies staying private longer expands private-market opportunity
- •Staying private longer is a double-edged sword: more access vs. DPI pressure
- •AI is the major change vs. when the growth fund started
- 1:33 – 3:04
AI infrastructure buildout: unprecedented CapEx as the new foundation
The discussion turns to the scale and structure of the AI buildout, emphasizing how different it is from prior cycles. Big Tech’s run-rate CapEx—hundreds of billions annually—largely funds AI data centers and infrastructure, creating a platform others can build on.
- •Big Tech is deploying massive CapEx into AI infrastructure and data centers
- •Scale of groundwork exceeds prior technology cycles
- •Large incumbents can tolerate overbuild risk better than weaker builders
- •Infrastructure buildout benefits downstream application and tooling companies
- 3:04 – 4:04
Input costs collapsing while model capabilities surge (faster than Moore’s Law)
David argues AI is uniquely powerful because cost and quality are improving simultaneously and rapidly. Model access costs have dropped ~99% in two years while frontier capability compounds quickly, creating a fertile environment for new products and categories.
- •Model access costs down ~99% over two years (~100x decline)
- •Frontier model capability improves on a rapid cadence (doubling-like progress)
- •Falling input cost + rising quality accelerates new product creation
- •Long-run view: AI becomes an ambient utility like electricity/Wi‑Fi
- 4:04 – 7:55
AI’s market size: shifting from software spend to white-collar payroll economics
They frame AI’s upside as far larger than traditional software because it targets a much bigger portion of GDP: labor. Most value accrues to end users, but even a fraction captured by vendors can create enormous market caps.
- •Mobile+cloud created ~10T in value; AI could be larger
- •Software spend ~1% of GDP vs. white-collar payroll ~20% of GDP
- •Rule of thumb: ~90% of value to customers, ~10% to suppliers
- •Consumer surplus examples: iPhone willingness-to-pay; Google value vs. monetization
- 7:55 – 11:53
Why this cycle differs from dot-com: demand signals and instant global distribution
Erik raises concerns about overbuild and historical parallels to early-2000s infrastructure gluts. David distinguishes the current cycle by highlighting stronger builders/tenants, lower systemic leverage risk (so far), and far faster demand realization due to internet+cloud distribution.
- •Key risk lens: leverage and who funds the buildout (banks/private debt/insurers)
- •AI rides on existing internet and cloud, enabling immediate global reach
- •ChatGPT scaled to massive usage far faster than earlier internet platforms
- •Faster observable demand reduces risk that supply buildout goes unused
- 11:53 – 18:17
Monetization reconfiguration: price discrimination, subscriptions, and AI commerce flows
They explore how AI products may monetize differently from prior consumer internet platforms. David points to emerging price discrimination (low-cost regions vs premium tiers) and potential new forms of ad/affiliate monetization, especially as AI becomes a superior discovery and shopping interface.
- •AI enables more granular price discrimination than classic consumer platforms
- •OpenAI tiering examples: low-cost regional plans vs $200–$300 premium tiers
- •AI-based “deep research” changes shopping/search behavior
- •Downstream impact: declining referral traffic to some websites as AI summarizes results
- 18:17 – 22:23
The biggest bottlenecks for AI: energy, construction speed, and cooling
A Q&A segment focuses on constraints that may cap AI expansion. Energy is identified as the near-term bottleneck, with nuclear and localized gas generation as promising solutions; construction logistics and cooling are highlighted as major limiting factors as density rises.
- •Energy availability is a core constraint for data center scale-up
- •Nuclear is a key bullish focus (restarts, siting near plants)
- •Physical buildout speed matters: labor, generators, supply chain improvisation
- •Cooling is an underappreciated next bottleneck (preventing chip/infra overheating)
- 22:23 – 28:46
Evaluating AI business models: retention, acquisition, and gross margin realism
David lays out how they underwrite AI-native applications amid high outcome variance. The core is enduring customer love (gross retention) and efficient acquisition; they’re somewhat more flexible on current gross margins because model input costs are expected to decline with competition.
- •AI era increases variance: harder winner-picking, bigger payoffs and risks
- •Top underwriting metrics: gross retention and ease of acquisition (pull vs push)
- •Gross margins matter, but may improve as model costs fall and competition persists
- •Coding is a key battleground; model competition (OpenAI/Anthropic/Google) pressures prices
- 28:46 – 34:00
OpenAI pricing vs cash burn: monetization upside and R&D discipline
They address fears that consumer pricing will compress while model labs burn unprecedented cash. David argues monetization is early (many users, few payers), leaving more upside in expanding paid conversion and monetizing free usage; he also expects competitive pressures to enforce economic discipline in R&D over time.
- •More upside in monetizing the user base than downside from near-term price pressure
- •p×q framing: user growth has limits, but price/ARPU can rise via segmentation
- •Big burn is largely R&D for frontier capability; consumer distribution offers durability
- •Labs increasingly act as rational capitalists due to competition, not pure research maximalists
- 34:00 – 38:50
Durability of AI app revenue: what’s sticky vs easily swapped
David contrasts sticky application categories (deep workflow integration) with more experimental or low-commitment usage. The stickier cases embed rules, integrations, brand constraints, and enterprise workflows; the less sticky cases are lightweight prototyping or tools with minimal switching friction.
- •Stickier examples: medical scribe, customer support workflows, financial analysis
- •Drivers of stickiness: integrations, rules engines, workflows, enterprise features
- •Brand constraints (tone/interaction style) increase switching costs in support
- •Less sticky: experimental internal tools and low-end prototyping/vibe coding
- 38:50 – 43:18
Pricing models in AI software: why “pay per task” is still early
They examine seat-based vs usage-based vs outcome/task-based pricing. David notes true task monetization is most advanced in customer support but remains early elsewhere; customers still prefer familiar pricing, and measuring “task value” objectively is difficult, which may leave surplus with buyers.
- •AI pricing discourse is ahead of reality; broad model innovation is limited so far
- •Customer support is furthest along in outcome-based/task pricing
- •Most buyers still want seats and consumption; vendors must meet the market
- •Hard-to-measure task value + competition likely leaves significant surplus with customers
- 43:18 – 45:49
Companies staying private longer: implications for exits, DPI, and secondary liquidity
The conversation shifts to market structure: private company lifetimes are lengthening even as growth accelerates, moving high-growth opportunity into private markets. David discusses how private markets mimic public liquidity via tenders and structured secondaries, while balancing what’s best for the company with the fund’s DPI needs.
- •Time to IPO has extended to ~14+ years; private market value has scaled dramatically
- •Only a small share of public software/internet companies grow >25% forward
- •Private-market tenders help with talent retention and partial liquidity
- •a16z focuses on company-first decisions while maintaining DPI discipline
- 45:49 – 53:01
Portfolio construction and AI strategy: momentum leaders vs elite teams with asymmetric risk
David outlines two AI investing buckets: undeniable breakout companies and very early bets on a tiny set of world-class teams. The latter has higher business variance but potentially asymmetric capital risk due to team quality and talent demand, while access and early relationships help secure allocations later.
- •Bucket 1: breakout momentum companies (e.g., Cursor, Decagon, Eleven, Abridge)
- •Bucket 2: early growth bets in top-tier teams (top ~5 globally), not a broad category
- •Early relationships create allocation access and “ball control” in later rounds
- •Portfolio balance: avoid only ‘safe’ outcomes; seek meaningful upside with controlled downside
- 53:01 – 56:39
Disruption framework for public software incumbents (and why public investing is rare)
Asked which public software companies are safe or vulnerable, David offers a framework rather than naming winners/losers. Disruption likely requires a reimagined UI/UX, new data advantages (especially unstructured), and potentially a disruptive business model—an uncommon combination.
- •Public investments are rare; private opportunity cost is too high unless thesis is exceptional
- •Potential disruption ingredient #1: proactive AI-first UI/UX vs form-based workflows
- •Ingredient #2: access to new data moats, especially unstructured data
- •Ingredient #3: disruptive pricing/business model (still early) to attack seat-based incumbents
- 56:39 – 1:03:40
Team culture, collaboration with early stage, and sector mix (AI, dynamism, health, crypto)
They close with how the growth team operates as a small, high-leverage unit tightly integrated with a16z’s early-stage practices. David previews likely portfolio mix driven by upstream early-stage conviction—AI infra/apps first, then American dynamism, selective AI-health, and growth-stage crypto alongside the dedicated crypto team.
- •Alpha sources: access (early relationships) and insight (market/product theses)
- •Close integration: most growth deals benefit from early-stage relationships
- •Expected mix: AI infra/apps largest; American dynamism next; growing AI-health interest
- •Crypto exposure is high-conviction, coordinated with the crypto team; stablecoins highlighted
