The Twenty Minute VCa16z's David George on the Most Controversial Bet at a16z & Do Margins and Revenue Matter in AI?
CHAPTERS
- 0:00 – 5:19
Fund size myth: why mega-funds can still produce venture-level multiples
David pushes back on the idea that large venture funds can’t deliver 5x outcomes. He argues the private market has expanded enough that big winners and repeated late-stage value creation can still drive strong multiples.
- •a16z’s larger funds have outperformed smaller ones in their experience
- •Example: a $1B fund as the firm’s best performer; Databricks and Coinbase as major return drivers
- •Value creation isn’t only early-stage: meaningful dollar gains occur in Series C+
- •Private market capitalization has grown dramatically, expanding the runway for large funds
- •Tech waves create larger-than-expected outcome sizes, enabling mega-fund economics
- 5:19 – 8:15
Staying private longer: competition, ownership, and liquidity trade-offs
Harry challenges whether longer private timelines increase competitive risk and reduce liquidity. David argues competition is largely independent of public/private status and says staying private can let investors increase ownership rather than exit early.
- •Competition dynamics (e.g., Flock vs Axon) aren’t primarily driven by being private vs public
- •Staying private can benefit investors by enabling ownership increases over time
- •a16z historically hasn’t taken much off the table in private secondaries
- •Some “stay private forever” narratives are overblown; many CEOs still want to go public
- •Public-company experience: David claims few CEOs he knows regret going public
- 8:15 – 11:13
Is public capital cheaper? The real benefits of going public vs remaining private
They debate whether public markets offer a lower cost of capital today. David contends public markets can still be cheaper, while private markets may offer accessibility at a higher implied cost; he also explains why some elite companies prefer private stability.
- •David’s view: public markets can provide cheaper cost of capital in many cases
- •Private markets can offer abundant capital but potentially at a more expensive cost
- •For companies like Stripe, avoiding stock-price volatility and employee management issues is a major private-market benefit
- •Transparency and storytelling can mitigate some public-market drawbacks
- •Examples of private-market advantage: Stripe, SpaceX, Databricks sustaining controlled pricing/ownership dynamics
- 11:13 – 19:42
Asset-class shift: shrinking public markets and the rise of private tech as the 'big leagues'
David argues the biggest structural change is that many high-quality, high-growth companies are now staying private longer, leaving fewer attractive small-cap opportunities in public markets. He uses declining public-company counts and ROIC trends to justify why venture/growth matters more to institutional allocators.
- •Number of public companies has fallen significantly over ~20 years
- •Returns and quality (ROIC) in broad small-cap indices have deteriorated over decades
- •Private tech now rivals or exceeds traditional PE tech opportunity sets
- •Venture/growth now requires companies to scale further privately (multi-product, multi-channel, international)
- •AI accelerates the pace at which private companies must mature operationally
- 19:42 – 22:10
How a16z Growth works: 'fix the mistake fund' and follow-on-heavy strategy
David explains the growth fund’s role in correcting early-stage misses in partnership with the venture team. He breaks down how much of the fund is follow-ons versus net-new and why pre-existing founder relationships matter for later-stage underwriting.
- •Growth fund explicitly helps address errors of omission from earlier rounds
- •Close coordination with early-stage teams (shared meetings, constant deal retrospectives)
- •Portfolio mix: roughly half follow-ons from venture; additional follow-ons from growth-originated positions
- •Net-new investments still typically involve founders known from earlier interactions
- •Example pattern: missing intermediate rounds (e.g., Deel) and re-entering later
- 22:10 – 30:23
a16z’s #1 rule: invest in 'strength of strengths,' not absence of weaknesses
They discuss what missing deals teaches and why VCs over-index on risks like hypothetical competition or market-size skepticism. David credits Ben Horowitz’s framework: back spiky founders and accept imperfections if the core strengths are extraordinary.
- •Core lesson: prioritize spiking founder/company strengths over eliminating weaknesses
- •Common mistake: over-weighting fear of theoretical future competition ('isn’t Google going to do it?')
- •Another recurring miss: underestimating market size (TAM)
- •Great companies can expand markets and outgrow prior mental models
- •The best decisions often look uncomfortable when evaluated with conventional checklists
- 30:23 – 32:28
Does revenue still matter in AI? Retention, engagement, and organic distribution as the new bar
With AI companies hitting revenue milestones faster, David says revenue only signals durability when paired with high engagement and retention. He argues the evaluation bar is higher because renewal history is shorter, so leading indicators matter more.
- •Revenue remains meaningful if retention and engagement are strong
- •AI changes diligence: focus shifts from long renewal histories to short-cycle retention + usage depth
- •Organic/low-cost acquisition is a key quality signal (market pull)
- •Examples cited: Gamma’s engagement + organic growth; similar dynamics in ElevenLabs, ChatGPT, xAI, Harvey
- •Fast growth can be durable, but requires more discerning analysis than prior SaaS eras
- 32:28 – 34:55
Growth expectations in the AI era: ROIC, customer acquisition efficiency, and momentum vs peers
Harry probes whether classic 'treble, treble, double, double' growth expectations still apply. David reframes around return on invested capital and argues that in fast-moving AI categories, momentum is strategic because it can help form moats—especially relative to direct competitors.
- •Primary metric: return on invested capital; early-stage proxy is acquisition efficiency
- •Not every company must go 0→100 quickly; market speed varies by category
- •In rapidly moving markets, slow growth increases strategic risk
- •Momentum matters most relative to peer set and category competition
- •Market pull—rather than paid sales hiring—drives the most compelling hypergrowth stories
- 34:55 – 39:26
Kingmaking debate: why capital alone rarely anoints winners (and when scale helps)
David argues investors should back companies that are already trending toward leadership rather than assume funding will create leadership. He contrasts preferential attachment and brand effects with SoftBank-style 'capital as a weapon,' which can create adverse selection.
- •Best thesis: invest in the winner; weak thesis: your money will make them win
- •Preferential attachment: leaders attract more resources, talent, and distribution advantages
- •Capital-as-weapon is hard in enterprise (requires real hiring/execution) and often ineffective in consumer
- •SoftBank Vision Fund: strong picks (e.g., NVIDIA exposure) but flawed assumption that cash alone creates dominance
- •a16z brand can function as a 'seal of approval' that helps hiring and credibility
- 39:26 – 43:48
Crowded categories (e.g., AI customer support): why the market is pulling and outcomes can still be huge
Harry questions the explosion of AI customer support startups and the logic of paying ahead of fundamentals. David argues the category is already better/faster/cheaper today, with strong conversion signals, and notes many SaaS markets are not winner-take-all—yet can still produce enormous companies.
- •Customer support AI works now with current model quality and economics
- •a16z sees strong enterprise interest and conversion in portfolio demos (EBCs)
- •Founders and execution speed matter heavily in crowded markets
- •Many cloud/SaaS categories split share rather than go pure winner-take-all (payroll as analogy)
- •David is structurally optimistic: best companies can surprise on growth rate and ultimate scale
- 43:48 – 47:02
Do margins matter in AI apps? Gross margin uncertainty, token economics, and 'SaaS-like margins' as a red flag
They tackle the criticism that AI applications have weak margins. David expects margins to improve over time, but notes token costs and reasoning usage create a muddy near-term picture; importantly, he says unusually high SaaS-like margins can indicate users aren’t actually using AI features.
- •Historical pattern: technology input costs decline and margins tend to rise over time
- •Near-term uncertainty: token costs down, but token usage up with reasoning models
- •Expected end-state: model APIs become an oligopoly like cloud; end customers still well served
- •a16z gives more margin 'pass' today than in prior SaaS eras if value + growth are strong
- •Counterintuitive diligence: SaaS-level gross margins can be a warning sign for weak AI usage
- 47:02 – 49:46
Biggest miss and model-market structure: Anthropic vs OpenAI (consumer vs enterprise)
David names Anthropic as a key omission and frames the model layer as an oligopoly similar to AWS/Azure/GCP. He expects differentiation between OpenAI and Anthropic while acknowledging intense competition across consumer, B2B APIs, and the application layer.
- •Anthropic cited as a notable miss; model market may support multiple massive winners
- •Analogy: owning AWS, Azure, and GCP would have been great—models could be similar
- •Expect divergence: OpenAI strongest in consumer brand (ChatGPT), Anthropic leaning harder into B2B/dev
- •B2B API and app-layer expansion will be highly competitive for both
- •Google likely remains a meaningful competitor in parts of the stack
- 49:46 – 52:50
How to price generational winners: OpenAI entry price, forecasting limits, and founder-led multi-product expansion
Harry asks when OpenAI becomes too expensive; David argues the answer requires continuous reassessment because the best companies repeatedly exceed expectations. He cites historical surprises in monetization expansion (Google/Facebook) and emphasizes backing teams that can discover adjacent products.
- •Entry price decisions require ongoing reassessment as opportunity size evolves
- •Databricks example: early investment case understated eventual magnitude
- •Monetization is hard to forecast; best platforms can expand ARPU dramatically over time
- •Preference for companies with clear theories for market expansion and/or founder advantage in new products
- •Conflict management at scale: a16z tries to avoid investing in direct competitors, especially with board roles
- 52:50 – 55:59
Most controversial bet: Waymo—internal disagreement, valuation skepticism, and autonomy as a 'mother of all markets'
David recounts the firm’s early-2020 Waymo investment and the internal debate with Mark and Ben about valuation versus market size. The compromise was a smaller initial check to maintain relationship, followed by a larger later investment as product reality and safety data strengthened the thesis.
- •a16z was the only VC in Waymo’s early-2020 round; product demos felt 'magic' even then
- •David worried valuation was too high; Mark/Ben emphasized limitless TAM and category leadership
- •Compromise strategy: invest smaller, stay close, then scale exposure later
- •Safety narrative: data suggests Waymo may be 7–10x safer than human drivers; regulatory implications
- •Autonomous driving + robotics framed as among the largest AI markets ahead
- 55:59 – 59:17
Why $300M into Adam Neumann’s Flow: founder 'spikes,' branding insight, and the renter experience thesis
Harry presses on Flow as a widely questioned decision. David ties it back to the firm’s 'strength of strengths' philosophy, arguing Neumann is exceptional at brand, product, hiring, and company building—and that renting is a massive, largely unbranded consumer category ripe for a premium experience.
- •Flow thesis anchored in backing an unusually high-skill founder despite known weaknesses
- •Neumann’s claimed spikes: brand building, company building, product instincts, and hiring
- •Market insight: rent is a top household expense but lacks a coherent branded experience
- •Bet that better product + brand can create a premium renter offering and scalable business model
- •Status update: team has proven initial value prop; focus shifts to scaling execution
- 59:17 – 1:11:52
Quick-fire wrap: models vs apps, memorable founders, internal dynamics, and what David’s excited about next
In lightning round format, David shares mind-changes about how models relate to application software, plus takes on investing craft and a16z’s internal scaling. He closes with personal optimism about AI-enabled personal health management and robotics as major upcoming opportunity areas.
- •Mind change: models won’t 'eat everything'—apps must solve surrounding workflows (radiology example)
- •Memorable founder meeting: Abridge’s Shiv as domain-authentic and aggressively execution-oriented
- •Internal strengths: Dixon’s clarity on early-stage approach; Mark’s future-sight vs Ben’s executive coaching
- •Scaling downside: decentralization reduced shared partner-meeting signal, requiring more deliberate coordination
- •Excitement areas: proactive personal health coaching and robotics as the biggest AI category over the next decade
