Dec 15, 2025 · 1h 11m · 20vc
a16z's David George on the Most Controversial Bet at a16z & Do Margins and Revenue Matter in AI? · 20VC with Harry Stebbings
0:00 / 0:00
▶ Watch on YouTube →
gold bands on the timeline = statements, start to end. Hover to read, click to jump. CC turns on captions
This video features a deep-dive interview between host Harry Stebbings and David George, General Partner at Andreessen Horowitz (a16z), discussing the economics of large venture capital funds, public versus private market dynamics, and a16z's core investment strategies. It covers their specific approaches to backing elite, "spiky" founders, evaluating high-growth AI companies, and managing the operational complexities of institutional-scale venture capital.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. Harry holds 23.4% of the talking time here. How this is scored →
Harry as informed peer 5.6 Guest teaching 5.5 Guest disagreement 2.9 Harry pushing back 5.0
speaking balance: gold is Harry, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 5:50 Guest rejects public/private competitive dynamic premise
David bluntly rejects Harry's thesis that staying private creates existential competitive risk, stating that public versus private status has little impact on market dynamics.
Hardest push from Harry ▶ 42:30 Host financial math pushback on AI app valuationsHarry runs step-down growth arithmetic on a $50M ARR company to show that paying current late-stage AI multiples yields poor opportunity-cost returns upon public exit.
Biggest teaching moment ▶ 11:45 Guest educates host on public small-cap ROIC decayDavid cites specific 30-year dataset on the Russell 2500 demonstrating that return on invested capital has fallen from 7.5% to 3% as top companies stay private.
Harry holds his own ▶ 8:54 Host hits back with public profitability compsHarry counters David's claim of cheaper public capital by comparing private valuations of Lovable and Rapid directly to Wix's $2B operating profit.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | Harry as informed peer | Guest teaching | Guest disagreement | Harry pushing back | Why |
|---|---|---|---|---|---|---|
| The Myth of Large Fund Sizes in Venture Capital | 5 | 6 | 3 | 4 | Harry opens by quoting partner Everett Randall's claim that large funds cannot deliver 5x returns. David counters forcefully with hard data on a16z returns, citing 7x from Databricks and 5x DPI from Coinbase in a single $1B fund, while breaking down the 47/53 gain split between early and late stage IPOs. | |
| Staying Private Longer as a Strategic VC Advantage | 6 | 5 | 4 | 7 | Harry cites the specific competitive battle between Axon and Flock Safety to question whether staying private hurts liquidity. When David downplays the public vs private distinction, Harry presses hard, expressing skepticism that public CEOs prefer being public. | |
| The Cost of Capital in Public vs. Private | 7 | 4 | 3 | 7 | When David claims public markets offer a cheaper cost of capital, Harry hits back with specific valuation comps, comparing Lovable and Rapid to Wix's $2B profit. David admits he is not close enough to those specific companies to argue. | |
| The Decay of Public Small-Cap Quality Over Time | 3 | 7 | 2 | 2 | David educates Harry on public market structural decay, citing how public companies have halved in 20 years and detailing how the Russell 2500 ROIC dropped from 7.5% to 3%. Harry listens and agrees. | |
| Endowment Asset Allocation Advice for a $10B Fund | 5 | 5 | 2 | 5 | Harry sets up a scenario as a $10B endowment LP and directly presses David on whether top private equity funds outperform top venture funds. The tone remains collaborative as Harry self-deprecatingly brings up passing on 11 Labs and Deel at seed. | |
| Justifying Venture Risk at Mature Growth Valuations | 6 | 5 | 2 | 5 | Harry questions taking venture risk at mature growth prices, probing whether liquidation preferences offset valuation risks. He cites internal team conversations with Brian Kim regarding the growth fund's charter to fix venture team errors. | |
| Investing in Strength of Strengths Over Weaknesses | 5 | 6 | 2 | 3 | David lays out Ben Horowitz's framework of investing in 'strength of strengths' rather than lack of weaknesses to overcome the TAM trap and fear of theoretical competition. Harry contributes his own lessons from missing 11 Labs and Deel. | |
| Analyzing the Three Disruptive Triggers of AI | 5 | 7 | 3 | 4 | Harry brings up the TAM trap and seat contraction in SaaS, leading David to detail the three disruptive triggers of AI (business model shift, UI/workflow, access to data). David educates on how incumbents face maximum risk from business model shifts. | |
| The Strategic Transition of Corporate Spend to AI | 6 | 6 | 4 | 6 | Harry argues that AI upside requires labor budgets shifting to tech budgets. David counters with the C.H. Robinson case study showing a 40% productivity bump, but Harry holds his ground by pointing to public market penalties for companies like monday.com and Duolingo that lack explicit labor replacement. | |
| The Speed of AI Revenue and the Retention Bar | 5 | 6 | 2 | 4 | Harry asks whether rapid ARR scaling (like Gamma's rise to $100M ARR) is transient. David explains that the bar for retention and engagement has risen significantly, contrasting hyper-fast organic acquisition with legacy paid growth. | |
| Evaluating the Classic SaaS Growth Playbook in AI | 6 | 5 | 3 | 6 | Harry asks if classic T2D3 SaaS growth is dead. When David suggests steady compounders still work, Harry pushes back on the opportunity cost of cash, arguing LP capital belongs in high-momentum category leaders like Gamma and Lovable. | |
| The Flaws of Capital as a Weapon in Venture | 5 | 7 | 4 | 4 | Harry introduces the concept of financier kingmaking. David rejects the premise, explaining why capital as a weapon failed at SoftBank by creating an adverse selection machine, contrasting it with organic preferential attachment. | |
| Venture Scale: Comparing a16z to Amazon's Retail Model | 6 | 6 | 3 | 5 | David compares venture scale to retail barbelling between Amazon and Chanel. Harry pokes fun at a16z being called 'Walmart,' which David deftly rebrands as Amazon. Harry then presses on the overcrowded AI customer support category. | |
| Navigating High Valuations and the Price of AI Growth | 8 | 4 | 4 | 8 | Harry challenges astronomical AI app valuations by walking through exact step-down growth math on a $50M ARR company, proving that standard public multiples yield poor dollar returns. David rejects the static model assumptions, pointing to 3x faster growth rates. | |
| Predicting the Future of AI App Gross Margins | 7 | 6 | 4 | 7 | When David talks about finding non-obvious beauty, Harry calls out a16z's top consensus positions in Anduril, Stripe, and OpenAI. David responds by detailing their initial entry points prior to consensus, such as investing in OpenAI before ChatGPT. | |
| The Foundation Model Wars: OpenAI Versus Anthropic | 5 | 5 | 2 | 3 | Harry asks about errors of omission, sharing his regret on Revolut. David names Anthropic, comparing foundation models to the oligopolistic cloud market where AWS, Azure, and GCP all thrive alongside each other. | |
| Constantly Reassessing the Ultimate Scale of Winners | 5 | 5 | 2 | 5 | Harry pushes on the upper pricing limit for OpenAI. David explains how initial growth fund assumptions on Databricks at $6B vastly understated the outcome, requiring constant reassessment of winner scale. | |
| Waymo: The Most Disagreed Upon Investment at a16z | 5 | 6 | 2 | 4 | Harry asks for internal disagreement stories. David reveals his initial pushback against investing in Waymo in 2020 due to high valuation, contrasting his conservative modeling with Mark Andreessen and Ben Horowitz's view on infinite TAM. | |
| Unpacking the Highly Controversial Investment in Flow | 6 | 6 | 4 | 7 | Harry confronts David on their controversial investment in Adam Neumann's Flow, noting that the rest of the world scratched their heads. David defends the deal using the 'strength of strengths' framework, citing Neumann's elite brand building in an unbranded residential rental market. | |
| Quickfire: Changing Minds and the Radiology AI Analogy | 6 | 5 | 4 | 6 | In quickfire, David argues foundation models will not swallow application software, using radiology work allocation as an analogy. Harry pushes back, citing recent moves by Gemini and OpenAI into vertical apps. | |
| The 'Killer' Founder Archetype | 6 | 4 | 2 | 3 | David describes Shiv from Abridge as a doctor who is also a total killer. Harry agrees and draws a parallel to Winston at Harvey, matching founder archetypes across vertical AI. | |
| The Best Picker at Andreessen Horowitz | 5 | 5 | 3 | 6 | Harry forces David to pick the best investor and sharpest internal player at a16z. David highlights Chris Dixon for clarity of thought and contrasts Mark Andreessen's futuristic vision with Ben Horowitz's executive management coaching. | |
| Decentralization Challenges in Scaling a16z | 6 | 5 | 2 | 5 | David candidly discusses the decentralization trade-offs required as a16z scaled, missing full-firm deal pitches. Harry asks if that hurt investor quality, before concluding with a humorous admission that he battered David with tough questions. |
Statements from this episode (48)
a16z gives startups more leeway on gross margins today
“On the gross margin point today, I'll say this. We give a little bit more of a pass than we used to.”
a16z's larger venture funds consistently outperform its smaller funds
“Our funds consistently beat small, large, Diversified, concentrated venture funds. So our larger funds have outperformed our smaller ones, and our larger ones actually have similar multiples of money to our smaller ones across, across strategies.”
a16z's best fund ever saw Databricks return 7x and Coinbase 5x
“Our best performing fund in the history of the firm is actually a one billion dollar fund. So it's a large fund, right? In that fund, Databricks has returned seven X the fund so far. Coinbase has returned already DPI five X of the fund.”
Tech private markets have reached over $5 trillion in market capitalization
“The private markets have grown 10 X over 10 years, right? It's over five trillion dollars of market cap now in our market.”
53% of dollar gains in top recent IPOs came post-Series C
“We actually just looked at the 50 top IPOs from 2017 to 2025. And if you disaggregate where the dollars of return come from, 47% of the dollars of gain happens between the seed in the series B and 53% of the dollars of gain happened from series C plus.”
Most AI value creation will happen in private markets, not public
“And if you look at what's happening in the private versus public markets now, the size of the winners from a new tech wave, that's going to happen in the private markets.”
Law enforcement tech shifted from a terrible category to highly lucrative
“And so it used to be that historically selling into law enforcement was a terrible category. And now it turns out that it's a wonderful category. If you actually have the most compelling products, you can get tremendous amounts of market share.”
a16z historically refuses to sell secondary shares in late-stage portfolio companies
“We could, but we historically have not. And so, you know, for the most part, for the companies that have decided to stay private, we've been really excited to stay in them, keep backing them. And that's probably the strategy that we'll, we'll continue to have.”
David George says no public CEO he knows regrets going public
“So, you know, I tell our CEOs all the time, I've been fortunate to work with a bunch of public companies. Never one of them has said, I regret going public.”
Recent a16z portfolio startups would access cheaper capital in public markets
“I can say in our portfolio, the companies that we have invested in over the last year or so, I'm pretty confident that if they were in the public markets, they'd probably have access to capital at a cheaper cost.”
Avoiding stock volatility is the primary advantage of staying private longer
“I think the biggest advantage is the avoidance of volatility in your stock price and sort of employee management. And so, you know, I think You know, if you can kind of steadily grow or control your stock price in the private markets, even if it's a slight dis…”
The number of US public companies halved over the last 20 years
“It turns out that the number of public companies has been cut in half over the last 20 years.”
Russell 2500 average return on invested capital halved over 30 years
“The ROIC of the Russell 2500 over the last 30 years, it's gone from seven and a half percent steadily down to three percent. So like more than cut in half.”
High-quality US private tech market cap dwarfs private equity technology
“If you just look at the size of The private technology, high quality companies, it dwarfs the size of private equity you know, private equity technology in the US.”
Top-performing venture capital funds historically outperform top private equity funds
“If you were to look at our returns or the top performing venture funds, let's just call it that relative to, you know, top performing PE funds, the top performing venture funds outperform.”
Harry Stebbings passed on investing in ElevenLabs at the seed stage
“Yeah, I turned it down at Seed.”
a16z paid growth prices for early-stage Character.AI to back Noam Shazeer
“My partner, Sarah led around a character AI and, you know, it was extremely early stage and, You know, we invested at a, you know, what you would call a gross stage price. But we knew that the likelihood of some degree of success in backing Nome was extremely …”
Only five founders globally justify paying growth-stage valuations at early stages
“The kinds of people like that, that warrant an investment decision, you know, a thought process like that, I think are extremely small. I mean, the list is five people, I think.”
Half of a16z's growth investments are follow-ons from its venture portfolio
“By the numbers, about half of what we do is follow-ons from existing venture companies. And then from a dollar standpoint, another 15% is follow-ons from existing growth stage companies. And then about a third or so is fully net new companies.”
VCs should invest in exceptional strengths rather than a lack of weaknesses
“When we make an investment, we should always be investing in strength of strengths as opposed to lack of weaknesses.”
Overweighting the fear of theoretical competition causes VCs to miss deals
“If you overweight the fear of future theoretical competition, you can always talk yourself out of making an investment.”
Business model shifts are the primary driver of AI software disruption
“If I were to rank order the level of disruption that is coming for these companies, business model shift is number one. And so, you know, we can talk about examples where that's most in practice today. You know, Sarah and Kimberly from our side led investment,…”
AI-native software companies will reach larger market caps than SaaS giants
“And so I have to think that this next wave probably presents the opportunity for this next generation to be much larger than the previous generation.”
C.H. Robinson saw a 40% AI-driven productivity gain in shipments
“They just disclosed in their last earnings that they saw a 40% productivity increased, increase measured in shipments per person per day in their core business since the end of 2022. 40% increase. It's an incredible and it's AI driven. And so what's actually h…”
Public markets assume legacy SaaS companies are doomed by AI
“In the public markets today, you are guilty until proven innocent. It's the full flip side of our criminal justice system where you are assumed that you are doomed by AI unless proven otherwise.”
a16z raised its underwriting bar for AI startups due to hypergrowth
“And so this is the bar has actually gone up significantly for us when we look at AI companies because it grown so fast. And so you can't actually look at years of renewal behavior, but you can look at shorter cycles of retention and you most importantly can lo…”
Breakout AI companies achieve massive scale with near-zero customer acquisition costs
“11 Labs has this, ChatGPT has this, XAI has this where it's organic customer acquisition or, you know, very low cost sales acquisition, Abridge, Harvey, companies where, like, the market is just absolutely starving for their product.”
Return on invested capital is the single best metric for evaluating companies
“The number one way to measure a company is ultimately return on invested capital.”
The original SoftBank Vision Fund correctly identified the massive AI opportunity
“They were early to figuring out that there would be a huge opportunity in AI. So, you know, they famously were in NVIDIA in, in that fund you know, and they did some really good investments, like Slack, like Garnet.”
Using massive capital as a weapon rarely succeeds in consumer tech
“Capital as a weapon in consumer, most of the time it's, it doesn't really work. Like I would say TikTok is maybe the exception, maybe Uber.”
Venture capital kingmaking creates adverse selection by backing fundamentally weaker startups
“We can king make if we just put the capital into the companies and then that will allow them to win. But that's a bit of an adverse selection machine where the companies that opt into that as their winning strategy are the ones that maybe don't have as good of…”
Unscaled mid-tier venture capital firms face high risk of being squeezed out
“Yeah, there's scale players, and then there's specialists. And I think, you know, we're obviously a scale player. I think the risk is, is everything in between, right? Department stores that have general merchandise, but don't have scale, for example. And that…”
Approximately half of SaaS and cloud markets are winner-take-most
“If you look at like SaaS and cloud markets, about half of them are winner take vast majority, like the overwhelming majority. And about half of them, you know, they're sort of a sort of breakup of market share.”
Top high-growth tech companies will trade above 6x public revenue multiples
“I also don't think that every great high growth company will end up trading for six times in the public markets. There are some that are going to trade higher based on very high growth rates or high cashflow.”
Winning AI applications are growing three times faster than predecessor SaaS startups
“For most of these companies that we've backed, these winning apps, they're growing faster, three X faster, you know, than predecessor SaaS and cloud companies.”
Gross margins for AI applications will rationalize and increase over time
“The history of technology inputs would suggest that the margins will rationalize and the margins are going to go up.”
Foundation model API providers will eventually operate as a high-margin oligopoly
“I think the market structure will end up sort of like cloud for the models where, you know, cloud costs for the average end customer are fine. And, you know, cloud's an oligopoly and they make high profits. I think the model companies, you know, that serve API…”
Traditional SaaS margins in an AI pitch indicate low AI feature usage
“If we ever see a company that pitches us as an AI company, and they have SaaS gross margins, we ask a lot of questions because it probably means that people aren't actually using the AI features.”
a16z's growth fund backed Anduril when it only sold border towers
“We invested in Andrel you know, in the growth fund when they had one program of record and it was border towers.”
a16z underwrote its later OpenAI investments primarily on consumer market potential
“As we under have kind of underwritten future rounds of open AI or later rounds of open AI is very much, you know, with the mind of consumer.”
Google and Facebook user monetization grew sevenfold over the last decade
“10 years ago, Google and Facebook were monetizing their users at like one-seventh of what they are today.”
Portfolio companies naturally diverge in product strategy more often than they converge
“The thing that we see that more often happens is companies diverge. More often than they converge. And so, you know, the perception of what a conflict can be in the future, like often doesn't come into play.”
Autonomous driving and robotics will be the largest markets created by AI
“Like I think autonomous driving and robotics are maybe the mother of all, you know, markets that are coming on AI.”
Independent AI application software companies will win across almost every category
“We fully changed our mind. I think there's going to be application software companies built on top of models in pretty much every direction.”
Radiologist jobs increased despite AI outperforming humans at reading scans
“Like AI has been able to do a better job than human radiologists. Like prior to this whole wave, like neural nets were able to do a better job than human radiologists and looking at scans. And yet since the, you know, this sort of proliferation of AI, the numb…”
Independent AI apps will thrive despite foundation models building competing products
“It's sort of like how AWS and the cloud, you know, have service offerings for basically everything that, that you could possibly have. And yet there's still tons of infrastructure companies that are independent.”
All-partner pitch meetings are not scalable for multi-sector venture capital firms
“I wouldn't change this, but one of the elements about us scaling has been, we've had to decentralize the way we run our business. Right. And so when I first joined the firm, you know, we used to sit around in partner meetings all day on Mondays and hear all th…”
Robotics will become the largest category in artificial intelligence within 10 years
“You know, we have not made a large investment in robotics, but I think it's going to be the largest category in AI, B to C, B to B. You know, there's still kind of debate on what the right form factors are. Whether it's at home help, whether it's industrial, l…”