Jul 31, 2025 · 1h 28m · 20vc
Why Apple Needs a Management Overhaul & Why Google is Catching Up with Hyperscalers · 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
In this episode of 20VC, host Harry Stebbings sits down with venture capital veterans Rory O'Driscoll O'Driscoll and Jason Lemkin to critique the massive shifts in the venture capital industry, the intense developer AI war, and the heavy macroeconomic imbalances of AI capital expenditure versus actual revenue.
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 12.9% of the talking time here. How this is scored →
Harry as informed peer 5.9 Guest teaching 5.3 Guest disagreement 3.4 Harry pushing back 4.4
speaking balance: gold is Harry, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 5:47 Rory calls out Harry for avoiding Jason's core question
Rory directly interrupts Harry's portfolio listing, telling him to stop because he avoided Jason's question about solo fund trade-offs.
Hardest push from Harry ▶ 51:11 Harry rejects Rory's dismissal of big tech oligopoly powerHarry forcefully pushes back against Rory's defense of big tech, citing Meta's unlimited hiring budgets pricing startups out of AI talent.
Biggest teaching moment ▶ 1:21:21 Rory analyzes AI CapEx using historical GDP boom metricsRory educates the panel with Paul Kedrosky's historical data, comparing current AI CapEx at 1.2% of GDP against the 1999 Telecom boom and 1840s Railroad mania.
Harry holds his own ▶ 5:09 Harry cites exact Benchmark portfolio ARR metricsHarry demonstrates deep venture expertise by listing specific ARR numbers and deal metrics across Benchmark's recent AI investments.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | Harry as informed peer | Guest teaching | Guest disagreement | Harry pushing back | Why |
|---|---|---|---|---|---|---|
| Show Subscribe Call-to-Action | 5 | 3 | 2 | 2 | Harry frames the news around Victor leaving Benchmark and names remaining partners to seed the topic. The guests share peer insights on brand value and solo GP incentives. | |
| Solo GP Funds vs. Institutional Brands | 7 | 6 | 6 | 6 | Harry reels off Benchmark AI portfolio ARR figures from memory, but Rory cuts him off for dodging Jason's main question on solo fund trade-offs. Harry holds his ground on why solo funds appeal to top investors. | |
| Elad Gil, Rapid Capital Deployment, and Short-Term Relevance | 7 | 4 | 2 | 4 | Harry cites specific deals across Elad Gil's portfolio and draws an analogy to Josh Kushner's real estate strategy. Rory expands on the tension between rapid deployment and long-term LP returns. | |
| Early-Stage Partnerships vs. Late-Stage Dictatorial Structures | 6 | 6 | 3 | 7 | Harry bluntly asks Rory whether dictatorial single-GP firms make him question his firm's equal partnership structure. Rory explains why early-stage investing structurally requires peer partnerships. | |
| Re-evaluating Tiger Global and the 2021 SaaS Bubble | 7 | 5 | 4 | 6 | Harry pushes back against Rory's critique of Tiger Global, pointing out that broad deployment captured top assets like OpenAI and Scale. Jason contextualizes the unsustainable 2021 SaaS growth environment. | |
| Anthropic's Growth, Vibe Coding, and Developer Token Spend | 6 | 6 | 4 | 5 | Jason shares direct vibe-coding experience to argue developer spend will reach $10k per month. Harry pushes back using Moore's Law, asking if falling token prices will lower per-developer spend over time. | |
| Cursor vs. GitHub and the Battle for Developer AI Prototyping | 7 | 4 | 4 | 6 | Harry directly asks if GitHub Copilot's loss to Cursor represents a massive execution failure for Microsoft. Guests agree, pointing out Microsoft's struggles to protect its early developer AI lead. | |
| Hyperscaler Showdown & The Economics of AI CapEx | 7 | 6 | 3 | 4 | Harry frames a buy/short trade across hyperscalers and references David Cahn's CapEx vs revenue analysis. Rory analyzes cloud growth trajectories and incoming depreciation headwinds. | |
| Are Tech Oligopolies Too Powerful? | 6 | 6 | 5 | 7 | Harry challenges Rory's defense of big tech, arguing Meta's unlimited hiring budgets price startups out of talent. Jason educates on oligopoly economics, noting feature competition reinvests profits into R&D. | |
| Why Apple Needs a Management Overhaul | 5 | 6 | 5 | 3 | Rory suggests Apple's board needs to evaluate if executive leadership is too old to execute on AI. Jason counters that Apple's 40% profit share from the App Store lets them capture AI upside without building models. | |
| Zuckerberg's Talent Siege & Existential Meta Bets | 5 | 5 | 2 | 3 | Harry asks when Mark Zuckerberg's talent acquisition spree will end. Rory explains that spending billions on AI researchers is rounding-error math compared to $40B annual CapEx budgets. | |
| The Margin Trap for Mature Tech Companies | 3 | 6 | 1 | 1 | Jason explains how public margin expectations trap companies like Adobe and Zoom, while founder-led firms like Meta and Oracle get hall passes. Rory adds Warren Buffett's framework on public capital allocation. | |
| The Figma IPO & Traditional SaaS Zeitgeist | 7 | 5 | 4 | 5 | Harry cites Figma's adjusted IPO pricing range and billion-dollar returns for early VCs. Jason contrasts the intense industry hype of the 2022 Adobe deal with today's shift toward AI prototyping. | |
| Quick-Fire Game: Cursor & Lovable Predictions | 5 | 4 | 4 | 3 | Harry leads an over/under game on Cursor and Lovable ARR targets. Jason jokes that Lovable could face a valuation down-round, prompting Harry to playfully shout for the producers to cut the audio. | |
| OpenAI's Valuation & The Fight for Global Capital | 7 | 5 | 3 | 6 | Harry takes a contrarian stance predicting OpenAI will hit an $800B valuation, citing GPT-5, Codex, and Jony Ive's hardware. Rory predicts under, noting compressing valuation step-up multiples. | |
| AI CapEx vs. Historical Economic Booms | 5 | 8 | 2 | 3 | Rory schools the room with Paul Kedrosky's macro dataset comparing AI CapEx at 1.2% of GDP against the 1999 Telecom bubble and 1840s Railway mania. Jason adds data on Apollo space program funding peaks. |
Statements from this episode (44)
O'Driscoll: Google has executed best in AI among major tech giants
“Ironically, Google, who we all piss on, has executed the best of the four. They have a model that works.”
O'Driscoll: Tech giants are on the back foot in AI
“I don't think these guys are too powerful. If anything, I think there are a bunch of rich people on the back foot behind the new trend, desperately trying to catch up.”
Lemkin: Venture partners with hot hands should leave funds early
“If you are going to leave a fund, and you can, you have a hot hand, better to do it early. Like staying longer doesn't help. Like it might seem like it helps the fund. It doesn't help you. You're probably not going to leave with much carry, right? You're going…”
O'Driscoll: Benchmark will hire replacement partners and remain successful
“Benchmark will be fine. So let's start with that. They'll find two other people to fill those slots and they'll be fine because they've been fine for 30 years.”
Stebbings: Benchmark has executed the best AI strategy of any VC firm
“I don't think any firm has played AI in the last 18 months as well as Benchmark have done, despite the team churn that they've had.”
Stebbings: Benchmark AI portfolio companies HeyGen, Sierra, and Fireworks hit $100M+ ARR
“McCaw over a hundred million in ARR. Hey Jen hitting a hundred million in ARR. Fireworks a hundred and forty million in ARR. Sierra with Brett Taylor over a hundred million in ARR.”
O'Driscoll: The next downturn will prove why traditional VC governance existed
“Rules of thumb, our heuristics are in place because across cycles, they've been proven to be correct. And my guess is some part of this trend makes total sense. And some part of it, you look back after next down and turn and say, oh yeah, that's why we had tha…”
O'Driscoll: No venture fund launched since 2010 has experienced a full cycle
“There's no data on anything through a cycle yet for anything that started from about 2010 on, right?”
O'Driscoll: Tiger Global and SoftBank are shadows of their former selves
“Tiger's the obvious example. Now, they're still around, but a shadow of their former selves. Their quote unquote relevance in 2021 meant that they saw every deal that year, but they picked badly, so they got sent home. Same thing with SoftBank.”
O'Driscoll: Market relevance in VC is short-term; only being right matters
“Relevance is a short term thing. And in the end, as I can say about this job, the only thing that's true about this job over the medium term is you have to be right.”
Stebbings: Recent top breakout VC firms are all single-person led
“When you look at, Bundly, the best performing managers, or the breakout managers the last few years, I think you put probably Thrive, Green Oaks, Founders Fund, and Eli Gill definitely up there. And all four of them are one person led.”
O'Driscoll: Late-stage venture carries 100% correlated valuation risk
“Remember you're only taking one risk as a late stage investor, which is valuation risk. And when it goes wrong in a hundred percent correlated fashion. Everything blows up in your face.”
Stebbings: Tiger Global will outperform expectations due to OpenAI, Scale AI
“Tiger will do better than people think, given their position, given their positions in open AI scale.”
Lemkin: The average public SaaS company grew 70% in 2021
“The average public SaaS company in 2021 was growing 70%.”
O'Driscoll: Early-stage firms scaling late-stage funds will outlast crossover investors
“And in my view, better late stage, I should put this better, more likely survivors than the late stage players who dash in from the public, think it's all the financial game and then get blown up, right?”
Stebbings: Thrive and Greenoaks made $500M+ buying distressed Carvana stock
“They made half a billion dollars plus from Carvana in the public markets buying when it was in the dumps and then writing that up.”
O'Driscoll: Anthropic grew revenue from $1B to $4B year-over-year
“I believe that there was a billion late last year and four billion this year.”
Lemkin: Tech companies will give developers $10k monthly in AI credits
“I think every developer at a top tech company growing is going to give them 10,000 dollars a month of AI credits. Everyone's going to get 10,000, not 200, which is what we're thinking. We thought that was a lot a couple months ago. They're all going to get 10,…”
O'Driscoll: Pure developer tools cannot support a $1T valuation
“I don't think you get to a month of a trillion dollars if you're just selling stuff to developers, which isn't to say I don't think you do extraordinarily well.”
Lemkin: Developer AI total addressable market will grow 50x
“It may not add up to two trillion, but I think the T TAM, the true TAM is 50 X what it is today without question”
O'Driscoll: Letting competitors enter your monopoly market is outright failure
“The thing about in any market, when you Microsoft, if you start with a monopoly and you fast forward three years and someone exists, by definition, you screwed up. Your job as a monopoly was don't lose the monopoly, right?”
O'Driscoll: AI CapEx reaches $600B against only $30B in revenue
“How long can you continue investing six hundred billion a year against a thirty billion dollar growing revenue line?”
O'Driscoll: Nvidia chip depreciation will trigger an AI CapEx slowdown
“Someone, at some point someone's going to depreciate all these Nvidia chips and someone's going to have to pay for them. Right? So I do. And I'm sorry to get, I do think there might be a slowdown at some point in time, but right now there's no evidence of it w…”
Lemkin: Tech oligopolies maximize innovation by competing on features, not price
“Is it because oligopolies maximize innovation? They don't maximize discounts, but they pour all the money back into R&D because they're competing on features, not price, right? If you believe in technology, you might almost want oligopolies, right?”
Lemkin: iPhone and App Store drive 75% of Apple revenue
“25% of the revenue is from the App Store, 40% of its profit, right? And 75% of its revenue is iPhone and App Store.”
O'Driscoll: AI talent bounties will end as marginal returns shrink
“My, I mean, my assumption is that at some point, the marginal return on the next scientist gets negative, so you probably stop spending a hundred million bucks a pop, but you know, I get what he's done.”
O'Driscoll: Google failing at AI would destroy its search business
“Like I would argue for Google, if they misfired on AI, they wouldn't have search.”
O'Driscoll: Misfiring on AI is not existential for Apple
“If Apple misfires on an open AI type AI product, they'll still be making hardware and it won't be existential.”
Lemkin: Mature public tech companies are trapped by high margin expectations
“It is almost impossible for most public companies to actually spend the cash there they're generating. So Zuck's got a triple down here because he's getting also a hall pass. The market's saying it's okay to spend everything, right?”
O'Driscoll: Rational capital allocation is management's key job post-growth
“Once a company becomes ex-growth, the most important thing to assess about management is their ability to rationally allocate capital, including sending that capital back to the shareholders”
O'Driscoll: Figma will price well and pop on its IPO
“My sense is it's hard not to imagine this thing pricing and then popping very well because the structure of how they went at it is such that they will almost certainly be that money left on the table feeling”
O'Driscoll: Four VC firms will each return $1B from Figma
“The reality of four VC firms, each making a billion dollars, that'll focus the mind.”
O'Driscoll: Nobody will build non-AI B2B software going forward
“Startup activity in the space that Figma plays is not going to be a thing anymore. You're not going to do a non, imagine you're not going to build a non AI collaborative design tool. You're not going to build a non AI, any kind of B to B software.”
O'Driscoll: VC holding periods now outlast full tech cycles
“Our holding period is now longer than the tech cycle. That's a terrifying fact. Figma's holding period is 12 or 13 years”
Lemkin: Figma and AI coding tools will compete within 12 months
“Today, they're great partners. It doesn't make sense that they will be great partners for prototyping in 12 months. It makes no sense.”
Lemkin: Replit's gross margins are positive and around 60 percent
“As someone who keeps paying these replet checks, they ain't negative. I guarantee you, reckless, reckless, gross margins are pretty good. Okay. They're pretty, I bet they're in the sixties even today.”
Lemkin: Lovable will reach $400M ARR by end of next year
“First, I have no doubts. That's easier to do the math. One to a hundred in six months. I mean, that would be a colossal F-up if it doesn't get to 400.”
O'Driscoll: OpenAI will not surpass an $800B valuation by late 2026
“Under. I'll tell you what I'm doing mentally. I'm just, I mean, there's no logic. I'm just looking at the step ups in the rounds as you'd expect are getting smaller, right? You know, you get two or five X and then you have, you go from 80 to one 50, which is a…”
Stebbings: OpenAI will top $800B valuation driven by new products
“I'll take the contrarian on you there and say that because of GPT-V and because of Kodak's being two very significant breakthroughs, they'll have enough, plus Johnny Ive coming out with something, a first product, the three combined will be enough to get enoug…”
Lemkin: Stakeholders will engineer an $800B valuation for OpenAI if needed
“And if open AI needs 800 to survive, open AI will find 800. It will find it. It may be through investors that get other things. It may be through existing investors. It may be through weird combinations of offshore money, but no one, No, no, no one can afford …”
O'Driscoll: AI CapEx relative to GDP exceeds peak dot-com bandwidth spending
“So we're running now at about 1.2% of GDP, which is basically 1.2% of GDP is going into making CapEx data centers, which is a little above the 1.1% of GDP that was going into bandwidth at the top of the 99 2000.”