Sep 25, 2024 · 1h 8m · news
Eric Vishria: Where is the Value in AI - Chips, Models or Apps? | E1206 · 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
Benchmark General Partner Eric Vishria joins Harry Stebbings to dissect his personal journey from startup founder to venture capitalist, highlighting his qualitative approach to early-stage investing. He provides a contrarian analysis of the AI value chain, explaining Benchmark's boutique partnership model, and why they prioritize infrastructure, founder quality, and high-trust teamwork over spreadsheet metrics.
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 16.3% of the talking time here. How this is scored →
Harry as informed peer 3.8 Guest teaching 4.1 Guest disagreement 1.9 Harry pushing back 2.6
speaking balance: gold is Harry, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 27:44 Rejecting the $600B AI Capex Premise
Eric explicitly rejects Harry's question regarding David Cahn's $600B AI capex thesis, flatly stating that he does not worry about the figure and denying that it is even the correct question to ask.
Hardest push from Harry ▶ 41:32 Challenging Partnership DynamicsHarry directly challenges Eric's framing of Peter Fenton's role, asking if a partner shouldn't act as a counterbalance rather than a Duracell battery for confirmation bias.
Biggest teaching moment ▶ 28:40 Search Engine Monetization ParallelEric educates Harry on tech history, explaining that early web search engines took over five years (1995-2001) to figure out monetization, reframing current AI revenue lag as a standard evolution during major platform shifts.
Harry holds his own ▶ 48:44 Harry's Stance on Early Gross MarginsHarry forcefully asserts his own venture experience, vehemently pushing back against spreadsheet-focused investors by explaining why gross margins at $3M ARR are completely meaningless.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | Harry as informed peer | Guest teaching | Guest disagreement | Harry pushing back | Why |
|---|---|---|---|---|---|---|
| Welcome Chat & Eric's CEO Reflection | 3 | 3 | 2 | 4 | Harry opens by probing Eric on his past CEO failures at Rockmelt and pushes back when Eric frames it as bad luck or timing, arguing that fundamental flaws in team or product must be to blame. Eric politely counters that startup success is non-deterministic rather than predictable. | |
| Evaluating a Founder's Distribution Strategy | 2 | 4 | 2 | 2 | Harry asks how to evaluate distribution strategies in founders and explores career investors versus operators. Eric clarifies that career investors are better at stock picking due to reps, but often make worse board members due to a lack of operational empathy. | |
| Sector Specialization vs. Benchmark's Core Model | 3 | 5 | 3 | 1 | Harry quotes Bruce Dunleavy regarding Eric's domain breadth. Eric reframes this by explaining why Benchmark intentionally rejects sector specialization, educating Harry on how rigid domain models get wiped out during platform shifts. | |
| Vertical SaaS & The Innovator's Dilemma in AI | 6 | 4 | 2 | 5 | Harry presents a detailed counter-thesis on vertical SaaS moats and deep data reserves. He also cites Mike Maples and Pat Grady when questioning if founder insights must be contrarian, prompting Eric to break down nuance across funding stages. | |
| Market Creation & The AI Medical Scribe Opportunity | 2 | 4 | 1 | 1 | Eric explains market creation using historical examples like Uber alongside modern AI medical scribes. Harry listens attentively as Eric outlines how to determine whether a brand new behavior will exist in three years. | |
| Filtering Through the Noise of AI Competitors | 4 | 3 | 1 | 2 | Harry brings up shared portfolio company 11x and references seeing 20 medical scribe companies, naming Nabla as a standout. Eric explains that crowded markets demand a significantly higher bar for founder belief and insight. | |
| Product Quality vs. Incumbent Distribution Moats | 5 | 5 | 2 | 5 | Harry pushes back on AI startups, arguing incumbent distribution moats like Microsoft Nuance will crush superior products through bundling. Eric agrees but breaks down the unit economics of developer tool spend versus total engineer cost. | |
| Evaluating Sugar High vs. Sustainable AI Revenue | 3 | 4 | 2 | 2 | Harry asks how to differentiate between sugar-high revenue scaling and sustainable retention. Eric shares an example of a team hitting $4M ARR in four months while cautioning that initial demand pull does not guarantee long-term advantage. | |
| The $600 Billion AI Capex Question | 4 | 5 | 6 | 3 | Harry raises David Cahn's $600B AI capex question. Eric flatly rejects the premise, stating he doesn't worry about it and schooling Harry with a historical parallel on how search engines took over five years to figure out monetization. | |
| AI Value Accrual: Infrastructure & Cerebras | 4 | 5 | 3 | 2 | Harry quotes Sarah Tavel regarding model commoditization. Eric corrects the quote to note foundation models are the fastest depreciating asset in history, before detailing Benchmark's positions in Cerebras and Fireworks. | |
| The Future of Foundation Model M&A | 5 | 4 | 3 | 5 | Harry challenges Eric on foundation model M&A, questioning antitrust barriers for $100B deals and asking if $50M AI checks break Benchmark's fund discipline. Eric counters that fund sizes are accounting artifacts and cites Charlie Munger. | |
| Benchmark's Concentrated AI Portfolio Strategy | 5 | 4 | 1 | 2 | Harry quotes Bill Gurley on playing the game on the field and asks if investors are overestimating AI in the short term. Eric contrasts Benchmark's extreme restraint in 2021 (3 deals) with their current high activity level. | |
| The Power of Partnership: Instincts & Memo Culture | 5 | 5 | 3 | 6 | Harry challenges Eric on partner dynamics, asking if a partner should be a counterbalance rather than a battery for enthusiasm. Eric counters by explaining why formal memo-writing cultures encourage irrelevant third-order analysis. | |
| Behind the Deal: The Cerebras Investment Story | 1 | 4 | 0 | 0 | Eric shares the narrative of the 2016 Cerebras deal, detailing how Peter Fenton went from trying to talk him out of the investment on Sunday to telling him to call for a vote on Monday afternoon. | |
| Partner Saves: Sarah Tavel & Gross Margins | 7 | 3 | 1 | 3 | After Eric tells a story about Sarah Tavel saving him on gross margins, Harry forcefully steps in to agree, passionately ranting about how early-stage gross margins at $3M ARR are completely irrelevant. | |
| Portfolio Support & Board Management at Benchmark | 5 | 4 | 3 | 6 | Harry questions Eric's capacity given his 12-13 board seats, calling it nuts, and asks if Benchmark's internal voting mechanism conflicts with non-consensus investing. Eric defends his time split and clarifies internal voting dynamics. | |
| Venture Legends Quick-Fire: Gurley, Fenton & Kohler | 3 | 4 | 1 | 1 | Harry asks quick-fire questions on lessons learned from Bill Gurley, Peter Fenton, and Matt Kohler. Eric provides sharp characterizations comparing Gurley's market-first focus to Fenton's people-first superpower. | |
| General Quick-Fire & Respect for Jim Goetz | 3 | 4 | 2 | 2 | Harry asks for contrarian beliefs and respected peers. Eric states his strong belief that Nvidia will not be the sole compute winner, and expresses deep gratitude for Jim Goetz. | |
| Unmade Decisions & The Ancient Chinese Horse Story | 4 | 5 | 1 | 1 | Harry asks a philosophical question about unmade decisions that weigh heavily. Eric reflects on Rockmelt's exit timing before sharing the ancient Chinese parable of the horse and the broken leg. | |
| Parenting Twins & The Fatigue Inversion | 2 | 3 | 0 | 0 | Harry asks about advice for parenting twins and favorite Benchmark memories. Eric describes how raising twins inverts weekend fatigue before recalling Jay Kreps calling him for Confluent's Series A. |
Statements from this episode (28)
Vishria: AI foundational models are fastest depreciating asset in history
“Foundational models are the fastest depreciating asset in human history.”
Vishria: Nvidia will not remain the only major AI infrastructure player
“I don't believe NVIDIA is going to be the only game to have content infrastructure layer.”
Vishria: AI platform shift could outgrow all previous tech shifts combined
“We have a major, major shift in AI, which could be bigger than any of these other shifts on maybe combined.”
Vishria: Benchmark is investing at its highest pace since 2011
“There's a lot of uncertainty, but we've been more active than we've been since 2010 and 2011.”
Vishria: I should have broken comp rules to hire key talent
“There were a couple times or two people and specifically that I think about where like what they wanted in comp and what I was like willing to give were separate. Like they were just off and they were so far out of market in terms of what their ask was. And I …”
Vishria: Career investors are better at investing than former founders
“They're better investors. Also not necessarily better board members, not necessarily better advisors, not necessarily a better bunch of things, but definitely better.”
Vishria: Career VCs often make worse board members due to lacking operational empathy
“For, well, going back to, they haven't actually done these things at depth. They haven't done them themselves. They often lack like empathy and understanding. I think they often think that things are more deterministic than I think they actually are. And so li…”
Vishria: Benchmark has no sector specialists and maintains four to six partners
“I am not a sector specialist and nobody at Benchmark is. And I think the fundamental idea with Benchmark is there's a small group of people, a small group of partners who are all equal. And right now it's five of us, but sometimes it's four, sometimes it's six…”
Vishria: Incumbents must burn down existing businesses to execute AI platform shifts
“And in order to make the platform shift for a company, like what you're articulating, You have to kind of be willing to burn down the existing business and at least in short order. And there's a whole bunch of entrepreneurs who don't do that.”
Vishria: Startups can succeed purely through aggressive execution without core insights
“There's a couple of cases where just like violent execution by a determined team in a hyper competitive market where like nobody has an insight has worked.”
Benchmark has met several AI medical scribe startups but passed on all
“I've met several of these companies over the years and we haven't invested in any.”
Vishria: Passes on viable startups if he cannot pitch recruits
“When you look at that, I ask myself that question. And if I answer is like, you know what, I think it's a cool business. It'll probably work, but I just don't understand how I can make that pitch. Then I don't do it. Like, that's my, like, I just, I don't do i…”
Vishria: Tech incumbents are too paranoid and alert to be easily disrupted in AI
“Today's incumbents are and you know, this is a trope and not really an insight. Everyone's saying the same thing, but today's incumbents are paranoid and on top of things and like, you know, like is, is Microsoft and Nuance gonna move slower and whatever than …”
Vishria: Replacing Developer Labor With AI Creates 20x More Economic Value
“So if the AI, if you can get AI good enough to eat much more of the 200,000 dollars of value attributed to your software engineer, IT person, it just follows that there's like 20 x more value creation and, like, there will be giant, giant outcomes.”
Vishria: No AI Company Has Yet Demonstrated Software Engineer Replacement
“We're a little ways away from like, from replacing the software engineer right now. We're like a bit away from that. And like, we haven't actually captured that much market value yet. And really no one's demonstrated anything close to that level. Quite yet.”
Vishria: Rapid zero-to-$4M ARR scaling in AI holds almost no value
“I was talking to a team that had gone like this last week, I was talking to a team that had gone like zero to four million, a four person team, zero to four million in four months. Amazing. Like just amazing kind of revenue trajectory. But I put like almost no…”
Benchmark AI startups achieve years of SaaS scale in under a year
“The quickness of the scale is, is, is unlike, it's like three, four, five years of what SAS company, like your traditional SAS companies were doing. We're seeing in under a year and we have a whole portfolio of companies that are like this.”
Vishria: AI monetization won't rely on $20 subscriptions or API calls
“And, like, it's not gonna be a 20 dollar a month subscription, that's not, like, the right way and it's not gonna be just, like, bundled API call, like, it's, there's gonna be much, much more sophistication and interesting models in that.”
Vishria: Benchmark holds no investments in foundational AI model companies
“In terms of value accrual, like, you know, I think for Benchmark, like, we have a, we have no foundational model investments thus far anyway.”
Vishria: Benchmark has written $50M checks despite boutique fund size
“Through, through 30 years of performance, I think we have Unprecedented flexibility in what we do, and so if we want to write a fifty million dollar check, we can write a fifty million dollar check, and we have in cases.”
Vishria: Benchmark never considers portfolio construction or fund timing
“You know, I think one of the things that we maybe think about almost not at all Is almost never think about like fun cycle or fun timing or anything else. And we almost never think about or talk about portfolio construction or anything else.”
Benchmark made only three new investments during 2021 SaaS boom
“I think in 20, 21, we made like three new investments as a firm. Three.”
Vishria: VC memo culture focuses on irrelevant details over key questions
“I think this is why the memo writing culture at a lot of firms gets you in trouble because you put a lot of information You basically are, it encourages putting a lot of information that is like third and fourth order stuff Into document as if that is impactin…”
Vishria: Sponsoring partner at Benchmark decides whether to call deal votes
“We have a system where, like, we talk, and then, like, you can call, like, the sponsoring partner basically can call for a vote or not.”
Vishria: Early unit economics never extrapolate to $200M revenue scale
“I've worked on five companies that have gone from zero to more than 200 and revenue, and like, in not a single case did the economics at the very early stage, like, extrapolate all the way. Like, it's just not a thing. Not, not even the economics from when the…”
Eric Vishria details Benchmark's unique 1-to-10 investment voting system
“Our voting system is you vote one to 10. You can't vote five. Six and above is yes. Four and below is no.”