Jun 21, 2024 · 1h 1m · more-or-less
#52: Elad Gil on What to Build in AI (Part II) · More or Less Podcast
Elad Gil · 21m spoken Sam Lessin · 16m spoken Jessica Lessin · 8m spoken Dave Morin · 6m spoken Brit Morin · 3m spoken
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
Investor and operator Elad Gil joins the More or Less podcast to debate the economic viability of the artificial intelligence boom, evaluating where enterprise value accrues across models, applications, and corporate buyouts. The co-hosts critically examine hyperscaler spending, startup defensibility, founder dynamics, and the broader societal consequences of algorithmic platforms.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. The hosts hold 61.1% of the talking time here. How this is scored →
The hosts as informed peer 6.4 Guest teaching 3.9 Guest disagreement 2.0 The hosts pushing back 3.9
speaking balance: gold is the hosts, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 45:20 SQL wrapper rejoinder and valuation clash
Elad abruptly shuts down Sam's skepticism about AI wrappers by retorting that all of SaaS was merely a wrapper on SQL databases, directly rejecting Sam's premise.
Hardest push from the hosts ▶ 38:00 Dave's refusal of weak data claims on teen mental healthDave adamantly rejects Elad's assertion that data on social media harms is weak, declaring the science solid and framing the issue as an urgent national emergency.
Biggest teaching moment ▶ 17:05 Payroll displacement and SaaS recreation thesisElad breaks down proprietary research demonstrating that capturing 10% of white-collar services payroll would recreate the entire $500B annual SaaS industry.
The host holds their own ▶ 45:24 Sam's five-times cash flow reality checkSam swiftly counters Elad's defense of SaaS by pointing out that SaaS multiples have cratered to 5x cash flow because investors mistakenly modeled infinite customer lifetime value.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | The hosts as informed peer | Guest teaching | Guest disagreement | The hosts pushing back | Why |
|---|---|---|---|---|---|---|
| More or Less Podcast Opening Title Sequence and Theme | 6 | 5 | 1 | 2 | Dave references Elad's classic end-of-cycle blog post and questions how market diversification signals apply to the current AI landscape. Elad clarifies the conceptual distinction between diversification across waves versus search functions within an active wave. | |
| Hyperscaler Capital Spending and the Reality of Enterprise Adoption | 6 | 6 | 3 | 7 | Sam aggressively challenges the massive capital expenditures in AI, calling them billions in 'fuck around dollars' driven by hyperscalers whose stock prices rise simply from buying H100s. Elad counters with hard numbers from Azure's earnings, showing genuine multi-billion-dollar cloud run rates even before full enterprise deployment. | |
| Antitrust Constraints on M&A and Incumbent Distribution Power | 6 | 4 | 2 | 5 | Brit and Sam probe M&A viability for application-layer AI, contrasting incumbent distribution advantages against regulatory friction. Elad points out that antitrust scrutiny makes public company takeovers difficult, driving private-to-private transactions like Databricks. | |
| Value Capture Across Waves and Displacing White-Collar Services | 7 | 7 | 4 | 6 | Sam argues historical tech waves primarily concentrate value into existing tech incumbents rather than creating new giants. Elad presents his team's analysis of a $3.5 to $5 trillion white-collar services payroll, demonstrating that converting even 10% into software would equal the entire existing SaaS market cap. | |
| Frontiers Beyond LLMs: Robotics, Biology, and AI Buyouts | 4 | 5 | 1 | 1 | Jessica asks about investment frontiers beyond LLMs. Elad outlines the potential of AI-driven buyouts, robotics breakthroughs, biological models, and synthetic avatars. | |
| Founder Runway, Startup Failure Norms, and Returning Capital | 7 | 5 | 2 | 5 | Elad argues founders should shut down struggling startups faster to conserve their most productive years, challenging the valley orthodoxy of endless grinding. Dave and Sam push back to analyze the underlying VC mechanics, such as free-option incentives, LP cash return demands, and fund recycling limits. | |
| Uncovering Investment Alpha, Backing Anduril, and Career Missteps | 4 | 4 | 1 | 3 | Brit and Sam ask how Elad finds alpha and demand an accounting of what he has gotten wrong in his career. Elad candidly discusses his early GAN investigations and backing Anduril, while admitting major missteps in studying biology and entering telecom at the dot-com crash. | |
| Reimagining Social Networks with AI Versus Algorithmic Addiction | 8 | 2 | 1 | 5 | Dave and Sam leverage their operational backgrounds in consumer social platforms to debate whether AI can build genuine social networks instead of hyper-optimized engagement feeds. Sam argues that algorithmic feeds like TikTok represent refined sugar that drives margin to zero, while Dave defends the potential for local-model innovation. | |
| The Teen Mental Health Debate and School Phone Bans | 7 | 4 | 5 | 8 | When the hosts raise the surgeon general advisory and school phone bans, Elad takes a contrarian stance by asserting that empirical data linking social media to teen harm is weak. Brit immediately challenges his premise, and Dave forcefully pushes back, stating the clinical evidence of an emergency is incontrovertible. | |
| Synthetic AI Companions, Societal Distortion, and Market Saturation | 6 | 2 | 1 | 2 | Sam forecasts deep societal distortion from upcoming synthetic AI companions and digital relationships, drawing parallels to shifts already seen with dating apps. The group engages in lighthearted banter regarding AI spouses and market oversaturation in tequila brands. | |
| Application Layer Moats, Workflow Ownership, and Bundling Risks | 8 | 6 | 6 | 8 | Sam dismisses basic generative AI workflow tools as easily replicated wrappers. Elad delivers a sharp retort that SaaS was just a wrapper on SQL databases, prompting Sam to fire back that SaaS multiples have collapsed to 5x cash flow while questioning AI startup stickiness against Salesforce bundling. | |
| The Structural Evolution of Venture Capital and Life Priorities | 6 | 5 | 1 | 2 | Jessica asks whether the traditional venture model is broken, referencing her historical scoop on DST's landmark investment. Elad gives a measured historical review of venture innovations and reflects on how his personal worries have evolved from career metrics to broad societal health. | |
| Cohost Debrief: Bullish Consensus, AI Winter, and Infrastructure Overbuild | 7 | 0 | 0 | 0 | In a post-interview debrief without the guest, the co-hosts critique Elad's bullishness, noting that an obsession with infrastructure over useful application is reminiscent of the crypto bubble and signals an impending AI winter. | |
| The AI Buyout Thesis, Metropolis Parking, and Flight Wrap | 7 | 0 | 0 | 0 | Concluding their debrief without the guest, Sam details his firm's investment thesis in Metropolis, where buying real-world parking assets and applying tech proved more effective than trying to sell software to reluctant incumbents. |
Statements from this episode (41)
Gil: 99% of AI Companies Will Fail, But 1% Will Be Huge
“And I think it's clear that 90, whatever, 99% of AI companies are not going to work out, but the one percent that will could be quite large.”
Gil: Big Wave of Enterprise AI Is Two to Three Years Away
“And I think from that is this under overhyped perspective, one could argue this is very under hyped in part because there's been so little true enterprise adoption. But despite that, you're seeing these massive waves of revenue and usage and everything else, a…”
Gil: Startup diversification differs across technology waves versus within one
“I think there's two types of diversification. There's diversification within a technology wave, and there's diversification across waves. And what I was referring to for that end of cycle post was diversification across waves... I think that's a little bit dif…”
Lessin: The AI Wave Will Sink $1 Trillion to Create $300 Billion
“The cost is going to be a trillion dollars of capital sunk to create that three hundred billion dollars.”
Gil: Microsoft Azure AI Services Lift Adds $1 Billion Quarterly
“Azure in their last quarter announced a twenty five billion dollar quarter of which a five percent lift came from AI. So they added something like a billion and a half per quarter in new revenue for what they considered AI services on their cloud.”
Lessin: An AI market crash is almost inevitable
“There's almost an inevitable crash, even if you do believe that eventually value gets created.”
Gil: Foundation model funding is mostly strategic hyperscalers, not VC
“Most of the money funding foundation models is strategic, right? Versus venture capital. Venture capital maybe funds things into the hundreds of millions, and then it gets handed off to Google and Amazon and Facebook and all the other folks to invest billions …”
Gil: Tech M&A will remain constrained by FTC and global regulators
“I think the M&A environment is going to be constrained due to our current FTC slash the more global regulatory viewpoint on whether they want M&A to happen or big tech to get bigger.”
Gil: Databricks Is Smart to Aggressively Acquire Startups Before Going Public
“And so it's actually harder to buy things as a public company as it is from a private company, which is why I think what Databricks is doing or some others in terms of buying things while you're private is really smart because it's going to get harder once you…”
Gil: Incumbents Can Launch AI Four Years Late and Still Win
“I also think honestly as an incumbent you have a couple years to figure it out because you tend to have so much distribution that you can launch four years later and still You know, when, as it were”
Gil: Startups captured internet value, but incumbents captured mobile value
“And in the internet wave, you could argue 80 or 90% of the value went to startups. And in mobile, it was the other way around. Almost all the value went to incumbents, and in part it was because it was an extension of access and UI.”
Lessin: Native mobile startups capped out at around $100B, led by Uber
“Then you have the mobile era companies, which are really the max is like a hundred billion dollars, right? Like Uber is the biggest, right? Of like that native wave there.”
Gil: Automating 10% of Services Payroll Recreates the Entire SaaS Market
“And if you just look at the headcount resident in those services, legal and accounting and certain types of sales and nursing, and you just kind of go through it all. And you add up just the payroll. For those areas, it's about three and a half to five trillio…”
Gil Predicts AI Cycle Will Take 10 to 15 Years to Fully Substantiate
“1015 years to fully substantiate, and the same thing is going to be true here.”
Gil: Easy capital leads founders to keep failing startups alive too long
“I think if things aren't working, I think because money has been so available, people have kept their companies going longer than they should. And in a lot of circumstances more recently, where sometimes the best thing you can do is just kind of quit and resta…”
Gil: VCs Push Struggling Founders to Grind Because They Are Free Options
“Investors have a strong incentive for you to keep going because you're just an option once they put money in. And so if they get 50 cents on the dollar, they don't care, and they'd rather you just keep trying because you're just a free option. They already hav…”
Gil: VCs avoid hard shutdown conversations to protect founder referrals
“Investors often don't want to have tough conversations because they want founders to refer their friends to them and they don't want to be the kind of tough person or whatever.”
Lessin: Seed Stage VCs Are Paid Unbelievably Well to Do Almost Nothing
“My view is seed investors get paid unbelievably well to do almost nothing. And like the trade is we get paid incredibly well when it works, but we lose money constantly. And like, that's completely fine.”
Gil: Nobody wanted to start an OpenAI competitor after GPT-3
“And I remember when GPT-III came out, I tried to both pitch a bunch of people to start effectively an open AI competitor, and nobody wanted to do it, because they didn't think it was a good business model, as well as it was such a big technology step up if you…”
Gil: Google canceling Maven prompted defense tech pivot and backing Anduril
“I started getting into defense tech when Google shut down Maven because I was like, oh my gosh, if Google isn't doing it, what a great startup opportunity, you know? So I started looking for defense companies and then I met Anduril and I was like, oh my God, t…”
Gil: Biotech Is Brutal; Regrets Spending Years on a Biology PhD
“I got a PhD in biology, and I think biotech is a really tough place to work, right? As is academic biology. So, you know, arguably I got that wrong, and if I could get those, you know, four or five years back, I would kind of thing.”
Elad Gil: Social networks became social media, leaving room for new networks
“All the things that used to be social networks turned into social media. You know, and so how we thought about networks versus broadcast shifted pretty dramatically, and there's probably still room for interesting networks.”
Morin: Overlaying AI primitives into social use cases is currently too expensive
“I just think it's expensive right now to do social use cases. I think that, you know, in order to overlay any of these AI primitives into social use cases, like it's very expensive.”
Morin: Apple's AI release lacks native access to local model APIs
“I was hoping actually that the Apple release would include Like native access to local APIs, but it doesn't which is unfortunate as I was like ready to write checks for social AI stuff”
Sam Lessin: TikTok shows engagement is like infinitely refinable sugar
“TikTok has shown us what the true north is of engagement. I think that is like refined sugar and you can just keep refining it.”
Gil: Data Linking Social Media to Teen Mental Health Issues Is Weak
“I think I tend to look at data for these sorts of things. And I think the data on a lot of these topics is pretty weak.”
Dave Morin: Social media causes irreparable harm to an entire generation
“I think the data is abundantly clear. The science is, like, pretty hard. It's like really good science, and the science says that I think we're doing irreparable harm to an entire generation, and that it's a national mental health emergency, and that we have t…”
Dave Morin: School phone bans will quickly become a national movement
“I think we should expect this to become a national movement very quickly because I think parents are panicking.”
Morin: Steve Jobs Warned in 2008 That Facebook Would Become Like Cigarettes
“Dave and I both worked at Apple back in the day, and I remember the town hall that Steve spoke at in like, oh, seven, oh, eight, where he specifically pointed out that Facebook was going to be seeing the equivalent of cigarettes, and like how prescient was tha…”
Lessin: Personalized Fantasy AI Companions for Children Will 1,000% Happen
“Wait until you have a fantiful character that knows you incredibly well, and everything about you, and is like your kid's best friend, but is like a fantasy made up character, which by the way will 1000% happen.”
Lessin: Marriage Is Declining Because Tinder Made Sex Too Accessible
“Right now we have a problem where, like, marriage is on the decline because Tinder has made sex too accessible, right?”
Lessin: AI Girlfriends Will Replace Dating Apps and Destroy Human Relationships
“Like, wait until, like, a generation of men, it's not porn and Tinder, it's just, like, your perfect AI girlfriend companion, and, like, you destroy human relationships.”
Elad Gil: Current generative AI is primarily good at automating repetitive email jobs
“If you ask, what is this wave of generative AI good at least on the LLM side... I think it's a lot of things which are highly repetitive email jobs. Right. That's what this wave and right now is really good at.”
Gil: SaaS Was Just a SQL Wrapper and Still Generated Massive Value
“People say that these things are just wrappers on GPT in some cases, and you're like, well, SaaS is just a wrapper in a SQL database, and that worked out just fine for lots of people.”
Elad Gil: AI application defensibility requires owning the workflow rather than the model
“For the application layer, can you eventually take over a workflow? What is the workflow that you're enabling? Because really what you want is if you swap in and out underlying models, you're still sticky somehow, and a model improving doesn't commoditize you,…”
Lessin: Investors Erred by Assuming Infinite Customer LTV in SaaS Businesses
“The big mistake everyone made in SAS as investors is they calculated infinite LTV. They're like, once we have a customer, they are never going to turn. Right. And therefore like we can pay a lot for them and we'll have them forever.”
Dave Morin: Tech Has Extracted Most Value from Current LLM Cycle
“I mean, I think we've extracted a lot of the value out of this LLM cycle, and there's a lot of people, like, trying to find what the next, you know.”
Lessin: Meta Will Realize Long-Term Value from Current AI Models
“Well, to be clear, I think we're going to see the value of it showing up in, like, meta earnings for a long time to come, which is incrementally what already exists, and, like, translation. There's a bunch of stuff that's, like, super valuable.”
Sam Lessin: Obsession with Infrastructure Over Applications Is a Negative Sign
“I always get very, very worried when everyone's obsessed with the infrastructure. Right? Like that, when that becomes an error. That happened in crypto too, right? This whole infrastructure, infrastructure, infrastructure, not what it's useful for. No one has …”