Aug 5, 2024 · 1h 13m · news
David Cahn: Why Servers, Steel and Power Are the Pillars Powering the Future of AI | E1186 · 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 the 20VC podcast, Sequoia Capital partner David Cahn discusses the physical, industrial, and financial realities of the AI boom—summarized as 'servers, steel, and power'—while sharing deep insights into venture capital strategy, deal-making tactics, and the psychological roots of his relentless work ethic.
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 25% of the talking time here. How this is scored →
Harry as informed peer 3.6 Guest teaching 3.8 Guest disagreement 0.8 Harry pushing back 2.3
speaking balance: gold is Harry, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 29:18 Rejection of Alex Wang's data bottleneck premise
The guest directly rejects Scale AI CEO Alex Wang's core thesis that data is the primary bottleneck, arguing instead that compute, models, and data have converged and proposing his own 'servers, steel, and power' framework.
Hardest push from Harry ▶ 20:24 Host refusal of software pricing power thesisThe host directly disagrees with the guest's pricing power model by citing Canva and Adobe, arguing that commoditization of AI features forces a race to the bottom regardless of value added.
Biggest teaching moment ▶ 12:30 Physical reality of data centers vs compute euphemismsThe guest reframes the host's abstract discussion of compute by educating him on the physical real estate limits, H100 vs B100 chip replacement challenges, and liquid cooling overhauls required in data centers.
Harry holds his own ▶ 39:25 Host highlighting VC logical inconsistency on AGIThe host demonstrates sharp domain analysis by citing Ethan Mollick to point out that VCs who claim to believe in imminent AGI while continuing to invest in standard SaaS accounting products are holding logically incompatible views.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | Harry as informed peer | Guest teaching | Guest disagreement | Harry pushing back | Why |
|---|---|---|---|---|---|---|
| 20VC Podcast Title Montage | 0 | 1 | 0 | 0 | The host opens with warm intro questions about the guest's childhood and twin background. The guest collaboratively explains how being a twin shaped his nonconformity and competitive nature. | |
| Family Heritage and Personal Drive | 1 | 2 | 0 | 0 | The host shares a personal story about his mother's illness to prompt the guest on personal adversity, then bridges to AI capex. The guest sets context on his six-year history investing in AI before framing the core capex payback problem. | |
| Oligopoly Game Theory and Speculative Overspending | 3 | 4 | 1 | 2 | The host brings up an insider perspective comparing AI spend to the Manhattan Project. The guest reframes recent big tech statements as proof of rational oligopoly defense rather than guaranteed revenue. | |
| Startup Benefits from Overproduced Compute | 4 | 3 | 1 | 3 | The host presents alternative viewpoints, asking whether overspending benefits startups or dangerously concentrates power. The guest agrees both dynamics exist and breaks down producer versus consumer compute economics. | |
| The Physical Reality of Compute and Obsolescence | 5 | 5 | 1 | 4 | The host contrasts the guest's thesis with Sam Altman's quote that compute is the currency of the future. The guest educates the host on the physical limits, GPU cycles, and liquid cooling trade-offs of data center construction. | |
| Model Efficiency and the Data Center Asset | 4 | 5 | 1 | 2 | The host asks if increasing model efficiency reduces compute replenishment needs. The guest counter-intuitively reframes that greater efficiency accelerates the obsolescence of data center architecture. | |
| Data Center Construction and Labor Supply Chain | 2 | 5 | 0 | 2 | The host asks how data center physical real estate evolves. The guest details the entire supply chain, mentioning developers, contractors, and Facebook ads for local electricians. | |
| AI Integration and Software Pricing Power | 6 | 4 | 2 | 6 | The host challenges the guest directly by citing Canva's AI integration, arguing that feature parity destroys pricing power and causes a race to the bottom. The guest responds with nuance regarding structural data moats. | |
| Vertical Integration of the Model and Infrastructure | 4 | 4 | 1 | 3 | The host asks about vertical integration and whether non-Mag 7 players can participate in the capex race. The guest explains why competing in frontier models requires non-AI cash cows like Instagram or AWS. | |
| Antitrust and the Future of AI Acquisitions | 4 | 4 | 1 | 3 | The host asks if acquisition is the only path forward for startups like OpenAI and Anthropic. The guest points out that antitrust pressure forces partial ownership structures rather than full takeovers. | |
| Wall Street Versus Silicon Valley Perspectives | 5 | 5 | 2 | 3 | The host brings up Scale AI CEO Alex Wang's thesis that data is the primary bottleneck. The guest gently rejects Wang's framing, offering his own three pillars: servers, steel, and power. | |
| Chip Wars, Nvidia's Dominance, and Geopolitics | 4 | 4 | 1 | 2 | The host asks about geopolitical risks and Trump trade policies regarding Taiwan chips. The guest outlines industrial supply chain dynamics, including supplier reluctance to build new factories without long-term guarantees. | |
| Investing in the physical AI Supply Chain | 4 | 4 | 1 | 3 | The host questions why investors spend time on software like AI CRMs instead of physical supply chain developers. The guest details his own focus on the physical supply chain, long-duration batteries, and utility leaders. | |
| Open-Source Models, AGI Panic, and Societal Impact | 6 | 3 | 1 | 3 | The host cites expert perspectives from Vinod Khosla, Marc Andreessen, Alex Wang, and Ethan Mollick to probe AI existential risk and VC logical inconsistencies. The guest agrees with Mollick's assessment while remaining pragmatic. | |
| China's Capabilities and US Competitive Advantages | 4 | 5 | 1 | 3 | The host asks about China's capabilities and probes the financial mechanics of rising capex. The guest demonstrates how simple math leads to a trillion-dollar capex projection via off-balance sheet lease structures. | |
| Data Center Debt Financing and Yield Spreads | 3 | 4 | 0 | 2 | The host asks about debt yield rates and market dynamic adjustments. The guest breaks down how real estate developers view data center leases as high-yield tech debt and references Asimov's psychohistory. | |
| Lessons in Deal Selection and Customer Value | 5 | 5 | 1 | 2 | The host transitions to guest's early Coatue career and lists major deals like UiPath, Databricks, and Snowflake. The guest explains deal selection lessons, noting customer calls often complain while customer behavior retains. | |
| Mastering Superpowers in Venture Capital | 3 | 4 | 1 | 2 | The host asks about zero-to-one learnings and managing distinct venture styles. The guest defines the concept of 'sluggers' in VC who generate billion-dollar returns. | |
| Constraints, Sequoia's Mold, and Long-Term Excellence | 4 | 4 | 1 | 3 | The host asks how Sequoia forces partners to put their neck on the line. The guest discusses investment constraints, PayPal history, and how Sequoia breaks down and rebuilds investors. | |
| Quickfire: Sequoia's Best Sorcerers and Pickers | 4 | 5 | 0 | 2 | The host prompts quick answers on Sequoia's top sorcerers and pickers. The guest shares a detailed 2x2 matrix combining science and intuition across human and tech dimensions to assess visionary founders. | |
| Is Venture Capital a Young Person's Game? | 4 | 3 | 1 | 3 | The host asks if venture is a young person's game and forces rankings of founder/market/product. The guest explains how young VCs pitch growing with founders over 20 years. | |
| Creative Tactics to Get a Founder's Attention | 3 | 2 | 0 | 1 | The host asks for crazy founder outreach stories. The guest reveals his tactics of sending personalized Cameo videos from actors and daily 'emo looms'. | |
| A Silicon Valley Morning Routine | 2 | 2 | 0 | 0 | The host asks about the guest's 7am office arrival and morning routine. The guest describes biking through Stanford up Sand Hill Road and recalls his first major deal in Starburst Data. | |
| Quick-Fire Round and Final Reflections | 3 | 3 | 1 | 1 | The host conducts a rapid-fire round covering belief in God, equity allocation, Mike Volpe, and driving lessons. The guest recommends founders give up to 5% equity to transformational hires. |
Statements from this episode (61)
Cahn: No AI Lab Will Train Frontier Models in the Same Data Center Twice
“No one's ever going to train a frontier model on the same data center twice, because by the time you've trained it, the GPUs will be outdated, and the data center will be too small.”
Cahn: Data Centers Are the Most Important Asset in AI
“The bigger these models get, the more that scaling laws become the dominant thing, and so I think there's a really good argument to be made that basically the data center is the most important asset.”
Stebbings Used 20VC Earnings to Pay for Mother's MS Treatment
“Well, I, my mother got MS, and I, we didn't make any money from the show, and I said, I'll pay for the show, and we used the podcast to pay for all of her treatment, hundreds of thousands of dollars.”
Cahn: The global SaaS market is currently $250 billion
“The entire SaaS economy is a two hundred and fifty billion dollar market, right?”
Cahn: AI Capital Expenditures Over Next 24 Months May Be Hard to Repay
“Believing in AI and believing that infinite capex in the next two years, those are very different concepts. You can believe in AI, but also believe that the amount of capex in the next 24 months may be difficult to pay back.”
Cahn: Big Tech AI Spending Is Speculative, Not a Guarantee of Revenue
“The wrong takeaway is, oh, Google knows something we don't know, and therefore is making this investment, and therefore hundreds of billions of dollars of AI revenue are guaranteed in the next two years. That's the extrapolation I think people were making that…”
Cahn: Big Tech's Overproduction of Compute Directly Boosts Startup Gross Margins
“These big tech companies are producers of compute. Startups are consumers of compute. So if you believe there is an overproduction of compute and that compute prices go down, startups win because they are buying those compute. Lower cost of compute directly tr…”
Cahn: Massive AI cloud capital requirements create significant barriers to entry
“Now, in order to be an AI cloud, you need to be willing to light a bunch of money on fire, right? And so I think that that is a pretty significant barrier to entry for new entrants.”
Cahn: The $250B Cloud Infrastructure Market Equals the Entire SaaS Sector
“This cloud business today is a two hundred and fifty billion dollar business. So the cloud business today is the same size as the SaaS sector. That's the business that Azure, Google and AWS control is the same size as the entire SaaS sector.”
Cahn: AI Data Centers Take Two Years and $2 Billion to Build
“Data centers take about two years to build. You are going to build, and it costs about two billion dollars per data center.”
Cahn: Researcher Mobility Has Eliminated Differentiation Between Top AI Models
“All these researchers, they jump from one lab to the other. It's not like there's that much differentiation anymore between the actual models that these companies are launching.”
Cahn: Amazon Announced $50 Billion in New Data Centers
“Amazon in the last six months has announced fifty billion dollars of new data centers, right?”
Cahn: DPR is the Largest General Contractor for Data Centers
“DPR is the biggest general contractor in building data centers.”
Cahn: Labor is the Single Largest Cost in Data Centers
“I mean, labor is the single biggest cost in data centers.”
David Cahn: Current AI evidence favors scaling laws over algorithmic breakthroughs
“The evidence today, I think, is more strongly in favor of scaling laws, and I hope to see more evidence of the, of these other breakthroughs being able to drive progress”
Cahn: Injecting AI for Marketing Adds Costs Without Pricing Power
“If you are simply injecting AI so that you can tell investors that you have AI and so that you can, you know, talk on your marketing page about how you have AI, then you're just injecting a new cost item with actually no benefit to the consumer. And if there's…”
Cahn: Current AI Products Are Not Yet Showing Strong Value Creation
“The burden to deliver value is very, very high in business. And we're not seeing that so strongly in AI yet.”
Cahn: Low-barrier AI sectors will see margins bid down to zero
“There will be some sectors where you're right, things get commoditized, and they just basically get bid down to zero percent gross margin.”
Cahn: AI Features Will Not Create Pricing Power for Weak Software Businesses
“If you had pricing power before AI, you'll have pricing power after AI. If you didn't have pricing power before AI, you probably won't have pricing power after AI.”
David Cahn: Nvidia CEO Jensen Huang is the Steve Jobs of our era
“Jensen is the Steve Jobs of our era.”
Cahn: Separate Data Center and AI Model Teams Will Fail as Models Scale
“I think the vertical integration piece is gonna matter in the sense of you can't have a separate team running the data center and a separate team building the model. That is just not gonna work as these models get bigger and bigger. You need to deeply couple t…”
Cahn: Competing in Frontier AI Requires a Massive Non-AI Cash Machine
“I think the evidence of the last couple of years is in order to play in the big model game, you need a cash machine and that cash machine cannot be the AI business, right?”
Cahn: Cloud Providers Are Trapped in an AI Capex Prisoner's Dilemma
“The cloud guys are stuck in more of a prisoner's dilemma. They have to keep investing. If they do not invest, they risk losing market share in one of the greatest businesses of all time.”
Cahn: A consumer killer use case for AI has not emerged yet
“I think the consumer killer use case hasn't emerged yet.”
Cahn: Tech Giants Will Avoid Full Acquisitions of Major AI Labs
“I think that will probably sustain because it's not really possible to acquire these businesses. Now you can acquire the smaller ones, and we've seen that with Inflection and Adept, you can acquire some of the teams, but it's going to be much more difficult to…”
Cahn: Rational AI strategy is unclear for non-cloud giants like Salesforce
“I mean, I do think the like tier below those three big companies are scared right now. I think there's probably like those boardroom conversations are probably tense, right? I think on the one hand, huge opportunity, right? Make the product better. You already…”
Cahn: No Major AI Model Company Holds a Data or Compute Advantage
“I think it's really hard to argue today that any of the big model companies has a data advantage. Compute, it's just a commodity that you pay for. So it's hard to argue that any of them has a compute advantage.”
Cahn: AI's true bottlenecks are servers, steel, and power
“And so I'll propose my own three things that I think are the three things that matter. And I would summarize it as servers, steel, and power.”
Cahn: Nvidia will continue to surprise positively on chip performance
“The price to performance ratio continues to improve. I don't see that stopping. NVIDIA is a fantastic company. They've, Have a track record of innovating. They have a track record of new products. You better believe Jensen's driving his team super hard on what…”
Cahn: Nvidia's fat gross margins will inevitably draw heavy competition
“NVIDIA earns a big gross margin. And when you, when anyone sees a fat gross margin in a big market, people come chasing.”
Cahn: Big tech's supply chain influence is an underestimated AI advantage
“The big tech companies actually do a very good job of this. And I think it's another advantage they have that's underestimated is kind of convincing the supply chain to change their behavior is something that those guys have a lot of practice doing.”
Cahn: CyrusOne and QTS will produce massive returns for KKR and Blackstone
“KKR owns Cyrus one, Blackstone owns QTS, fantastic investments. There's gonna be huge money-making investments for those firms.”
Cahn: Coatue-Backed Battery Startup Built a $1B Factory in 12 Months
“I visited this company that's building long duration batteries that I had invested in at KOTU. They literally built a billion dollar factory in the last 12 months.”
Cahn: Factory Construction Rates Will Reach All-Time Highs Within 12 Months
“I think in 12 months you'll see all these charts coming out of, like, more factories are getting built than ever before, and all this stuff. The industrial revolution is just getting started.”
Cahn: US power supply is insufficient for AI compute demands
“We don't, and I, you know, there's sort of an interesting phenomenon here, which is, you know, the Biden administration passed the IRA, the Inflation Reduction Act, and this was this huge stimulus bill that basically incentivized everyone to build solar and bu…”
Cahn: AI Demand Will Drive Green Energy Faster Than Government Regulation
“And I think one of the great ironies, and this is me as a capitalist speaking, right, is that the forces of capitalism, AI, will drive more energy revolution than any amount of political regulation could have, right?”
Cahn: AI will cause gradual energy grid expansion, not an acute crisis
“And so I don't know that it's going to be this like loud crisis that people anticipate. I think it's going to be a slow track for more renewable energy, more power on the grid. And we're just going to, we're going to plot our way there.”
Cahn: AGI Is Decades Away and Not Happening Anytime Soon
“It's like, we're probably gonna be fine. AI is probably gonna change our lives and make it better. It's gonna improve productivity. Maybe AGI is gonna happen when you and I are like, 90, but I don't think AGI is happening like tomorrow.”
David Cahn predicts China will catch up in AI development
“I think that China is going to catch up.”
Cahn: GPUs Account for Half of Total AI Data Center Costs
“About half the cost of the data center is GPUs, half is everything else.”
Cahn: Tech Giants Will Increasingly Use Off-Balance-Sheet AI Financing
“And so I think one thing that we're going to see more of is off balance sheet financing.”
Cahn: Real estate investors supply significant capital for AI data centers
“A lot of the capital going to fund data centers comes from real estate developers. Real estate investors”
David Cahn: AI Data Center Debt Yield Spreads Will Compress
“So I think prices will get bit like the spread between the risk-free You know, return and the sort of risk adjusted return will compress, which just means that it's going to be less of a good deal as more people do it.”
David Cahn: Expect More Deals Like Blackstone's CoreWeave Debt Financing
“And so you know, Blackstone did this core weave deal. They were one of the first and Blackstone are generally very good at structuring these type of creative deals. Good on them. And I think we'll see more of that type of deals going forward.”
Cahn: Focus on Software Customer Behavior Over Stated Churn Intentions
“So I think the lesson is you have to listen to what customers do. They continue to pay for the, especially expensive products. If people are paying for expensive products and telling you they're going to churn, you probably shouldn't listen to that. They're pa…”
Stebbings: Investors miss deals by wrongly assuming incumbents will build software
“So many of the great investments have been missed because people say, ah, but they could build it themselves. And they so rarely do it.”
Cahn: Replicating 80% of 10 peers' distinct superpowers drives investor success
“I'm probably not gonna be as hundred percent as good as you if this is your superpower, but I'd like to be 80% as good at you at your superpower, and if I can be 80% as good as 10 people on each of their 10 superpowers, then I'll probably be pretty, pretty goo…”
Cahn: Generating $1B+ Gains Is the Only Definition of Success in VC
“There is a definition of success in this business. You're either a slugger or you are not. If you are generating billion dollar gains, you are a slugger. Doug Leone is a slugger. Pat Grady is a slugger. And Jurita is a slugger, right? So you either are sluggin…”
Cahn: Hardest part of VC is taking personal risk, not finding good deals
“And something I like to say to people is I don't think the hardest part of venture is figuring out what are the good companies. I think the hardest part of venture is figuring out when are you going to put your neck on the line for this company and all good pa…”
Sequoia hires investors, breaks them down, and rebuilds them
“There's this great line that people at Sequoia say, which is Sequoia hires investors, breaks them down, and then rebuilds them in the Sequoia mold.”
Cahn: Sequoia Defines Success as Backing a Top 50 Global Company
“The bar for success is so, so, so high. Unless this company is one of the 50 most important companies in the world. You haven't done your job well.”
Cahn: Optimal Venture Strategy Today Is a Barbell Approach
“And so I think the right strategy is this barbell approach, right? Where you're doing early companies and you're also doing late companies.”
David Cahn: Building a $10B+ company requires four distinct founder attributes
“I think if you have one of these, you can be very good. I think if you have two of these, you can be great. I think if you have three of these, you can build a billion dollar company. You need all four of these to build a ten billion dollar company, a hundred …”
Cahn: Deal selection is more critical than winning or sourcing in VC
“Selection is the most important and I think, and then I think winning comes behind that and sourcing probably is actually, maybe it's tied, but I think it's probably third.”
Cahn: Joining Sequoia Did Not Increase Personal Deal Flow
“I think my deal flow basically didn't change at all.”
Cahn: The best founders do not want to meet with VCs
“The best founders don't want to meet with VCs. That's the reality of it. You have to convince them.”
Cahn Used Personalized Cameo Videos to Get Founders' Attention Early On
“One thing I used to do is I went through this phase where every time I met a founder that I really liked, I would go through their Twitter. I'd figure out something they really liked, let's say some TV show, and then I'd go in cameo and I would find the actres…”
Cahn recorded one Loom video every day for over a year for founder outreach
“There was a phase that I went through for more than a year where I recorded a Loom video, one Loom video every single day. And it was just at seven PM. I would record a Loom video. That was what I needed to do. And I would find the coolest product I could find…”
Cahn: I Believe in God, While Most in Silicon Valley Do Not
“I believe in God. Most people in Silicon Valley do not.”
Cahn: Founders Should Offer Up to 5% Equity for Transformational Hires
“I tell founders all the time, if you find someone who's going to transform your business, give them five percent of the company. Now I say that it's an extreme statement. Maybe it's four percent, maybe it's three percent. But I think we get very married to lik…”
Cahn: ChatGPT Is Now a $3.4 Billion Revenue Business
“Definitely OpenAI. I mean, OpenAI has a brand, it has consumer distribution, it has product market fit with ChatGBT, which is a 3.4 billion dollar revenue business now. I think ChatGBT will definitely be around. Yeah, I think a bunch of the others will too for…”
Shorts cut from this episode
▶ Do you agree with Elon’s advice? 🧠 · 20VC with Harry Stebbi (@19:15)
▶ Simple trick to boost your replies! 🚀 · 20VC with Harry Ste (@1:05:15)
▶ Would you give away 5% of your business? 👋 · 20VC with Harr (@1:09:58)
▶ How to build an AI data center 🤖 · 20VC with Harry Stebbing (@16:32)
▶ How to play the AI game 🎮 · 20VC with Harry Stebbings (@0:16)