#863: Elad Gil, Consigliere to Empire Builders — How to Spot Billion-Dollar Companies Before Everyone Else, The Misty AI Frontier, How Coke Beat Pepsi, When Consensus Pays, and Much More
The myth-busts, hot takes, explainers, and tools worth keeping.
•topic· 9
•topic05:00
The Compute Constraint and Why No Lab Can Pull Ahead
Gil explains the AI supply-chain bottleneck: right now it's a specific memory made largely by Korean fabs, expected to bind for about two years. Because every lab is equally constrained, none can buy 10x the compute of another, so OpenAI, Anthropic, and Google should stay roughly close for the next two years.
Training output is astonishingly a small flat file encoding humanity's knowledge plus reasoning
The current bottleneck is memory; earlier it was packaging, later it may be power
Fabs take years to build, so there's no quick workaround to the memory constraint
Equal constraint creates an artificial ceiling that keeps the labs close for ~2 years
“you have an artificial ceiling on how big a model can get in the short run”
#ai#compute#supply-chain#labs
•topic13:30
The Value-Maximizing Window: Sell or Never Sell
In every tech cycle 90-99% of companies go bust, so founders must ask if they're one of the durable handful. If not, there's usually a 6-12 month window where value peaks before a headwind hits, often visible first in the second derivative of growth.
1,500-2,000 companies went public in the dot-com era; only a dozen or two survived
Durable companies should never sell; everyone else should find their peak window
The plateau shows up first in the rate of growth, not the growth itself
Unprecedented buying power: 1% of a $3T market cap is $30B
“for every company there's a value maximizing moment where they hit their peak. And it's usually a window”
#exit#timing#m-and-a#growth
•topic28:30
Market First, and How Gil Got Into Airbnb and Stripe
Gil explains his market-first, team-second thesis and how his earliest deals came organically from helping founders. He introduced Airbnb to investors when they were eight people and cold-emailed Stripe's Patrick Collison offering to talk — both led to invitations to invest.
Great teams get crushed by bad markets; index on the market early
Access came from being useful, not from chasing deals
Offer advice and you often get to invest; offer money and you get asked for advice
He avoids 'science projects' with stacked science and market risk
“the first few things that I did were very organic where the founders were like, we want you on board”
#investing#market-first#access#origin
•topic26:30
Go to the Cluster: Geography Is Destiny for Access
Gil calls moving to an industry's physical cluster the single most important move for breaking in, dismissing 'work from anywhere' as BS. His team's analysis shows 91% of global private AI market cap sits in one 10x10 Bay Area zone.
Every industry aggregates: film in Hollywood, finance in New York, AI in the Bay Area
91% of global private AI market cap is in one 10x10 area
Defense tech clusters near SpaceX and Anduril in Southern California
Being in the cluster puts you in the networks where access happens
“all the advice that you can do anything from anywhere and everything's remote is all BS”
#clusters#geography#network#career
•topic49:00
Diligence That Collapses to One Belief
Gil does enormous diligence — CFO meetings, financial models, customer calls, even cash reconciliations no other fund does — but insists every deal collapses to one core belief. If it takes three things to be true it's too complicated; if zero, there's no thesis.
The power law is real: ~10 companies drove ~80% of two decades of returns
Regret is about under-investing in winners, not about losses
One-line theses: Coinbase = index on crypto; Stripe = index on e-commerce
Don't waste founders' time on questions that don't move the core belief
“What is the one thing I need to believe about this company that makes me think it's going to continue to be really big?”
#due-diligence#conviction#power-law#investing
•topic1:08:30
Why Now, Fake TAM, and Selling Labor Instead of Software
Gil frames great markets through 'why now' — regulatory, technology, or incumbency shifts that open a closed market. He distinguishes real from fake TAM (Coca-Cola's share-of-liquid reframing), and explains how generative AI shifted business from selling seats to selling units of labor, reopening markets like legal.
Markets open via regulatory, technology, or incumbency/competitive shifts
Fake TAM is a fraction of a giant aggregate you don't actually serve
Coca-Cola redefined its market from soda to all liquid sold, expanding ambition
AI shifted the sale from software seats to work-product and labor hours (Harvey/legal)
“what AI did is it shifted things from selling tools to selling work product or selling units of labor”
#why-now#tam#market-opening#ai-business-models
•topic1:00:00
Boards as In-Laws You Can't Fire
Gil argues founders should build boards as deliberately as they hire, writing a board job spec. Reid Hoffman: a board member at best is a co-founder you couldn't otherwise hire. Naval: valuation is temporary, control is forever — so take a better person over a slightly higher price.
Most companies build boards reactively instead of proactively
An investor with a contractual seat can't be fired for a decade
Write a board-member job spec like you would for any role
Get to know angels first — Gil added BlackRock's Sue Wagner to Color that way
“valuation is temporary but control is forever”
#boards#governance#control#founders
•topic1:04:30
The Distribution Engine That Gets Edited Out of TED Talks
Gil punctures the 'it just grew organically' myth: category winners pair a great product engine with an aggressive distribution engine. Google paid to bundle its search toolbar everywhere, Facebook bought ads against people's own names, ByteDance spent billions distributing TikTok, and Snowflake spent billions on sales.
Almost every mega-cap company took an aggressive approach to distribution
Distribution can be product-built (Cursor word-of-mouth) or bought (ads, sales, channel)
Google distributed its toolbar by paying nearly every internet company to bundle it
Sometimes the best product loses to whoever out-distributed it
“the companies that are really good have an enormously good product engine. And then they have an amazing distribution engine”
#distribution#product#go-to-market#growth
•topic1:19:00
How Gil Consumes Information and Uses Multiple AI Models
Gil's information diet collapsed into X, technical papers, talking to smart people (20 minutes beats exhaustive research), and increasingly using multiple AI models at once for research — asking for primary sources and summary charts, then double-checking. He also experiments with cold-reading founders from photos.
20 minutes with a smart person beats exhaustive solo research
He runs 2-3 models at once, requests primary literature, and cross-checks
He keeps a roster of go-to experts for specific topics like longevity
For fun he prompts models to predict founder personality from micro-features in photos
“20 minutes with somebody really smart on a topic gives me more information and insights and leads on what to go read about than doing…”