Why The Laws of Startup Physics Have Changed | Ben Horowitz Interview

Summary

• AI’s economic impact arrives in the next 12–24 months, Horowitz argues, because unlike cars (roads) or the internet (fiber, smartphones), there is no infrastructure to build for adoption: “the internet is here… there is no infrastructure that needs to be built to adopt the thing.” Alongside lower energy prices, less regulation, and a friendlier tax code, “the economy is in much better shape than people realize.”
• The laws of physics of company building changed. The old certainty — you can’t throw money at software (“what’s a man year? 700 IBMers before lunch”) — has changed: with data and enough GPUs, Elon caught the big models and “got in the game very fast.” Markets might be $5 trillion instead of $50 billion, yet leaders can now be caught — “these are just concepts we’ve not dealt with,” changing how venture thinks about valuations and long-term value.
• The market is overreacting on incumbent software — Salesforce and SAP are “extremely hard targets… even with AI” — while AI-native revenue compounds at unprecedented speed: Cursor passed $1B in revenue “in no time” in a category whose prior leader took 12–15 years. And billion-dollar researcher pay is rational math: building giant models is “a little bit alchemistic,” unlearnable in school — “what if there were only 40 of them in the world?” against a $4 trillion company.
• Policy is the tail risk in a rerun of the industrial revolution, which America won and converted into military, economic, and cultural dominance — “we’re there again.” Venezuela was the fourth-richest country in the world before communism, and the Biden executive order requiring federal approval to sell a GPU meant “we were that close to being basically out of the global chip game.”
• Job-doom is falsely confident: ImageNet was 2012, ChatGPT 2022 — “where’s all the job destruction? Why hasn’t it happened yet? And why are you so sure it’s going to happen next?” He also rejects the permanent-underclass meme: hyperbolic new technology is exactly how small capital compounds — “if you bought Bitcoin for a nickel, you did really well. And all you needed was a nickel.”
• The Andreessen Horowitz playbook, restated: venture was “a great product for LPs, but not a great product for entrepreneurs,” so they built the founder-to-CEO product and marketed when nobody did. The next job is supporting companies longer in private markets — the IPO threshold moved from $50M revenue to roughly $1B — while refusing the AI private-equity rollup wave despite calling it “a really good business idea”: it’s the cultural opposite of a firm where “price doesn’t even matter.”
• The Las Vegas police project is his proof that technology beats policy: a drone on scene within 90 seconds of any 911 call or gunshot, crime, he thinks, down over 50%, police shootings of suspects down close to 75%, and a 94% murder clearance rate. “Defund the police did not make anybody safer — technology does.”

Deep dive

1. America 2026 is healthy — and AI needs no infrastructure build-out

• Ben’s read on the country: the tech sector is “very very healthy,” competitiveness strong, the entrepreneurship culture “outstanding.” Everyone abroad wants Silicon Valley, and Europe has the talent and universities — but a worsening regulatory environment and a cultural gap: “succeeding, doing something larger than yourself, making the world a better place — those aren’t things that young people feel like society values,” so founders may struggle to recruit people for such missions.
• The economy is better than people realize: lower energy prices, much less regulation, a more user-friendly tax code — “that’s all starting to kick in now.”
• Why the 12–24 month window matters: past technology deployments needed infrastructure — roads and traffic lights for cars, fiber and smartphones for the internet. “The internet is here. If you want to apply AI to your business, you just do it.”

2. Policy can ruin it — and the stakes are the industrial revolution rerun

• His father’s warning frames the risk: “a bad government, no matter how many smart people you have… can ruin the whole thing.” Venezuela was the fourth-richest country in the world; Romania and Hungary produced geniuses and then went “from inventing everything to nothing overnight” once the communists took over.
• The near miss: the Biden executive order that said you could not sell a GPU without federal government approval. It was reversed, “but we were that close to being basically out of the global chip game. It is fragile.”
• Technology solutions beat policy solutions: lockdowns versus a vaccine; European emissions cuts that “didn’t do anything because China didn’t reduce emissions” versus building safe nuclear or fusion; “defund the police did not make anybody safer — technology does.” The corollary: “If you want to change the world for the better, it’s never been a better time to be an entrepreneur.”
• The through-line of his ambition: “We won the industrial revolution. We really did.” Ford and Edison’s technology “led to a military lead, led to an economic lead, led to cultural dominance” and winning World War II. AI is the equivalent shift — “we’re there again” — and if America isn’t the provider of this technology, it stops being the superpower and standard of the world.

3. The laws of physics of company building just changed

• Everything is up for grabs, but the market is overreacting on incumbents: “it’s not that easy to take out Salesforce or SAP — you would be surprised, even with AI, how much heavy lifting that is.” Meanwhile the number of possible interesting companies “went up a lot,” and AI companies grow revenue faster than prior technology companies: Cursor, “ostensibly an IDE” — a prior IDE took maybe 12–15 years to build — went over $1B in revenue “in no time.”
• The broken law: you used to know you couldn’t throw money at software (“what’s a man year? 700 IBMers before lunch”) — Google hiring 2,000 engineers would never catch a great small team. Now, with data and enough GPUs, “you can solve damn near anything”: Elon caught the big models with money, a really good data-center design, and smart engineers. “He got in the game very fast. That would have never happened in the past.”
• Both tails widened at once: “when you calculate the long-term value, what if this market wasn’t $50 billion — what if it was $5 trillion? And on the other end, what if somebody could catch you? These are just concepts we’ve not dealt with.”
• The billion-dollar researcher, demystified: building a giant model is “a little bit alchemistic in nature… a little bit of an art” you can’t learn in school — if you weren’t one of the main people handed hundreds of millions at Google, Facebook, OpenAI, or Anthropic, you probably can’t do it. “What if there were only 40 of them in the world?” and you have a $4 trillion company. “We track all of them.”

4. Inequality is a feature — and the job-doom math doesn’t check out

• AI extends the Kobe Bryant effect: in Naismith’s day you earned from whoever showed up to the game; television and global audiences made LeBron a billionaire. The internet gave products very fast global distribution; AI is “another layer on top” — the inventor gets internet-company-plus rich. The offset is day-one democratization: “super intelligence in your phone” and “a super advanced amazing tutor” for every child — “I don’t think we’ve ever seen a bigger opportunity equalizer than AI.”
• The father’s lesson, applied to systems: “Son, life isn’t fair” — and a system built to correct that “doesn’t make things more fair, it just transfers all the power to the person running the system.” Stalin, likely Ceaușescu, Pol Pot, Mao: “not an accident that every single system like that went bad.” What you want instead: “Don’t give me no chance. Give me some chance.”
• On the Twitter meme that you have a couple of years to get capital or join the permanent underclass: “I don’t think the door is going to close behind you — opportunities tend to multiply.” Crypto minted people who “literally didn’t have much to start with,” because hyperbolic growth is precisely where small capital compounds: “if you bought Bitcoin for a nickel, you did really well. And all you needed was a nickel.”
• Labor predictions carry false certainty: the speaker cites likely ImageNet in 2012, the natural-language work around 2015, and ChatGPT in 2022 — he estimates 95–96% of all US jobs were once in agriculture, and nearly all those jobs were eliminated; people doing agriculture wouldn’t even consider today’s work to be jobs. Then the timing rant: “where’s all the job destruction? Why hasn’t it happened yet? And why are you so sure it’s going to happen next? And why are you so sure no jobs are going to be created?”

5. Management is psychologically brutal, not conceptually hard

• From Andy Grove — “maybe the greatest tech CEO we’ve had” — the core insight: management concepts need “an eighth-grade education… it’s not like physics,” but the psychology is “extremely difficult, particularly for a young person”: super confrontational, and “the good of the whole supersedes the good of the individual.” Ben calls his foreword to High Output Management “the best thing I ever wrote”; The Hard Thing About Hard Things “was basically intended to be the updated version of it.”
• The Grove specimen, as told: Intel’s lowest-scoring Santa Clara facility, a roll of toilet paper placed under his chair, and when the excuses started — “I reached under my chair, put all the toilet paper up. I said, clean up your [__] and tell me when you’re going to be up to code.” Up to code in two months, highest-rated facility thereafter, and a 1992 “Manager of the Year” award on the wall of the world’s biggest CEO.
• The founder failure loop: early mistakes → lost confidence → hesitation → deferring to executives who lack the leader’s full context → a political vacuum where “you’re not making the decision, I’ll make the decision.” The hardest case is the reorg: someone good will lose power and “be pissed,” and if you compromise the org so they keep it, “you’ve just redistributed power from the people doing all the work to the executives — and that’s a catastrophe.” Inexperienced founders pick “the known avoid-hurt over the theoretical avoid-hurt… and that’s when you wreck your company.”

6. The Andreessen Horowitz origin: venture was a great product for LPs, a bad one for founders

• The context: no genuinely new top-tier firm since the 80s (Benchmark, 1995, was really Merrill Pickard alumni). Reputation was the moat — you can’t retroactively have backed Apple, Cisco, and Google — and tier two won’t last, because “the best entrepreneurs will only work with the top-tier firms… that’s why the tier ones always have better returns.”
• The wedge: “venture capital is a great product for LPs, but it’s not a great product for entrepreneurs.” Build the firm designed to give a founder “enough confidence, power, network reach, advice that you could actually be a CEO.”
• The second insight — VCs never marketed — Ben traces to the first financier class (JP Morgan, Rothschild, Goldman Sachs) keeping quiet after “financing both sides of World War II” [as said], a silence that carried through Arthur Rock. So when Andreessen Horowitz talked, “everybody covered it.” Even the name was a fix: 2009 LPs feared the founders would bail to start another company, so — “why don’t we just name it with our names, and then they know we’re safe.”
• The early arc, mistakes included: fund one ($300M — Skype, Slack, Okta, Stripe) had “too many good things… not to blow the doors off”; fund two was weaker; fund three was “scary for a while” amid contention over the GP profile before Coinbase, Databricks, Lyft, and GitHub made it great. The confessed error was requiring investors to have founded or run companies: “most CEOs aren’t as interested in investing as they think they are” — and aren’t as good at teaching the job.

7. Venture at scale: support companies longer — and skip the rollups

• Grove’s lesson frames the firm’s ambition today: “if you’re the leader in the industry, then the growth of the industry is dependent on you… nobody else is going to do it.” The structure that enabled scale: multiple 4–5-person investing teams, each with its own founder-support platform — “a little bit of a novel idea” begun in earnest with the crypto fund around 2018, now how the whole firm runs.
• Capital markets flipped underneath venture: Ben went public “at 18 months old with $2 million in trailing revenue — that wasn’t a good idea,” but companies “used to go public routinely with $50 million in revenue.” Now the bar is around $1 billion, “and you’re kind of small if you don’t have that” — so companies need far more support from private markets than VCs were built to supply. At $200–300M of revenue they must go multi-product, multi-channel, multi-geography: “How do I get to Japan? How do I get to South America?”
• On the “big huge wave” of AI private-equity rollups: “just like the spreadsheet created the original private-equity business, AI is creating a new private-equity business… a really good business idea” — and not something Andreessen Horowitz will do. It’s the cultural opposite (Mark Rowan at dinner: “entry price, entry price, entry price,” versus venture where “price doesn’t even matter — as long as the thing succeeds, you’re going to do well”), and Ben won’t be in a business whose edge is optimizing an existing thing and laying people off: “I’ll leave that to the other smart guys in the industry.”

8. Culture is a set of actions — and the father who wired the worldview

• The most important insight, from Bushido: “a culture is not a set of ideas. It’s a set of actions.” Corporate values — integrity, do the right thing — are “a bunch of [__] platitudes”; culture must be defined as exact behaviors, down to the SLA on returning a Slack message and showing up to meetings on time.
• The Andreessen Horowitz behaviors: never late to an entrepreneur meeting (Ben fined people $10 a minute in the early days); every “no” must be explained, and the entrepreneur gets surveyed afterward to confirm it; trash a founder publicly (“he’s selling dollars for 85 cents”) and you’re fired — “We’re dream builders, we’re not dream killers.” Enforcement is personal: nobody joins without signing the culture document, and Ben spends an hour teaching it to every single employee.
• The source code: his father was a “red diaper baby” — communist grandparents with party cards, a grandfather fired as a teacher in the McCarthy era — who edited Ramparts and was involved with the Black Panthers before reemerging on the right. The line that made Ben a systems thinker: pick any book on socialism and “you will find page upon page, chapter upon chapter of how to divide the wealth. You will not find a single sentence on how to create it.”
• The parenting register, as told: 102 degrees, broken air conditioning, a gallon of apple juice steaming out of the rug — “Son, you know what’s cheap? Flowers. You know what’s expensive? Divorce.” (“He had been married four times, so he knew what he was talking about.”)

9. The creative frontier: coding turned a corner; Hollywood and music are next

• Over the winter break, coding crossed a threshold: “really really good programmers were going, whoa… I just became a hundred times more productive. I can’t remember any technology where all of a sudden you wake up and the whole world just changed” — and that, he says, is now happening on a pretty regular basis.
• AI is “a little underestimated as a tool for creatives”: an actress who shoots a scene 15–20 times could shoot it three and let AI generate variations. In music, his hip-hop parallel — dismissed at birth as “a bunch of [__], a novelty,” but actually postmodern art: “the invention of the new art form is when it gets really exciting.”
• The Nas friendship doubles as sourcing: listening to Rakim (kind of like the likely John Coltrane of rap), Nas decoded a line Ben had heard a thousand times — “he’s passing out cigars at the birth of knowledge.” Two weeks after Nas called to learn about Bitcoin, Chris Dixon mentioned one of the Coinbase founders, Fred, was “really into hip-hop”; Ben invited Nas to a boxing match at his house — “that’s how we got that deal.” His cross-domain comp for Nas’s timelessness: Jensen — same identity across gaming, Bitcoin, and AI eras, and “it never felt like he’s trying to be current, which — like Nas never feels like he’s trying to write a hit.”

10. Las Vegas: proof that intelligence, not force, makes policing safer

• Why Vegas was the testbed: the force answers to an elected sheriff, not the mayor, so it never got caught in defund-the-police, never militarized, and does community policing — visible in a 94% murder clearance rate versus ~75% in San Francisco, the 30s in Chicago, and below 60 nationally. Sheriff Kevin McMahill’s explanation: “when somebody is murdered, there’s always somebody who knows who did it. They just don’t talk to the police. But they talk to us, because we’re part of the community.”
• Ben is personally funding “the highest-tech police force in America, hopefully the world”: a drone program, Prepared 911, and AI cameras (likely Flock Safety). Any 911 call or gunshot puts a drone on scene within 90 seconds, its video feed instantly on every nearby officer’s phone. Since the program began, he thinks crime is down over 50%, and police shootings of suspects are down close to 75%.
• The surprise finding: bad descriptions “cause like half the violent confrontations” — the 2004 blue Hyundai that’s really a 2008 green one, pulled over with a baby in the back seat and a gun in the car. With AI cameras, “we know that’s the car, and we know there’s a baby in the car,” so a full squad apprehends safely: “policing is inherently dangerous, but intelligence makes it dramatically safer” — for suspects, citizens, and police alike.
• The knock-on effect: recruiting flipped. Between the state-of-the-art drone center and “amazingly futuristic” Cybertrucks, “everybody wants to be a police now” — standards rose instead of falling, drawing on the country’s highest concentration of veterans.