The Answer in AI Is Probably

David George is a general partner at Andreessen Horowitz, where he leads the firm's growth fund. He joined in 2018 from General Atlantic, a more conservative growth shop, and was earlier an investor in Uber. The growth fund's first deal was Databricks at about a $6 billion valuation, and a16z has since done nearly every round through Series M. He estimates the firm holds about 20% of its growth market, backed by 650 employees who work with portfolio companies. These signals are from David George's interview with Jack Altman on Uncapped with Jack Altman.

1. It's All Going to Work. George agrees with Benchmark's Eric Vishria that open source and frontier models, Nvidia and new chip companies, will all succeed. His version: if your question is whether this or that will work, "the answer in AI is probably." AI infrastructure spending has passed the railroads as a share of GDP, at around 3%, and he expects it to grow 10x over many years. OpenAI, Anthropic and SpaceX AI together run at roughly $120 billion of revenue and add more each month than every hyperscaler except Amazon.
2. 30 Million Coders. Nearly all of that enterprise revenue comes from about 30 million coders, and within them a power law where the top million spend massively. The world has 1.5 billion knowledge workers, and George puts AI diffusion into the B2B economy below 5%. Legal is the next function in takeoff, about 12 months behind coding. Demand is "totally insatiable," while supply is short at every step of the chain, with new data center capacity unavailable until 2028.
3. The Bear Case Fails. George sees two ways the token boom could stall: coding turns out to be an anomaly, or an algorithmic breakthrough slashes compute needs. He calls the first extremely unlikely and the second possible in training, since people learn on 2 or 3 orders of magnitude less data than models do. Inference is the bigger market, and more reasoning at inference time keeps producing better answers. When Altman raised the risk that token sellers get rich while token buyers don't, George said that view assumes writing far more code far more efficiently produces nothing, which he called "fundamentally wrong."
4. Shorting Your Own Future. The best companies right now, Stripe among them, spend their AI productivity on building new products. A new product can bring 10x or 100x more revenue, while cost savings have a ceiling and will still be there later. George says a company focused only on cost is "shorting your own future." Procter & Gamble probably isn't seeing big gains yet, and companies like it will need the handholding that the labs, Databricks and Palantir provide.
5. The Frontier Premium. The 3 frontier labs take 95% or more of AI dollars, about $125 billion, and OpenAI and Anthropic alone have raised around $350 billion, more than the entire downstream market. Heavy users pay 10x more for frontier tokens because the work is worth so much and because Codex, Claude Code and Cursor tie their harnesses tightly to the models. N minus 1 and open source models will still be huge once usage grows 50x and cost starts to matter. Customer support stopped needing the top model about 3 years ago, and application companies watch this closely because token cost is their gross margin.
6. The Blast Radius. George expects the labs to build first-party products in coding and its "blast radius," plus horizontal tools for all 1.5 billion knowledge workers, in the mold of Microsoft Office and Google Apps. A vertical like legal sits somewhere around priority 6 to 15 for them. a16z is a large investor in Harvey, whose clients now demand their law firms use it; Kirkland's plan to spend $500 million on its own stack was a validation event. The moats are the last details of the product and real go-to-market work on the ground. Even inside the blast radius, Replit and Cursor are working, and he cites the old Microsoft rule that a platform is only a platform if what's built on it earns more than it does.
7. Consumer Is Still Skeuomorphic. More than a billion people use ChatGPT, mostly the way they used a search engine. George's 10-year-old son writes full rap songs and makes music with it. The shift comes when assistants move from reactive to proactive, and voice now handles interruptions well enough for natural conversation. He thinks the biggest prize will be killer consumer products, paid for by subscriptions and an ad format nobody has invented yet, since describing a social feed ad to a newspaper classifieds team would have made no sense. Of the automations he's built for himself, he gets real value from 2 out of 10.
8. Barbaric to Drive. Waymos have logged millions of miles at 10 to 14 times safer than human drivers, and George says history will judge harshly anyone who blocks them. Uber and Lyft grew San Francisco's $100 million taxi market to $1 billion within 3 years, and he expects autonomy to 10x ride hail again. Owning a car costs about $0.80 a mile fully loaded against more than $2 a mile for Uber, in a highly elastic market. With 17 million new cars sold a year in the U.S. and 250 million on the road, he puts the value of full autonomy at a minimum of $10,000 per car. The U.S. has fewer than 10,000 Waymos today.
9. Robots Beat Language. George expects robotics to become a bigger market than language models, with a ChatGPT moment for robots hopefully within 5 years. Home robots are far off, but defined, repetitive factory work pays today. a16z, through his partner Sarah Wang, backed Mind Robotics from Rivian founder RJ Scaringe, which puts robots on Rivian's assembly floor and trains models on that work. On jobs, he notes about 60% of today's jobs didn't exist 70 years ago.
10. Half the Returns Come Late. a16z's analysis showed half of private market returns come from seed through Series B and half from Series C onward, and with companies staying private longer George expects that to reach 70/30 in favor of late stage. The power law is stronger now because the cycle is early and because in AI, more money makes a company better. Pouring $40 billion into Salesforce would just buy sales reps; $40 billion of training makes models much better. He thinks the Vision Fund had the right idea at the wrong point in the cycle. When Altman mentioned Benchmark's first growth fund, George said, "Welcome to the dark side."
11. Founders Are the Asset Class. George rates the product cycle a 9 or 10 out of 10, against 1 in 2021 and 8 in 2010, and the capital cycle about a 6. The last wave of mobile, social, e-commerce, SaaS and cloud produced $25 trillion of market cap, and AI, autonomy, robotics, bio-health and defense should produce far more. a16z pays fair prices for great companies, and the alpha comes from revenue growth. When things go right, the founder usually finds the next thing: Ali Ghodsi's Databricks reaccelerated seven years after a16z's first growth check, and Starlink wasn't even generally available when the firm backed SpaceX.
12. Vibes Matter. In 2018, before George joined, Ben Horowitz asked about his marketing plan and called "we don't need one" the dumbest idea he'd ever heard. Vibes drive fundraising, valuation, hiring and retention, and sometimes customers. A hedge fund manager told George he never shorts a messianic founder or a product people love. Nobody knows what Palantir's "ontology" means, yet Alex Karp's following helped make Palantir the place CEOs trust to implement AI. George tells founders to go direct, own the narrative and be the face of the company, the way Palmer Luckey is for Anduril.

Watch the full video at https://www.youtube.com/watch?v=4H1FbNRdtM0.