@hpatel What’s the time horizon where margins go down to zero? When everyone is doing this one day pricing pressure will drive margins to unsustainable levels
Chegg, the homework site pretty much every college kid paid for at some point, was still worth about $3 billion the month ChatGPT came out. Today the whole company is worth about $80 million.
Over those same few years, four MIT friends built a coding app on top of the exact same AI models and grew it from $100 million to $4 billion a year in revenue. SpaceX bought it this summer.
Same technology. Greg was at a dinner in SF where someone asked the table one question that explains how it wiped out one company and made the other. If a new model drops next month, does your company get better or worse?
Half the table went quiet. You don't go quiet on a question like that unless you already know your answer.
Think about where the upgrade actually shows up. When OpenAI or Anthropic or Google ships a smarter model, Cursor plugs it in, and within days every customer is coding with a better assistant. Cursor didn't pay a dime for that research. The $20 a month people were already paying just bought them more.
Chegg sold students the answer to a homework problem. A smarter model hands out that answer for free in a chat window, so every new release made ChatGPT better at Chegg's job and gave millions of students one more reason to cancel.
Startups got hit the same way. Jasper raised money at a $1.5 billion valuation in October 2022 to sell AI-written marketing copy, mostly at around $80 a month. Six weeks later ChatGPT launched and did a decent version of it for free. Jasper's revenue had just doubled to about $80 million, and then growth stalled out. The next year the company marked down its own valuation by 20%.
Greg also noticed everyone's chasing $1 billion in revenue in 12 months and hardly anyone's talking about retention. Put that next to the dinner question and you can see the problem coming. Money that shows up because a model just got good enough can walk right back out when the next model does the job for free.
The founders with a good answer all built the same way. Greg says the best teams treat the model as maybe 20% of the product. The other 80% is the unglamorous stuff, like hooking into an industry's data, tests written by people who actually do the job, rules about what the AI is allowed to touch, and a screen where a human approves the work. When a better model comes out, it upgrades their 20% for free, and nobody at OpenAI can ship the rest because it lives inside the customer's systems.
So here's the test for any AI company you're thinking about working at or putting money into. When the next model comes out, who gets the upgrade first, the company or its customers? If it's the customers, they'll figure out pretty fast they don't need the company anymore.