Great analysis as always. But here’s what I don’t understand: if you’re the cursor founders or early investors, why would you sell equity in your new round at a $50b valuation and then at the same time sell the entire company at $60b? Didn’t you just give investors in the latest round a 20% ROI with virtually no risk?
I get it - there are operational conditions and benchmarks to hit to get that $60b sale, but they can’t be very tough to hit if everyone’s calling this acquisition a fait accompli.
Help me understand what just happened here? Someone just left a ton of money on the table.
"Generative AI subscriptions are like if Uber charged users $20 a month for 100 rides of any distance under 100 miles, and if gas was $150 a gallon, and Uber paid for the gas because somebody insisted that oil would one day be too cheap to meter.
Uber would, eventually, decide to start charging users a monthly subscription to access rides, and bill them for the gas that they consumed. Suddenly users would go from paying $20 a month for 100 rides to paying $20 to access a driver and $26 for a 10 mile drive."
I'd also argue that a 40% increase is nowhere near sufficient to get tokens to break even.
Nice narrative. I completely disagree with the attempt to turn this single case into a broad parable. Cursor knew their arbitrage was temporary, found a buyer needing what they had, and took the logical step of cashing in at a good time. This says nothing about application businesses that add real value: one case is not enough to support the general claim that "the application layer doesn't get champions". (It might be the case that the examples cited here are also just thin wrappers and vulnerable, but the claim is too strong as stated.) I am also not convinced that Anthropic's strategy was focused on Cursor much; responding to OpenAI seems enough to explain the history.
it will be interesting: cursor is trying hard to have more compute for "developers". Claude, once revered as the best developer ai, has been branching around adding more features for the general users (cowork, office app integrations, etc.).
Maybe an area to consier is Anthropic blocks China and Hong Kong, so we have not choice but to use Cursor, maynbe there is an opportunity for an Asia based push on this front who is not Racist.
Great writeup. One issue though - cursor has already tried hard to get their users to use other models (their own and chat-gpt) essentially acting as a token marketplace. However - users keep choosing anthropic. They can change the default to grok, but if users keep choosing anthropic then they are still in trouble. They could wholesale force grok usage (or their grok GPU trained models) but that will cost them customers and might really endanger their enterprise contracts and future sales (enterprises want optionality).
wow super fun read
Great analysis as always. But here’s what I don’t understand: if you’re the cursor founders or early investors, why would you sell equity in your new round at a $50b valuation and then at the same time sell the entire company at $60b? Didn’t you just give investors in the latest round a 20% ROI with virtually no risk?
I get it - there are operational conditions and benchmarks to hit to get that $60b sale, but they can’t be very tough to hit if everyone’s calling this acquisition a fait accompli.
Help me understand what just happened here? Someone just left a ton of money on the table.
My gut feeling is that whoever led the 50b round has vested interest and facilitated the merger. A16z, thrive, battery ventures
the Cursor-shaped reason to subsidize claude code might be gone, but the Codex-shaped reason might very well take its place
Interesting read, though the Uber comparison rings a bit hollow. Ed Zitron's article (https://www.wheresyoured.at/ais-economics-dont-make-sense/#generative-ai-subscriptions-are-nothing-like-uber) on the subject is a great read:
"Generative AI subscriptions are like if Uber charged users $20 a month for 100 rides of any distance under 100 miles, and if gas was $150 a gallon, and Uber paid for the gas because somebody insisted that oil would one day be too cheap to meter.
Uber would, eventually, decide to start charging users a monthly subscription to access rides, and bill them for the gas that they consumed. Suddenly users would go from paying $20 a month for 100 rides to paying $20 to access a driver and $26 for a 10 mile drive."
I'd also argue that a 40% increase is nowhere near sufficient to get tokens to break even.
Nice narrative. I completely disagree with the attempt to turn this single case into a broad parable. Cursor knew their arbitrage was temporary, found a buyer needing what they had, and took the logical step of cashing in at a good time. This says nothing about application businesses that add real value: one case is not enough to support the general claim that "the application layer doesn't get champions". (It might be the case that the examples cited here are also just thin wrappers and vulnerable, but the claim is too strong as stated.) I am also not convinced that Anthropic's strategy was focused on Cursor much; responding to OpenAI seems enough to explain the history.
Doesn't this thesis fall apart against musk giving anthropic access to their data center?
it will be interesting: cursor is trying hard to have more compute for "developers". Claude, once revered as the best developer ai, has been branching around adding more features for the general users (cowork, office app integrations, etc.).
Great read. If Cursor helps prop xAI to become a credible competitor to OpenAI and Anthropic, that’s a good outcome. Markets doing their work.
Maybe an area to consier is Anthropic blocks China and Hong Kong, so we have not choice but to use Cursor, maynbe there is an opportunity for an Asia based push on this front who is not Racist.
great read
Great writeup. One issue though - cursor has already tried hard to get their users to use other models (their own and chat-gpt) essentially acting as a token marketplace. However - users keep choosing anthropic. They can change the default to grok, but if users keep choosing anthropic then they are still in trouble. They could wholesale force grok usage (or their grok GPU trained models) but that will cost them customers and might really endanger their enterprise contracts and future sales (enterprises want optionality).