All unsuccessful startups are alike; each successful startup is successful in its own way.
Sam Altman reflected a few months ago on the advice he’s given over the years and, ultimately, what has led him to (even more) success with OpenAI.
I feel so bad about the advice that I gave while running YC that I’m thinking about deleting my entire blog. There were a lot of things that we really held dear — you have to launch right away, you’ve got to launch a first version you’re embarrassed about, raise very little capital upfront, don’t take big R&D risk, you’ve got to immediately find product-market fit. OpenAI raised a billion dollars of capital before any product at all. It took us 4.5 years after we started to release something, and when we released it, we didn’t talk to users for awhile. We didn’t do it the same way, and it still worked. — an interview with Sam Altman
So, what exactly has OpenAI done differently? Expanding on Altman’s comments and adding a few others.
- Going to market with a consumer and an enterprise product. ChatGPT Enterprise just launched. There’s ChatGPT Plus for $20/month for consumers. Is ChatGPT just product-led growth for the enterprise product, or will OpenAI run two playbooks: one to become the next Google and the other to become the next Microsoft?
- Fundraising through complicated financial structures (in complex ways). OpenAI started as a nonprofit organization with over $1 billion in donation commitments (it received $130.5 million of those) before transitioning to a “capped profit” structure in 2019. Not to mention the 49% stake and profit-sharing agreement it made with Microsoft in January 2023.
- No commercial product for the first 4.5 years. OpenAI released its first product, an API, in June 2020.
- Product behind a login wall. You have to sign up before you use it. Today’s products have significant low-friction freemium motions. That might be open-source for developers or a page where people can spin the wheels on the product before committing further. The last company to have this sort of success behind a login wall was probably Facebook.
- No social, sharing, or other viral features (initially). No other product in the history of consumer products has grown so quickly without any social features. There’s no viral loop. No asking to access your contacts and crawl your existing social network. No notifying your coworkers, friends, or family that you’ve made a text completion.
- Capital-intensive business. The opposite of the 2010s-era Lean Startup advice.
- Solution in search of a problem. Developers, enterprises, and consumers are still figuring out exactly how to use these reasoning machines. Some early use cases have real traction (code completion), but others are still nascent.
- No proprietary data to train models on. The company was in a unique position because of the expertise and network of its key employees, so it’s possible they had access to privileged data sources (like Reddit, a longtime YC company and an early user of GPT-3). But for the most part, the company had no data of its own. Before OpenAI, the rule was that a unique and proprietary dataset was much more important than anything else in machine learning startups.