Who is behind OpenAI, and why does Microsoft have such a big stake?

When people say "OpenAI made ChatGPT," it's natural to picture an ordinary tech startup. OpenAI's setup is more unusual than that, and it explains why one of the world's largest software companies is so closely tied to it. Understanding the arrangement also helps you make sense of why the same AI keeps turning up in so many different places.

Two layers: a nonprofit on top, a company below

OpenAI started in December 2015 as a nonprofit research lab. Its founders included Sam Altman, now its CEO, along with Elon Musk, Greg Brockman, Ilya Sutskever, Wojciech Zaremba and John Schulman.

By 2019 the lab had concluded that building advanced AI would cost billions of dollars over the coming years, for cloud computing, specialized supercomputers and staff. Donations were not going to cover that. In March 2019 it created OpenAI LP, which it called a capped-profit company, a hybrid of a for-profit business and a nonprofit.

Here is how the two layers fit together:

• OpenAI Inc., the original nonprofit, sits at the top. Its board controls the company below it.
• OpenAI LP, the capped-profit company, raises money from investors and employs most of the staff. Its first duty is to OpenAI's charter, its statement of mission, not to its investors.
• Investors' returns are capped. For the first round of investors, the cap is 100 times what they put in; OpenAI said later rounds would have lower caps. Any profit above the cap belongs to the nonprofit.

A rule also limits conflicts of interest: only a minority of the nonprofit's board may hold a financial stake in the company.

How Microsoft became the biggest partner

Microsoft has invested three times. In July 2019 it put in $1 billion and became OpenAI's exclusive cloud provider. A second investment followed in 2021. In January 2023, Microsoft announced a third, multiyear, multibillion-dollar investment. It didn't state the amount; news reports earlier that month had described talks of up to $10 billion.

Each side gets something it badly needs.

OpenAI gets computing power. Training large models like GPT-4 takes an enormous amount of computing. Microsoft builds supercomputers for OpenAI's research on its Azure cloud, and Azure remains the exclusive cloud for OpenAI's research, products and API.

Microsoft gets the models. It can build OpenAI's technology into its own consumer and business products, and it sells access to the models to other companies through its Azure OpenAI Service. Microsoft says models trained on Azure already power ChatGPT, DALL·E 2 and GitHub Copilot.

The deal is not a purchase. OpenAI remains a separate organization that sells its own products, such as ChatGPT Plus and its API, and both companies may commercialize the resulting technology independently.

Where OpenAI's models show up in Microsoft products

As of mid-May 2023, here is a snapshot of the main places.

Microsoft stresses that 365 Copilot is more than ChatGPT pasted into Office. It combines the language model with your work files, email and calendar, and Microsoft says it only shows data you already have permission to see and that customer data is not used to train the models.

What does that mean for the tools you use?

First, you are probably meeting OpenAI's models in more places than you realize. A chat answer in Bing, a code suggestion in GitHub Copilot and, soon, a draft paragraph in Word may all rest on related technology. Learning to write clear prompts and check answers in ChatGPT carries over directly to those tools.

Second, the same model can behave differently depending on where you use it. Microsoft wraps the models in its own instructions, search results and safety rules. Bing's chat, for example, searches the web and shows citations, while ChatGPT without plugins answers from what it learned in training. Judge each tool on its own results rather than assuming they are identical.

Third, keep an eye on where your data goes. Your conversations in ChatGPT fall under OpenAI's policies; Microsoft's business tools fall under Microsoft's commitments to its customers. If you use these tools at work, ask your organization which ones are approved and what information may go into them. The partnership ties the two companies together, but the rules for your data still depend on which door you walk through.