OpenAI is one of the best cloud customers in the world right now. That could change in the future, though.
In a clear warning shot over future competition, OpenAI Chief Financial Officer Sarah Friar said cloud providers have been “learning on our dime,” and she wants to ensure the startup isn’t giving away its AI know-how to these tech giants.
OpenAI has massive computing needs to support its booming ChatGPT service and a growing enterprise AI business. That means the startup works with most of the major cloud providers now, including Microsoft, Google, Oracle, and CoreWeave.
“As we help our partners design the AI data center, as they learn, as we train in those data centers or even as we inference… they are, I sometimes think of it as a little bit learning our dime,” Friar told a packed crowd at a Goldman Sachs conference in San Francisco on Tuesday.
“What I want to make sure is that we’re not giving that IP away,” she added, referring to OpenAI’s intellectual property.
First-party builds
This is pushing OpenAI to start building its own AI data centers, Friar explained, referring to this as “first-party builds.”
The massive Stargate projects, which currently involve several large cloud providers, are something that OpenAI is beginning to talk about as a potential first-party build, the CFO noted.
Related stories
Business Insider tells the innovative stories you want to know
Business Insider tells the innovative stories you want to know
She laid out three steps involved in this transition. Initially, OpenAI buys compute capacity from existing cloud service providers “off the shelf.”
Next, OpenAI works with partners more closely so it can learn how to design and run data centers.
“And then I think you’ll see us do more 1P ourselves over time,” Friar said, referring to first-party data center buildouts.
This could be a long-term risk for some cloud providers. At the moment, AI companies are desperate for compute and are willing to work with outside providers and pay handsomely for AI infrastructure rented out via the cloud. However, if OpenAI starts building its own data centers in the future, it could become a rival, rather than a lucrative customer.
Still, this potential outcome is probably many years away. That’s because AI data centers take a long time to build, and construction is currently being restrained by a limited supply of AI chips, energy, and labor.
Sign up for BI’s Tech Memo newsletter here. Reach out to me via email at abarr@businessinsider.com.