Microsoft’s hedge against OpenAI makes perfect sense

At the drop of a hat, OpenAI’s board could decide to stop sharing its technology with Microsoft in much the same way it decided to fire Altman. That puts CEO Nadella in a precarious position — and makes his efforts to build powerful AI internally all the more reasonable.

By Parmy Olson

When Microsoft Corp. invested more than $10 billion for a chunk of OpenAI, scientists inside its storied research division were rankled about being shoved aside for a newer player from outside the company. Microsoft Research was more than 30 years old and stuffed with esteemed scientists who’d won Turing Awards and Fields Medals, and here was Chief Executive Officer Satya Nadella last summer berating them for falling behind Alphabet Inc.’s Google on AI research. But Nadella wasn’t switching his focus from one research team to another. He was doing what he does best: branching out.

It appears that Nadella was already working on a strategic hedge against OpenAI, building its own large language model technology. Microsoft Chief Technology Officer Kevin Scott confirmed in a LinkedIn post that the company was working on an AI model called MAI-1, which, according to an earlier report in The Information, would be large enough to “compete” with OpenAI.

That same fear of missing out is what fueled Microsoft’s first investment in OpenAI in the first place. In June of 2019, Scott warned Nadella about the advancements OpenAI was making, according to emails released from the US Justice Department’s antitrust investigation into Google. Scott admitted to being highly dismissive of the company’s work at first, before realizing both Google and OpenAI could now process human language in ways that “we couldn’t easily replicate.” As he looked closer at the gap between Microsoft and the frontrunners, "I got very, very worried,” Scott added.

Nadella then looped in Microsoft’s chief financial officer, Amy Hood, to the email chain. “[This is] why I want to do this,” he said, referring to what would become Microsoft’s initial $1 billion investment in OpenAI the following month.

But while some CEOs might have let their new partner forge the company’s strategic path, Nadella didn’t. His radical drive to diversify Microsoft’s business in the last 10 years reflects that approach to curbing the company’s reliance on a single dominant product like its Windows operating system. When he saw declines in PC shipments on the horizon in 2014, Nadella pushed Microsoft into the cloud business, eventually grabbing more than 20% of the global cloud-computing market.

That same safeguarding strategy could also help if OpenAI pulls the rug out from under Microsoft again. Last year’s dramatic firing and return of OpenAI CEO Sam Altman may have led Microsoft to getting an observer seat on its more corporate-friendly board, but the AI firm’s unusual structure still makes it a ticking time bomb for Nadella.

It remains “an entirely independent company” governed by its nonprofit board, according to OpenAI’s website. That board decides if and when OpenAI has finally built human-level “artificial general intelligence,” or AGI, to which Microsoft isn’t entitled. “AGI is explicitly carved out of all commercial and IP licensing agreements,” the company says.

At the drop of a hat, OpenAI’s board could decide to stop sharing its technology with Microsoft in much the same way it decided to fire Altman. That puts Nadella in a precarious position — and makes his efforts to build powerful AI internally all the more reasonable.

The success of that effort is another matter. Microsoft’s internal AI teams have baggage to work through: the humiliation of falling behind Google, the rebukes from Nadella and the historic gaffes with projects like Tay, the chatbot that spouted racist, abusive messages. Mustafa Suleyman, a co-founder of Google DeepMind, is the new face of Microsoft’s AI efforts. Former colleagues call him a charismatic Pied Piper for engineers. That bodes well for making Nadella’s Plan B to work — if it’s ever really needed.