OpenAI drift from Microsoft to Amazon turns aggressive after subtlety
Avisek Das | Light Rocket | Getty Images
OpenAI chief revenue officer Denise Dresser said the AI company agreed to roll out its models on Tuesday. Amazon had nothing to do with the announcement made the day before that the startup was restructuring its relationships Microsoft second time in six months.
“The two are not related at all,” Dresser said in an interview with CNBC following OpenAI’s announcement with Amazon.
Analysts aren’t so sure.
A lot has happened since late October, when OpenAI completed its recapitalization, giving Microsoft a 27% stake in the for-profit side of the AI company. As part of this arrangement, OpenAI agreed to purchase $250 billion worth of additional Azure services. And a revenue-sharing agreement will remain in place until an independent panel confirms that OpenAI has achieved artificial general intelligence, or AGI.
One of the biggest developments since then has been OpenAI’s move closer to Amazon, Microsoft’s biggest rival in cloud infrastructure.
In November, OpenAI announced a $38 billion commitment with Amazon Web Services. In late February, Amazon said it would invest $50 billion in OpenAI, which in turn would use 2 gigawatts worth of AWS’s custom Trainium chips to train its AI models.
Amazon and OpenAI also agreed to jointly develop “customized models” for Amazon’s engineering teams to power consumer products, increasing OpenAI’s spending commitment to AWS by $100 billion.
“That was the biggest thing that happened,” RBC Capital Markets analyst Rishi Jaluria, who recommended buying Microsoft shares, said in an interview.
This week’s one-two punch is the starkest sign that a dramatic shift is underway in the decade-old relationship between Microsoft and OpenAI.
It all started with OpenAI in 2016. to start runs its big experiments on Azure. Three years later, Microsoft invested its first $1 billion in OpenAI; this figure would increase to $13 billion over the next few rounds.
But in 2024, Microsoft began calling OpenAI a competitor in its financial disclosures, and early last year the software giant lost its title as OpenAI’s private cloud provider. In an internal memo earlier this month, Dresser wrote that OpenAI’s partnership with Microsoft “is foundational to our success” but “also limits our ability to meet organizations where they are.”
Against this backdrop, the latest deal between the two companies “appears to be fairly fluid and could change again within six months for all we know,” UBS analysts wrote in a note on Monday.
Additional elements of the agreement include ending Microsoft’s exclusive license to OpenAI’s intellectual property and ending Microsoft’s revenue sharing payments to OpenAI. Microsoft will also no longer be the sole cloud provider for API products built with third parties.
“While some changes seem inevitable, Microsoft appears to have made more concessions than earnings,” UBS analysts, who have a buy rating on Microsoft, wrote.
Amazon CEO Andy Jassy called Monday’s announcement “very interesting” in a post on X and added that more details would be shared on Tuesday.
Hours later, his company moved to announce a service for building AI agents with OpenAI models.
‘principal partner’
Microsoft CEO Satya Nadella (right) greets OpenAI CEO Sam Altman at the OpenAI DevDay event in San Francisco on November 6, 2023.
Justin Sullivan | Getty Images News | Getty Images
For years, developers interested in these models had to go through Microsoft’s Azure cloud or work directly with OpenAI. Now companies with large AWS investments will be able to more easily adopt models while taking advantage of bulk spending plans.
Speaking at an Amazon event, Dresser said OpenAI’s reshuffle of its deal with Microsoft was not inspired by increased collaboration with Amazon.
“Microsoft is our original partner,” he said. “They are an incredible partner for us. They will be a premier partner as we move forward. Our focus is on ensuring that we meet our customers where they are while having access to the environments in which they work. And we want to make sure that we deliver the best models in the best environments for our customers to be successful.”
Financial Times It was reported that Microsoft was considering legal action regarding OpenAI’s plans with Amazon, and Microsoft told the newspaper that it was “confident that OpenAI understands and respects the importance of complying with the standards.” [its] legal obligation.” Microsoft had no comment beyond Monday’s announcement.
Microsoft is similarly making moves to move away from OpenAI.
In September, Microsoft said it had begun leveraging an AI model from Anthropic to answer some queries in its 365 Copilot assistant for business customers. Two months later, Microsoft agreed to invest up to $5 billion in Anthropic, which has committed to purchasing $30 billion of Azure computing capacity.
Capitalizing on the growing popularity of Anthropic’s Claude Code, Microsoft announced an offering in collaboration with Anthropic called Copilot Cowork in March.
One downside of increasing demand for Claude is that reliability suffers. The company has reported partial or major outages in 37 of the last 90 days. Amazon, an early Anthropic partner and investor, recognized this.
AWS vice president Anthony Liguori said his team, which built the Bedrock service to work with AI models, moved to OpenAI Codex as its primary development platform after relying on Claude Code and Amazon’s own Kiro tool.
The reality for all the major parties involved is that they need each other.
Capacity is so limited that OpenAI and Anthropic must work with all major cloud vendors to secure as much compute as possible. Microsoft and Amazon need simple access to all major models to serve their huge customer bases.
Even though Microsoft and OpenAI are drifting apart, Jaluria was quick to note: “Microsoft still needs OpenAI, and OpenAI still needs Microsoft.”
WRISTWATCH: Private investors don’t believe OpenAI is as valuable as it seems, CFR’s Sebastian Mallaby says
