Microsoft is in a singular place as AI overtakes the tech trade. It’s one of many world’s largest cloud suppliers and software-as-a-service firms, whereas additionally holding priceless stakes within the two largest AI labs, OpenAI and Anthropic.
These incentives are beginning to conflict as Microsoft posts blockbuster monetary outcomes. The corporate just reported an extremely profitable quarter with $90 billion in income and internet revenue of $35.8 billion. For the fiscal 12 months, which ended June 30, Microsoft reported $331.8 billion in income with a internet revenue of $133.7 billion for the 12 months.
And CEO Satya Nadella just isn’t about to let the trajectory of Anthropic and OpenAI — that are increasing into purposes and agentic infrastructure that would finally allow them to personal buyer relationships — derail that form of money.
Nadella has been preaching to enterprises to make use of a number of fashions and to cease counting on the frontier AI labs for the agentic harness/app layer.
Doing so is harmful, he’s been saying, as a result of it requires firms to share too a lot of their inside secrets and techniques with mannequin makers of doubtful trustworthiness. He is aware of his clients. Enterprise IT fears each information leaks and being locked right into a vendor.
Now he has overtly instructed Wall Road analysts in the course of the firm’s quarterly convention name Wednesday that this is a chance for Microsoft to promote clients its personal homegrown fashions, alongside brokers, AI safety and extra, whereas promising decrease prices.
In different phrases, he’s pitching Microsoft as an alternative choice to lots of the upscale providers that OpenAI and Anthropic are growing for their very own progress.
When UBS analyst Karl Keirstead particularly requested Nadella to weigh in on the open vs. closed-sourced debate roiling the AI trade, and the way Microsoft will profit from it, Nadella got here out swinging.
“The purpose is to have the agency be in command of their very own future,” the CEO mentioned of enterprises. “We’re very, very clear in regards to the architectural type of design of the platform, which is you bought to maintain your harness separate from the mannequin … which means any mannequin at any given time is swappable.”
Microsoft, after all, sells a menu of harnesses (aka AI brokers), too, below the Copilot identify, together with its coding agent GitHub Copilot. Coding brokers are the place a lot of the AI {dollars} are being spent at this time.
And he used the high-profile incident from final week as proof of his warnings.
“For those who look even on the Hugging Face incident, the most important factor that we should always take away from that’s you’ll be able to’t type of rely upon anyone mannequin,” Nadella mentioned. “You’ll perhaps want a number of fashions to even remediate some challenges that get attributable to one mannequin. Like that’s the best way to consider it, proper? Which is you’ll be able to’t be topic to a refusal of 1 mannequin.”
The incident concerned an unreleased mannequin from OpenAI breaking out of its sandbox and efficiently mounting a full-scale hack on Hugging Face, all in pursuit of besting a benchmark. Making an attempt to know what occurred, Hugging Face at first tried to make use of a non-public frontier mannequin (which it hasn’t named) that refused to assist it. So it turned to the Chinese language open-source mannequin Z.ai GLM 5.2 to investigate logs and defend its infrastructure. The incident has so shocked the trade that even Sam Altman is now saying that perhaps AI improvement ought to decelerate a bit.
Nadella additionally made clear that Microsoft is fortunately promoting its personal homegrown fashions, the MAI household, by itself homegrown AI chips, Maya, and pitching them as cheaper alternate options.
“Each buyer needs the precise mannequin for every process based mostly on high quality, latency, value, and compliance. We provide the broadest mannequin catalog within the cloud with over 11,000 fashions, together with the leads from OpenAI, Anthropic, Mistral, xAI, in addition to our personal MAI household,” he mentioned.
He added: “We’re additionally accelerating our personal mannequin improvement. We introduced greater than a dozen new fashions throughout picture, voice, transcription, coding, safety, together with our first reasoning mannequin, MAI pondering one, all with cost-efficient inference on the core for the enterprise use instances. We’re co-designing these fashions with our silicon, and we’re seeing 40% higher efficiency per watt when working MAI fashions on Maya 200.”
As for Mythos? Nadella pointed to Microsoft’s new Mythos competitor announced earlier this week, MAI Cyber One Flash. It “achieves higher efficiency than the a lot bigger Mythos mannequin, however at half the price when mixed with our multi-agent safety harness,” he mentioned.
Certain, the Microsoft CEO says that enterprises ought to use the frontier fashions that OpenAI and Anthropic supply of their combine. However his greater message is: don’t belief them sufficient to depend on them.
Once you buy by hyperlinks in our articles, we could earn a small fee. This doesn’t have an effect on our editorial independence.
