As AI overtakes the know-how trade, Microsoft is in a novel place. The corporate is without doubt one of the world’s largest cloud suppliers and software-as-a-service corporations, in addition to holding precious stakes in two main AI labs, OpenAI and Anthropic.
These incentives are beginning to collide as Microsoft releases blockbuster monetary outcomes. firm We just reported that we were very profitable. Income for the quarter was $90 billion and web revenue was $35.8 billion. For the fiscal yr that ended June 30, Microsoft reported income of $331.8 billion and web revenue of $133.7 billion.
And CEO Satya Nadella has no intention of letting Anthropic and OpenAI’s trajectory (which in the end expands into the functions and agent infrastructure that permits its personal buyer relationships) derail these funds.
Nadella has been advocating for corporations to make use of a number of fashions and cease counting on Frontier AI Labs for the agent harness/app layer.
Doing so is harmful, he stated, as a result of it will require corporations to share too many inner secrets and techniques with mannequin makers of questionable credibility. He is aware of his clients very nicely. Company IT departments are involved about each information breaches and vendor lock-in.
He overtly advised Wall Avenue analysts on the corporate’s quarterly convention name Wednesday that this is a chance for Microsoft to promote its homegrown fashions to clients alongside issues like brokers and AI safety, whereas promising price financial savings.
In different phrases, he is pitching Microsoft as an alternative choice to most of the premium companies that OpenAI and Anthropic are creating to develop their corporations.
When UBS analyst Karl Keilstedt particularly requested Nadella to think about the open supply vs. closed supply debate roiling the AI trade and the way Microsoft stands to profit from it, Nadella appeared perturbed.
“The aim is to empower corporations to regulate their very own future,” the CEO stated of corporations. “We’re very particular concerning the architectural type of design of the platform, which implies we’ve got to maintain the harness separate from the mannequin, which implies it may be swapped out on any mannequin at any time.”
In fact, Microsoft additionally sells a menu of harnesses (aka AI brokers) underneath the Copilot title, together with the coding agent GitHub Copilot. Presently, a lot of the AI spend goes to coding brokers.
And he used final week’s high-profile incident as proof for his personal warning.
“If you happen to have a look at the face-hugging incident, the most important factor we’ve got to be taught from it’s that we should not depend on one mannequin,” Nadella stated. “You may want a number of fashions even to repair some challenges posed by one mannequin. That is the best way you concentrate on it, proper? It isn’t topic to rejection of 1 mannequin.”
The incident concerned OpenAI’s unreleased mannequin breaking out of the sandbox and a profitable full-scale hack into Hugging Face, all with the aim of exceeding benchmarks. In an try to grasp what occurred, Hugging Face initially tried utilizing the Personal Frontier mannequin (which it has not named), but it surely refused to assist. So the corporate turned to China’s open supply mannequin Z.ai GLM 5.2 to research logs and defend its infrastructure. The incident shocked the trade a lot that even Sam Altman stated that AI improvement may must decelerate a bit.
Nadella additionally revealed that Microsoft is prepared to promote its MAI household of fashions powered by its Maya AI chip, touting it as a less expensive different.
“Each buyer desires the proper mannequin for every process primarily based on high quality, latency, price, and compliance. We provide probably the most in depth mannequin catalog within the cloud with over 11,000 fashions, together with main fashions from OpenAI, Anthropic, Mistral, and xAI, in addition to our personal MAI household,” he stated.
He added, “We’re additionally accelerating our personal mannequin improvement. Now we have introduced greater than a dozen new fashions throughout pictures, audio, transcription, coding, and safety, together with our first inference mannequin, MAI Considering One, all with cost-effective inference for enterprise use circumstances at their core. We’re co-designing these fashions with our silicon and Maya. “Working the MAI mannequin at 200 has a 40% improve in efficiency per watt.”
What about Mythos? Nadella pointed to rivals to Microsoft’s new Mythos announced Earlier this week, MAI Cyber One Flash. “It delivers higher efficiency than the a lot bigger Mythos mannequin, however when mixed with our multi-agent safety harness, it prices half as a lot,” he stated.
Certainly, the Microsoft CEO has stated that corporations ought to use a mix of OpenAI and the frontier mannequin offered by Anthropic. However his bigger message is: “Do not belief them sufficient to depend on them.”
If you happen to purchase by means of hyperlinks in our articles, we could earn a small fee. This doesn’t have an effect on editorial independence.
