Microsoft builds its own Artificial Intelligence stack

Microsoft introduced in-house Artificial Intelligence models and a new quantum chip as it works to reduce reliance on OpenAI. The move is positioned as a way to lower costs, improve margins, and strengthen its enterprise technology strategy.

Microsoft was one of the first big tech companies to go all in on Artificial Intelligence through an early investment in OpenAI and a partnership with the large language model maker. Access to OpenAI’s models helped draw customers to Azure cloud computing and power Microsoft’s Artificial Intelligence assistant co-pilots, but it also left the company relying on technology it did not fully control. Microsoft still owns about 27% of OpenAI and has privileged access to its models and an exclusive license to its intellectual property through 2032, but OpenAI is now working with other cloud providers, including Amazon and Alphabet.

Looking to reduce its dependence on OpenAI, Microsoft unveiled seven new in-house Artificial Intelligence models at its Build developer conference. It said its flagship MAI-Thinking-1 model was made from scratch with no distillation from other Artificial Intelligence models. Distillation is when an Artificial Intelligence model is trained to act and reason like an already established, more complex model. Using McKinsey benchmarks, Microsoft said it was able to outperform OpenAI’s GPT 5-5 while delivering 10 times better cost efficiency. The company also introduced models aimed at coding, image, voice, and transcription, among other areas.

With its own models, Microsoft should be able to reduce costs by lowering the amount it pays third-party Artificial Intelligence model companies like OpenAI. This, in turn, should help improve margins and boost profits. In addition to introducing its new Artificial Intelligence models, the company made it known that it’s not forgetting about the race in quantum computing. At the event, it introduced its new topological quantum chip, Majorana 2. It said the new chip was developed with the help of agentic Artificial Intelligence and claimed that the chip’s qubits are 1,000 times more stable than those of its earlier generations. It is looking to have a scalable quantum computer by 2029.

Microsoft’s stock has lagged over the past year as investors have fretted over the software-as-a-service model, aggressive spending on Artificial Intelligence infrastructure, and the lack of its own top-tier Artificial Intelligence models or chips. The company remains well-positioned because of its deep presence in the enterprise market. More seriously entering the Artificial Intelligence race is presented as a positive step as Microsoft shifts from being mainly an Artificial Intelligence user toward becoming an Artificial Intelligence creator.

78

Impact Score

Climate case tests Artificial Intelligence discovery in court

A federal judge paused an order requiring the Conservation Law Foundation to produce an expert witness’s generative Artificial Intelligence prompts in its climate lawsuit against Shell. The dispute could shape how courts treat Artificial Intelligence data in expert discovery.

Trump signs Artificial Intelligence cybersecurity order

The White House moved to strengthen Artificial Intelligence-enabled cybersecurity across government systems, with specific protections for rural hospitals. Health Catalyst also agreed to divest Vitalware, while digital health companies announced food access, workforce and cybersecurity initiatives.

Getty ruling leaves Artificial Intelligence training law unresolved

A UK High Court decision narrows one route for copyright claims against generative Artificial Intelligence models but leaves the legality of training on copyrighted works unanswered. Stability Artificial Intelligence avoided broad copyright liability, while a limited trade mark finding keeps output risk in focus.

Nvidia unveils physical Artificial Intelligence research tools

Nvidia is rolling out physical Artificial Intelligence research tools, agent workflows and open source models aimed at autonomous vehicles, robots and vision systems. The releases emphasize virtual training and testing before real-world deployment.

Contact Us

Got questions? Use the form to contact us.

Contact Form

Clicking next sends a verification code to your email. After verifying, you can enter your message.