Latest
Wednesday, November 18, 2026
  • New YorkNY
  • LondonLDN
  • TokyoTYO
Technology

Microsoft Unveils Customizable AI Models at Ignite Conference

Microsoft's latest initiative allows businesses to tailor its cutting-edge AI models with their own data, marking a departure from more restrictive proprietary models.

Microsoft’s new Licensed Weights lets businesses customize AI models with their own data
Source: ZDNet

Microsoft's latest move has taken center stage at its Ignite conference, where CEO Satya Nadella unveiled several new AI models from Microsoft AI (MAI), a lab within the company. Amidst the flurry of announcements, one development stands out: a game-changing initiative that allows businesses to customize these cutting-edge models with their own data.

The breakthrough lies in an initiative called Licensed Weights, which enables customers to tailor MAI's frontier models to suit their specific needs and run them securely on Azure. This approach contrasts with the more restrictive proprietary models from OpenAI and Anthropic, yet doesn't grant full access like Hugging Face's offerings do. Companies can get started by working through their Microsoft account representatives.

Microsoft AI CEO Mustafa Suleyman emphasized that customers retain control over who accesses their customized models, ensuring data security and ownership remain in their hands. This selective approach to open-weight AI has sparked interest among businesses, which will now need to partner with MAI's Frontier Company for delivery of custom models embedded within their products.

These collaborations promise a seamless integration of tailored AI capabilities into various applications, underlining Microsoft's commitment to empowering its customers through cutting-edge technology.

Businesses are increasingly seeking AI solutions that allow them to maintain control over their intellectual property. To meet this need, Microsoft has introduced Licensed Weights, a customizable AI model framework that lets companies tailor their own AI capabilities using their proprietary data.

According to Suleyman, "no one wants to pour their IQ into someone else's model." This sentiment echoes the views of Palantir CEO Alex Karp, who has repeatedly argued that businesses should be able to utilize their valuable information within their own AI models. By offering Licensed Weights, Microsoft is providing a middle ground between proprietary and open-source AI options.

The traditional narrative surrounding AI has pitted expensive, proprietary models with peak capabilities against cheaper, open-source alternatives that offer customizability but often lack performance. However, by introducing a customizable option, Microsoft may be able to bridge this gap and carve out a unique position for itself in the market.

Western research labs are now developing more competitive open-weight models, including those from Thinking Machines and Mistral. Nonetheless, Anthropic and OpenAI still hold a significant majority of the enterprise AI market share. With Licensed Weights, Microsoft is diversifying its offerings and differentiating itself from other major players in the industry.

Microsoft's new Licensed Weights offering has allowed businesses to customize AI models using their own data, setting it apart from other major players in the industry.

The company's MAI Code 2 model, unveiled at Ignite 2026 alongside Copilot Workspaces, boasts impressive performance numbers on widely used benchmarks. According to Suleyman, the model achieves 98% of the performance of GPT 6.1 Sol at just 40% of the cost, and significantly outperforms Opus 5.5 in terms of price and quality.

With Licensed Weights, Microsoft is now positioned as a reliable choice for organizations seeking a tailored blend of ROI tools, customization capabilities, security controls, and flexibility - all while keeping costs low.

Facts based on reporting originally published by ZDNet.

You may republish this story, in full or in part, if you credit News Central Site and link to it (licence CC BY 4.0). Photos are not included.