Microsoft embraces ‘off-frontier’ AI strategy, prioritizes practicality

Microsoft is shifting its artificial intelligence (AI) approach to focus on practical applications rather than pursuing the most advanced models.

Mustafa Suleyman, CEO of Microsoft AI, outlined this ‘off-frontier’ strategy during the company’s 50th-anniversary event.

This approach emphasizes developing AI solutions that are slightly behind the cutting edge but more cost-effective and tailored to specific uses. ​

Unlike firms competing to create the world’s most powerful models, Microsoft intentionally pulls back from the cutting edge. Instead of immediately building the most advanced systems, it takes three to six months to do anything after the latest models are released.

“After you’ve let the frontier go first for the first three months, the first six months, it’s cheaper to give a specific answer then,” Suleyman said in an interview. “We call that off-frontier.”

This strategy enables the tech firm to sidestep the vast expense of being a first mover while at the same time establishing powerful, useful AI systems. It also allows the company to concentrate on creating artificial intelligence tools that the world needs rather than trying to ride the hype wave.

Suleyman said they had never attempted to win any leaderboard; instead, they focused on solving real people’s real problems.

Microsoft is adding personalized features to empower Copilot

Copilot (Microsoft’s AI assistant for Windows, Office, and other apps) is one of the best examples of the practical AI vision the firm tries to achieve.

At its anniversary event, Microsoft revealed that it will soon be adding “memory” to Copilot, allowing it to retain key details about users as time goes on. That means the assistant can provide more intelligent, customized assistance — like remembering the names of people you frequently email, how you usually format documents in Word, or which tasks you do most often in Excel.

Those features were originally introduced in OpenAI’s ChatGPT, which Microsoft deeply embeds. Now, comparable powers are coming to the firm’s products.

OpenAI released the “o1” model in September — a system focused on reasoning that takes longer to create considered answers. Then, a few weeks later, Microsoft announced “Think Deeper,” a similar capability for Copilot. Instead of racing to be first, the company is repackaging ideas that work in ways that match its ecosystem.

Copilot’s improvements prove the tech company is interested in usability rather than Flash. This is about tools that work and feel like magic — AI that helps people do their work better.

Microsoft balances partnerships with in-house development

Microsoft’s close relationship with OpenAI has spurred its speedy advancement in AI in the past two years. The firm has poured over $13 billion into OpenAI since 2023, embedded its models in Bing, Word, Teams, and Windows, and powered everything from generative tools like image generation to AI-powered search.

But that relationship is beginning to change. Microsoft targeted OpenAI in July 2024, just a few months after it targeted Google. In January 2025, OpenAI struck a significant deal with Oracle to help power the $500 billion Stargate AI project — a deal that raised the possibility that the startup was seeking partners beyond Microsoft’s Azure cloud.

Still, Suleyman said the partnership with OpenAI remained strong, stating that they had been deeply partnered with the company until at least 2030 and that OpenAI had enjoyed an enormously successful relationship with Microsoft.

It’s worth noting that the company is not resting on its laurels; it is now leaning into developing its proprietary AI capabilities, too. Suleyman added that it is mission-critical for the company to develop its own long-term AI. He said that Microsoft had “thousands” of Nvidia GPUs, critical hardware for AI training, and is also developing a separate class of smaller and open-source models that can run on machines without cloud access.

These models avoid being expensive and resource-intensive, and they can be a nice option for businesses that don’t need to own and operate AI at a massive scale but do want powerful tools.

Cryptopolitan Academy: Want to grow your money in 2025? Learn how to do it with DeFi in our upcoming webclass. Save Your Spot