If your company runs on Outlook, Teams and Excel, you’ve probably seen the Copilot button creep into every corner of your workday. Most people never ask which AI model sits behind it.
For Microsoft, that may be the whole point.
The AI race has been framed as a contest to build the smartest model, but the companies paying the bills increasingly care more about cost and results.
Microsoft (MSFT) is already cashing in on the boom. Revenue rose 18% to $90 billion in its fiscal fourth quarter, with Azure and other cloud services up 43%, according to its earnings release.
“We are advancing the frontier on the cost-to-outcome curve, ensuring every customer can turn tokens into business results,” CEO Satya Nadella said in the release.
Still, a nagging worry has followed the stock. Cheaper open-weight models, which anyone can download and run, could shrink what companies spend on AI and leave Microsoft with costly data centers it can’t fill.
According to a Morgan Stanley research report shared with me, that shift toward many models may end up working in Microsoft’s favor.
Morgan Stanley says multi-model AI favors Microsoft, keeps $600 target.
Morgan Stanley sees a multi-model world helping Microsoft
Analyst Adam Wood reiterated his Overweight rating and $600 price target in the Oct. 7 report, titled “Battle of the AI Stack: Built for a Multi-Model World.” The target sits about 13% above Microsoft’s $529.76 close on Oct. 7, according to FinancialContent data.
“A multi-model AI world plays to MSFT’s full-stack strengths across Azure, Fabric, Foundry, and Copilot, w/ expanding monetization opportunities across the stack and Maia + Cobalt adding further differentiation over time,” wrote Wood.
His argument rests on four layers. Azure and Microsoft’s own Maia and Cobalt chips run the models, Fabric holds a company’s data, Foundry picks which model handles each job, and Copilot delivers the answer to workers.
Related: Microsoft cutting off popular software with no extension Oct. 13
Every layer can bill separately, which means Microsoft could get paid several times for a single AI task. Wood calls the loop an “Enterprise AI Flywheel,” with the resulting usage driving more Azure consumption.
His $600 target is 25 times his fiscal 2028 earnings estimate of $24.06 a share, a slight premium to large software peers.
Open-weight models are gaining ground with businesses
The shift is already visible. Open-weight models jumped to about 65% of token volume among the 50 most popular models on the OpenRouter platform from roughly 38% a year earlier, according to Morgan Stanley’s analysis, which notes those users likely skew toward cost-conscious developers.
More Artificial Intelligence:
- Trump’s ‘Super Intelligence’ order rolls through the tech world. But will it last?
- Meta Muse gives AMD and Intel investors a reason to cheer
- Microsoft’s $665 target hinges on a new AI advantage
Wood also cited a McKinsey survey showing 63% of technology leaders already use open-weight models, often alongside closed ones. MIT and Boston University research cited in the report found the cheapest price for 2023-level frontier intelligence fell about 1,000 times in two years.
Microsoft’s own numbers point the same way. Foundry, its platform for building AI apps and agents, passed 100,000 customers and more than doubled revenue in the fiscal fourth quarter, and customers using models from multiple providers rose fivefold since the start of the year, MarketBeat reported from the earnings call.
“The company does not necessarily need to own the single winning model if it can own the infrastructure that runs the models, the enterprise context that makes them useful and the control plane through which they are consumed,” wrote Wood.
Given that backdrop, it’s easy to see why Wood is upbeat on Fabric. The data platform passed a $2 billion annual revenue run rate growing about 60%, and more than 17,000 customers now use both Fabric and Foundry, up 60% from a year earlier, according to the report.
Chief information officers back him up. In Morgan Stanley’s AlphaWise survey, 47% of CIOs named Microsoft their preferred vendor for building custom AI apps, versus 10% each for Amazon and Google, and 88% expect to use Microsoft 365 Copilot over the next year, up from 72% a year ago.
Amazon and Google are building the same AI middle layer
Microsoft isn’t alone in chasing this layer. Wood points out that Amazon Bedrock and Google’s Gemini Enterprise Agent Platform offer much the same model choice and routing, while independent gateways such as Cloudflare and OpenRouter want to sit between apps and models too.
His answer is that Microsoft wins on everything around Foundry: Azure, Fabric, its security tools and the 30 million-plus paid Copilot seats that put AI in front of workers every day.
Much will likely depend, however, on whether all that usage arrives fast enough to justify the spending. Microsoft expects capital expenditures of about $175 billion in fiscal 2027, MarketBeat reported, and Wood’s own model shows gross margin slipping to 65.7% this fiscal year from 67.9%.
In my view, that’s the real bet here. Cheaper models are great news for Microsoft’s customers, but the flywheel only pays off if falling prices per task create enough extra volume to fill all those new data centers.
Wood thinks they will, and he made the case plainly.
“Microsoft has the potential to monetize the same AI workflow at multiple points across its stack, making its opportunity less dependent on any particular model,” wrote Wood.
Related: Microsoft’s $665 target hinges on a new AI advantage
