Microsoft is weighing a strategic shift to open-weight AI models as the software giant seeks to reduce its dependence on OpenAI while Chinese competitors like Moonshot AI's Kimi K3 erode the cost advantage of proprietary systems.
Back
Microsoft considers open-weight AI models after Azure surges 40%
Microsoft Corp. is considering releasing some of its internally developed MAI artificial intelligence models as open-weight, a strategic pivot that would reduce the company's dependence on OpenAI and respond to a wave of capable, low-cost Chinese alternatives that are reshaping the global AI market.
"We are very pragmatic. Our core task is to deliver the best performance for customers," Mustafa Suleiman, Microsoft's chief executive of AI, said in an interview following the company's Q4 FY2026 earnings report. Suleiman, who co-founded DeepMind and later Inflection AI before joining Microsoft in 2024, said the company is "building more agent-like software," a shift that could put Microsoft in more direct competition with its own model suppliers. "First-party models will create independent capabilities for us over the long term. We cannot rely on third-party suppliers. That is not sustainable for a multi-trillion-dollar company."
The potential move comes as Microsoft reported Azure revenue growth of more than 40% in the fiscal fourth quarter, with capital expenditure coming in below analyst expectations. Shares surged more than 15% in intraday trading on July 30, adding roughly $450 billion to the company's market capitalization. Microsoft has invested in OpenAI since 2019 and deepened the partnership over successive funding rounds, but the two companies have increasingly competed in overlapping product categories.
The open-weight debate has intensified since the release of Moonshot AI's Kimi K3, a Chinese model with 2.8 trillion parameters that rivals systems from OpenAI and Anthropic at a fraction of the cost. Moonshot, founded in 2023 by Tsinghua and Carnegie Mellon graduate Yang Zhilin, paused new sign-ups after demand surged. Unlike most US frontier labs, Chinese AI companies have broadly embraced open-weight distribution — releasing trained model parameters for developers to download, customize, and run locally. The approach has helped Chinese models gain adoption among price-sensitive developers and enterprises worldwide.
Why open-weight models are reshaping the competitive landscape
Open-weight models give developers far greater control than proprietary systems, allowing them to inspect how the AI functions, run it on their own infrastructure, and build products without depending on a single provider. They are also typically cheaper. For Chinese labs, open weights serve as a distribution strategy to reach developers quickly and build an ecosystem, according to Poe Zhao, a China tech analyst. US companies have generally kept their frontier models closed, though the calculus may be shifting.
A coalition of 25 technology companies — including Microsoft, Nvidia Corp., Meta Platforms Inc., and International Business Machines Corp. — recently signed an open letter urging policymakers to avoid premature restrictions on open-weight AI. "Software developed by the open-source community now supports most of the internet," the letter said. Nvidia CEO Jensen Huang made his first-ever post on X to endorse the initiative, arguing that open models expand competition and speed AI adoption.
Anthropic CEO Dario Amodei has not signed the letter, citing safety concerns. In a blog post, Amodei said he has "never advocated" banning open-weight models but proposed stricter export controls on advanced chips to authoritarian governments, a crackdown on "industrial-scale distillation" — where AI models are trained using outputs from more advanced systems — and mandatory safety testing for sufficiently capable models. Anthropic has accused both Moonshot and DeepSeek of illicitly extracting capabilities from its Claude model.
What the shift means for investors
Microsoft's potential pivot carries significant implications for the AI investment landscape. The company's MAI model family, developed under Suleiman, has so far been proprietary, though Microsoft previously released open-weight versions of its smaller Phi models. A broader shift would put Microsoft in direct competition with OpenAI while potentially accelerating enterprise AI adoption through lower-cost, customizable models.
Suleiman dismissed concerns that high AI service costs could weaken future demand, arguing that the industry is at the beginning of a major inflection point. "People severely underestimate the fact that we are at the start of a huge turning point — models will be consumed at massive scale across all industries and verticals, and inference demand will explode," he said.
Microsoft shares, trading at roughly 35 times forward earnings, have gained more than 30% year to date. The company's Azure business, which generated more than $70 billion in revenue over the past four quarters, remains the primary vehicle for its AI monetization. If Microsoft follows through on open-weight releases, the move could pressure proprietary model providers — particularly OpenAI and Anthropic — while expanding the total addressable market for AI infrastructure that companies like Nvidia supply.
This article is for informational purposes only and does not constitute investment advice.
[1] AI战略或将转向，报道：中国AI崛起下，微软考虑支持"开放权重"AI模型[2] What Chinese AI model Kimi’s success says about the next phase of US-China AI race[3] Anthropic CEO Dario Amodei Says He 'Never Advocated' Open-Weight AI Ban, Calls...[4] The AI giants’ new problem: open AI[5] Jensen Huang Just Revealed Nvidia’s Real Endgame — And the Risk It Creates for U...
