On June 2nd local time,6th month2, at Microsoft’s Build developer conference in San Francisco, Microsoft announced a strategic partnership with NVIDIA, rolling out a unified tech stack to deploy Agentic AI across every environment—from Windows devices to the cloud and local infrastructure.
That same day, Microsoft debuted its first dedicated AI reasoning model, Mai-Thinking-1. Microsoft’s head of AI noted that this mid-sized reasoning engine boasts 350 billion active parameters. It’s engineered to deliver peak performance and efficiency while significantly driving down token consumption costs.
Currently, MAI-Thinking-1 is live for private preview via Microsoft’s Foundry platform. Foundry is Microsoft’s developer hub built to help enterprises bolt AI models directly into their applications. Clients can submit requests to join the testing pool today, getting early access before the official public launch. Microsoft emphasized that users can also pipe their own proprietary datasets into MAI-Thinking-1 to sharpen its reasoning accuracy.
During the event, Microsoft also showcased its inaugural code generation model —— MAI-Code-1-Flash . Capable of auto-generating application and website source code based on user-submitted natural language prompts. Right now, this model is fully integrated into GitHub Copilot AI coding assistants and the Visual Studio Code editor.
Tailored for next-gen AI hardware, Microsoft also unveiled the “Project Solara” platform. On this setup, chips run AI Agent software rather than traditional apps, communicating directly with cloud data centers. The project features a series of prototype devices ranging from smart-speaker size down to employee-badge dimensions, utilizing chip solutions from Qualcomm and MediaTek. Live demos showed a stark contrast to smartphones: these gadgets run on AI agents, offloading specific tasks to cloud systems without needing a conventional OS or app layer. In healthcare, for instance, the device automatically logs clinician-patient interactions and drafts clinical documentation.
Beyond its fame for Windows and office software, Microsoft also launched a novel AI assistant named Scout. Scout proactively filters through emails and messages, automatically organizing actionable items that require user decision-making, ultimately helping teams streamline their workflow.
Microsoft also introduced an upgraded quantum computing chip, the Majorana 2 generation ( Majorana 2 ). The core breakthrough of this new chip lies in: qubit coherence time breaking past 20 seconds, and the qubit count jumping from 8 in the previous generation to 12 . Leveraging the rapid progress made with the Majorana 2 chip, Microsoft plans to develop a scalable, practical quantum computer by 2029 .
Microsoft CEO Satya · Nadella summed up the vision on stage:“This marks a serious paradigm shift. We firmly believe companies shouldn’t just passively consume frontier models—they need to actively build and shape the entire ecosystem around them.”
Industry watchers note that Microsoft is clearly trying to grab more control over the AI value chain and compete head-on with rivals in the custom model space. By giving developers direct access to these self-built models, Microsoft ensures everything runs natively on its own Azure cloud backbone—sidestepping hefty third-party licensing fees tied to players like OpenAI .
In fact, this isn’t happening in a vacuum. Google rolled out its own task-executing Gemini 3.5 Flash model back in May , deploying it straight across its internal data centers to keep compute costs and control in-house.
To put things in perspective, Microsoft has already poured 130 billion US dollars into OpenAI and another 50 billion US dollars into Anthropic , while simultaneously offering both companies’ models to enterprise customers over Azure .
On May 11th local time, Michael · Wetter, Microsoft’s executive overseeing commercial transactions, confirmed the company has staked over 100 billion US dollars into the OpenAI partnership. That figure covers the initial equity stake, plus the massive capex required to build out infrastructure and provide compute hosting for OpenAI . Wetter noted this represents the cumulative spend through the current fiscal year (wrapping up in 2026 June). He pointed out that a huge chunk of this cash burned well before any revenue came knocking:“We had to stand up the entire Azure backbone before we could even start delivering services to OpenAI ,”
On the flip side, leaked internal financials from OpenAI show Microsoft raked in roughly 300 billion US dollars from OpenAI -related services between 2023 and 2025 .
Both OpenAI and Anthropic are currently scaling fast and eyeing the public markets. Anthropic quietly filed its IPO paperwork earlier this week, while OpenAI is steadily pushing forward with its own listing plans, potentially going public as soon as later this year.
Backing this up, Microsoft reported its Q3 earnings on April 29th local time. Revenue hit 82.9 billion US dollars (+18% YoY), operating profit climbed to 38.4 billion US dollars (+20% YoY), and GAAP net income landed at 31.8 billion US dollars (+23% YoY). Excluding OpenAI investment impacts (non-GAAP), net income sat at 317.9 billion US dollars (+20% YoY).Meanwhile, quarterly capex surged to $31.9 billion, with Microsoft forecasting full-year 2026 capital expenditure to reach $190 billion—a staggering ~137% jump compared to 2025 .
