Microsoft Azure leadership reorganization ties to Maia AI chips

The core change is a bigger job for Ulag. Under the new plan, he oversees several senior Azure leaders, including Girish Bablani, who runs the group that builds software for Microsoft’s cloud servers, Azure technology chief Mark Russinovich and security head Charlie Bell. Some units are also being merged.

Microsoft has not said how many teams or employees are affected, or when every change takes effect. It is an internal reshuffle, not a new product, but it puts databases, core infrastructure and engineering under tighter, more central control.

Azure’s top ranks have already shifted this year. In April, Eric Boyd, the executive behind the Azure AI platform, quit to run infrastructure at Anthropic, maker of the Claude chatbot.

On the day of the reshuffle, Guthrie also unveiled the Maia 300, the third generation of Microsoft’s in-house AI chip. In a blog post he said Microsoft will scale the Maia family to deployments of several gigawatts starting in 2027, making it the main system for running AI in Microsoft’s products and services, Copilot included.

Maia 300 is built for inference, the work of running an AI model once it has been trained, every time someone asks Copilot a question. Microsoft says it should produce 40% more tokens (the chunks of text an AI generates) per megawatt than leading GPU-based systems. A full rack of 96 Maia 300 chips is meant to deliver up to seven times the output of a rack of the older Maia 200.

Each chip carries 18 TB/s of peak memory bandwidth, which matters for long, complicated AI tasks, and its compute parts are made on TSMC’s N2P process. Deployment is due to start after the turn of the year.

Its predecessor, Maia 200, launched in January. It already runs some Microsoft Copilot work, OpenAI models and Microsoft’s own AI models in the US, Europe and Australia. GitHub Copilot and Microsoft Foundry are due to move onto it by the end of 2026.

Building AI data centers is extremely expensive, and a second report a day later suggested Microsoft is looking for savings. According to The Information, Bablani told the managers under him to come up with savings of 20% or more. His group has about 11,500 staff.

The savings would reportedly help pay for AI data center spending and fund higher pay to keep engineers being courted by Meta and Anthropic. Options under discussion include ending contracts with outside firms. The report did not say whether jobs would be cut.

Microsoft disputes the account. A spokesperson said Azure leadership has not told teams to draw up cost-cutting plans, adding that responsible companies prepare for many scenarios. Microsoft’s last deep cuts came in July 2025, when it laid off about 9,000 people.

If you use Copilot at work or at home, the chip shift is the part to watch. Running AI on cheaper, more efficient in-house chips could help Microsoft keep prices in check and add capacity faster, without waiting in line for scarce Nvidia GPUs. Microsoft says Maia 200 already beats leading GPUs on throughput per watt for core AI features in Teams and Copilot.

For businesses on Azure, the reorganization may mean fewer gaps between database services and core infrastructure, since one leader now oversees both. It also signals that Microsoft wants Azure lean: more output per dollar and per megawatt.

Azure is now big enough to report on its own. In September, Microsoft cut its business segments from three to two, Agents and Infra plus Devices and Consumer, and started breaking out Azure on its own. Azure took in $29.4 billion between April and June, 42% more than a year before.

Microsoft plans to share more about Maia 300 at Ignite, its annual conference for business customers, in San Francisco next month. Expect details on where the chips will run first and which Azure services move onto them.