Microsoft Ramps Up Homegrown AI Chip Push

Summary
Bloomberg Tech anchor Ed Ludlow interviews Bloomberg News reporter Matt Day regarding Microsoft’s announcement of its third-generation custom silicon, the Maia 300. The segment examines Microsoft’s strategy to deploy the inference-focused chip at multi-gigawatt scale beginning in 2027, comparing its energy efficiency to Nvidia hardware. Day explains that Microsoft plans to utilize the chips internally for products like Copilot and GitHub rather than renting them to external cloud clients.

What is shown

• [00:00] Anchor Ed Ludlow introduces Microsoft’s Maia 300 announcement and its planned deployment scale.
• [00:10] Reporter Matt Day discusses the system-on-a-chip architecture and its dedicated inference purpose.
• [00:30] Full-screen graphic titled “MICROSOFT’S NEW MAIA 300 AI CHIP” highlighting:
  • Third-generation, homegrown AI chip
  • Designed to power AI inference workloads
  • Rollout begins in 2027
• [01:06] Microsoft intraday market graphics showing stock trading near $540.84 (+1.08%).
• [01:37] Footage and product renders of Azure Maia 300 silicon packaging and liquid-cooled data center server racks.

Claims & numbers

• Ed Ludlow states that Maia 300 promises 40% more tokens per megawatt than leading GPU systems and uses HBM4 memory [00:29].
• Studio graphics state that the Maia 300 is a third-generation homegrown AI chip designed for inference, with rollout scheduled to begin in 2027 [00:29].
• Matt Day states that Microsoft has approximately 12 gigawatts of capacity across its data centers today, making multi-gigawatt Maia deployments a significant portion of its fleet [01:30].
• Matt Day reports that Microsoft announced it will not rent Maia computing power directly to third-party cloud customers, instead dedicating capacity internally to services such as Copilot and GitHub while preserving Nvidia GPUs for external clients [02:16].

Notable quotes

• [00:12] “It’s designed for AI inference, though. We’re not talking about training new models. This is about running existing ones efficiently.” — Matt Day
• [00:29] “Maia 300 promises 40% more tokens per megawatt than what Microsoft said was leading GPU systems...” — Ed Ludlow
• [02:16] “...they are not going to rent the computing power of their chips to third parties. They think there is enough internal demand for things like Copilot, for things like GitHub...” — Matt Day

Assessment
This is a broadcast news report and analysis segment rather than an official hardware benchmark demo. The discussion relays specifications and strategy details released by Microsoft, illustrated with studio graphics and B-roll imagery of server hardware without presenting independent live benchmarks.

Described by gemini-3.8-flash on 2026-10-12 from the video’s audio and frames.