With the Surface Ultra and Surface RTX Spark announcements at Build, I saw a Microsoft returning to its development and server roots. Finally, the company that brought us IIS, SQL Server, Windows Server, Azure, VS Code, Visual Studio, and more will finally “get serving AI locally on the edge right.”
Yet the Surface videos and Build announcements gave me pause. For all the pomp and circumstance over the hardware, the software details are scant. Here’s what we see:
- Windows Subsystem for Linux
- vLLM
- VS Code
What here projects any of Microsoft’s strengths?
This is the company that fought tooth and nail to prove they were the best server and desktop operating system combo. The best integrated IT solution for your company. The best hosting for your organization. The go-to software for productivity.
And now they say they’re offering the best solution with what? Linux and AI server software they didn’t write. Heck, they don’t even appear to be loudly committing to contributing to solutions to optimize them for Windows. Remember when they did that after open-sourcing .NET?
I’m hoping this is all a rouse — Microsoft must be working on something they can’t yet announce. Something they’ll announce with the Surface releases this Fall. At least, I hope so. They could finally be the company to “do AI right.”
For most developers, fumbling around with Ollama and LM Studio and hoping they work at least 90% of the time, is tough. We don’t want to also be IT managers managing vLLM instances. We just want AI to work, 100% of the time. Five 9’s for AI, if you will.
Argue about my being lazy as much as you want — “You just don’t like Linux” or “You just don’t understand”. No — I just want it to run. Easily selecting a model, with an AI server solution I don’t have to think much about. Even supporting an “Auto” mode, choosing the local model I need, and for larger operations, asking if I want to move part or all of my jobs to the Cloud. That’s something Microsoft can do. It’s a strength.
It could be woven into their other solutions. Native support and understanding of what Local vs. Cloud vs. “Trusted Cloud” for LLM servers from an IT and data privacy management perspective. End-to-end AI support, with local and cloud.
Because eventually somebody will do this. Because it’s needed. Because the integration of agentic development into the developer workflow will likely never be uncoupled. And there’s an opportunity now for Microsoft to own what they were built for.
Looking forward to the Surface releases to see if I got this one right. ☕
Auri's Blog 