Sweet Dreams of Data Centers for Clippy Version 2: The Agentic Operation System
November 13, 2025
Another short essay from a real and still-alive dinobaby. If you see an image, we used AI. The dinobaby is not an artist like Grandma Moses.
If you have good musical recall, I want you to call up the tune for “Sweet Dreams (Are Made of This) by the Eurythmics. Okay, with that sound track buzzing through your musical memory, put it on loop. I want to point you to two write ups about Microsoft’s plans for a global agentic operating system and its infrastructure. From hiring social media influencers to hitting the podcast circuit, Microsoft is singing its own songs to its often reluctant faithful. Let’s turn down “Sweet Dream” and crank up the MSFT chart climbers.
Trans-continental AI infrastructure. Will these be silent, reduce pollution, and improve the life of kids who live near the facilities? Of course, because some mommies will say, “Just concentrate and put in your ear plugs. I am not telling you again.” Thanks, Venice. Good enough after four tries. Par for the AI course.
The first write up is by the tantalizingly named consulting firm doing business as SemiAnalysis. I smile everything I think about how some of my British friends laugh when they see a reference to a semi-truck. One quipped, “What you don’t have complete trucks in the US?” That same individual would probably say in response to the company name SemiAnalysis, “What you don’t have a complete analysis in the US?” I have no answer to either question, but “SemiAnalysis” does strike me as more amusing a moniker than Booz, Allen, McKinsey, or Bain.
You can find a 5000 word plus segment of a report with the remarkable title “Microsoft’s AI Strategy Deconstructed – From Energy to Tokens” online. To get the complete report, presumably not the semi report, one must subscribe. Thus, the document is content marketing, but I want to highlight three aspects of the MBA-infused write up. These reflect my biases, so if you are not into dinobaby think, click away, gentle reader.
The title “Microsoft’s AI Strategy Deconstructed” is a rah rah rah for Microsoft. I noted:
- Microsoft was first, now its is fifth, and it will be number one. The idea is that the inventor of Bob and Clippy was the first out of the gate with “AI is the future.” It stands fifth in terms of one survey’s ranking of usage. This “Microsoft’s AI Strategy Deconstructed” asserts that it is going to be a big winner. My standard comment to this blending of random data points and some brown nosing is, “Really?”
- Microsoft is building or at least promising to build lots of AI infrastructure. The write up does not address the very interesting challenge of providing power at a manageable cost to these large facilities. Aerial photos of some of the proposed data centers look quite a bit like airport runways stuffed with bland buildings filled with large numbers of computing devices. But power? A problem looming it seems.
- The write up does not pay much attention to the Google. I think that’s a mistake. From data centers in boxes to plans to put these puppies in orbit, the Google has been doing infrastructure, including fiber optic, chips, and interesting investments like its interest in digital currency mining operations. But Google appears to be of little concern to the Microsoft-tilted semi analysis from SemiAnalysis. Remember, I am a dinobaby, so my views are likely to rock the young wizards who crafted this “Microsoft is going to be a Big Dog.” Yeah, but the firm did Clippy. Remember?
The second write up picks up on the same theme: Microsoft is going to do really big things. “Microsoft Is Building Datacenter Superclusters That Span Continents” explains that MSFT’s envisioned “100 Trillion Parameter Models of the Near Future Can’t Be Built in One Place” and will be sort of like buildings that are “two stories tall, use direct-to-chip liquid cooling, and consume “almost zero water.”
The write up adds:
Microsoft is famously one of the few hyperscalers that’s standardized on Nvidia’s InfiniBand network protocol over Ethernet or a proprietary data fabric like Amazon Web Service’s EFA for its high-performance compute environments. While Microsoft has no shortage of options for stitching datacenters together, distributing AI workloads without incurring bandwidth- or latency-related penalties remains a topic of interest to researchers.
The real estate broker Arvin Haddad uses the phrase “Can you spot the flaw?” Okay, let me ask, “Can you spot the flaw in Microsoft’s digital mansions?” You have five seconds. Okay. What happens if the text centric technology upon which current AI efforts are based gets superseded by [a] a technical breakthrough that renders TensorFlow approaches obsolete, expensive, and slow? or [b] China dumps its chip and LLM technology into the market as cheap or open source? My thought is that the data centers that span continents may end up like the Westfield San Francisco Centre as a home for pigeons, graffiti artists, and security guards.
Yikes.
Building for the future of AI may be like shooting at birds not in sight. Sure, a bird could fly though the pellets, but probably not if they are nesting in pond a mile away.
Net net: Microsoft is hiring influencers and shooting where ducks will be. Sounds like a plan.
Stephen E Arnold, November 13, 2025
Comments
Got something to say?
-
- Subscribe to Beyond Search
Feature archive
News archive
-
