Who Owns The Plug? #3 - Holger Mueller, Constellation Research Group

By Raju Vegesna, Chief Evangelist, Zoho

In this episode of "Who Owns the Plug?", I sit down with Constellation Research analyst Holger Mueller to unpack what's actually changing in the sovereignty conversation, from Amazon's massive sovereign cloud investment in Germany to why "data residency" laws are already outdated in the age of AI.

Sovereignty is now a spectrum, and it's the AI era that's forcing that realization. Data-residency laws feel outdated once you treat data as a "book" — a country can require the book stay within its borders, but if a model reads it and the knowledge (the intelligence extracted from it) leaves, sovereignty has already been lost even though the data never moved. Holger adds a legal dimension: without judicial recourse against foreign actors, and with AI's black-box nature making harm nearly untraceable, sovereignty gaps become effectively unenforceable.

Underneath this sits a familiar tension: a handful of private companies now control the digital "backbone" — networks, cloud, and models — the same way a handful of firms once controlled physical trade routes, echoing the discomfort nations felt with the East India Company. That imbalance, paired with "data gravity" and now "GPU gravity" (fast infrastructure attracting more workloads, since a well-resourced agent can simply out-negotiate a slower one), is pulling the industry back toward on-premise and edge deployments — mirroring the historical swing from mainframes to PCs to cloud and back. It connects to a deglobalization thesis: just as cheap shipping once pulled physical production toward wherever labor was cheapest, AI is now making digital "creation" itself cheap, which could pull production back closer to where the demand actually is.

We both agree the technology is still immature relative to the capital pouring into it — comparing it to a gold rush, with today's compute investment less durable than historical infrastructure bets like fiber optic, since chips depreciate fast as next-gen hardware arrives annually. The practical center of gravity for enterprise AI right now, they agree, is narrower than the hype suggests: it works best where outputs can be verified quickly (like code), and agents are becoming "the new office suite" rather than a wholesale replacement for infrastructure or workers. This is as an early, unevenly distributed phase that will likely take a decade or more to fully play out, with genuine efficiency gains (in tokens, energy, and cost) still needed before the investment thesis holds.

The full discussion can be viewed below: