The four stories on this desk share one thread: the physical distribution of computing is being redrawn. Nvidia is looking at old telco buildings for capacity and tightening its watch over where its AI chips end up, Microsoft is meeting resistance at a data center construction site in Amsterdam, and United is putting Spotify into seatbacks over Starlink. Each is a different edge of the same shift - compute is moving out of the purpose-built hyperscale campus and into inherited, dispersed, or otherwise existing infrastructure, and the politics, logistics and business models are following it there.
The reallocation of existing space
Start with the most concrete example. As Tom's Hardware reported, Nvidia wants to repurpose more than 130 old telco buildings owned by Brightspeed as small data centers, with Nvidia estimating that the facilities could produce up to 1GW of compute capacity. That is a striking figure for buildings that were never designed for this purpose, and it says something about the economics of AI infrastructure in 2026: the constraint is no longer only chips, and it is increasingly land, power and the time it takes to bring a facility online. Existing telco real estate comes with power connections, physical security and a distribution footprint already in place. The trade-off is that these are small, awkward sites, not the clean, dense halls that GPU clusters prefer. Nvidia's estimate of up to 1GW is a claim about potential, not a promise, but the direction is clear.
For US technology companies, this matters because it reframes where AI capacity comes from. If distributed, retrofitted facilities become a real part of the stack, the companies that can operate many small sites - and the vendors that supply and track them - gain leverage relative to those betting only on a handful of giant campuses. It also spreads the load, and the local opposition, across many more communities.
Tracking as an operating condition
The second story is about visibility. As Tom's Hardware reported, Nvidia is requiring Southeast Asian cloud providers seeking Nvidia Cloud Partner status to enroll in its Fleet Intelligence tracking program, giving the chipmaker greater visibility into where its AI GPUs are deployed. The context is Washington's pressure over suspected chip smuggling into China, with investigations involving Megaspeed International and Super Micro highlighting how hard export restrictions are to enforce.
The connection to distributed infrastructure is direct. The more compute is spread across small sites, partners and retrofitted buildings, the harder it is to know where a given GPU physically sits and who is using it. Tracking becomes a condition of doing business rather than an afterthought. For US companies, this cuts two ways: it lowers compliance risk for buyers who can demonstrate where their hardware is, but it also entrenches Nvidia as the party that decides who is eligible for partner status. The enforcement burden is being pushed down the supply chain, to cloud providers that never had to think of themselves as export-control intermediaries.
The friction is local
The third story shows the cost of siting. As The Verge reported, protesters broke into a construction site of a new Microsoft data center in Amsterdam, with the Geef Tegengas group helping organize opposition to three 85-meter towers on the city's outskirts, and around 100 people entering the site. This is not a story about a single permit dispute. It is the predictable result of a buildout that has to happen somewhere, and increasingly happens near people who did not ask for it.
The implications for US firms are practical rather than abstract. Every large campus carries reputational and schedule risk, and a delayed project is a costlier project. That pressure helps explain the interest in smaller, inherited sites like the telco buildings - less visible, faster to stand up, and easier to spread across jurisdictions. It also raises the value of firms that can demonstrate local legitimacy, not just capacity.
