The number grabs attention immediately: almost half of the office space that US tech companies are currently pursuing is now in the hands – or rather, the expansion plans – of artificial intelligence firms. This is not a widespread trend; the data, emerging from real-estate analytics platform VTS, concentrates on a handful of streets. San Francisco, New York, a few corridors in Seattle and Austin. The effect is a marked densification, reshaping tech geography far more sharply than previous software booms did.

The phenomenon speaks to something much larger than a real-estate bubble. When an AI company hunts for office space, it’s not merely placing desks. The hardware needed to train and serve LLMs – high-density GPUs, cooling systems, low-latency connections – demands redesigned physical spaces, often more akin to light data centers than open-plan offices. The concentration in a few blocks is no accident: proximity to engineering talent, access to network backbones, and the ability to iteratively build on-premise clusters all reward those who secure a specific address.

The paradox of the physical cloud

For years, the narrative held that the cloud would render location irrelevant. Generative AI is reversing that trend: the cost and latency of large-scale inference push many organizations to rethink deployment. This isn’t just about TCO. Controlling the metal enables better management of model versions, quantization, and fine-tuning without depending on someone else’s GPU quotas. The roads to on-premise development increasingly pass through these streets.

Those who arrive first in those neighborhoods gain a soft but real competitive advantage: the ability to iterate on models without logistic bottlenecks. Unsurprisingly, companies not born as pure AI players are also seeking space in the same blocks, creating a network effect that draws further investment. The flip side is pressure on real-estate costs and potential systemic fragility, if the entire ecosystem plays out over just a few square miles.

For those evaluating on-premise deployment, AI-RADAR provides analytical frameworks that help map trade-offs among geographic location, infrastructure, and data sovereignty, without prescribing single choices but illuminating the variables that matter.

The VTS report is not just a euphoria indicator: it shows how AI infrastructure is literally cementing a new hierarchy among cities, streets, and even individual buildings – and doing so at a speed rarely seen in commercial real estate.