A large-scale buildout of AI infrastructure is underway, and it’s becoming harder to separate individual company announcements from a broader pattern. Over just the past few days, several developments, spanning the U.S., Europe, and the semiconductor ecosystem, have exemplified that the competition around AI is shifting focus to physical capacity.
In Texas, Meta has raised its planned investment in a new AI data center from $1.5 billion to $10 billion, with a target of bringing 1 gigawatt of capacity online by 2028. That kind of scale shows just how much computing power the company expects to need as it expands its AI efforts.
At the same time, in Finland, Nebius has announced plans for a 310-megawatt data center, one of the largest in Europe, as part of a broader push across the region to build out AI infrastructure despite higher energy costs and tighter grid conditions.
Alongside these construction efforts, the underlying technology stack is evolving in parallel.
NVIDIA, which already sits at the center of much of the AI hardware market, is expanding its ecosystem through partnerships like its latest deal with Marvell. The collaboration is focused on integrating custom silicon, high-speed interconnects, and networking into what NVIDIA describes as scalable “AI factory” systems, essentially the architecture that will power many of these new facilities.
Taken together, we see the AI race accelerate by watching who is securing the land, energy, hardware, and systems needed to run these models at scale, and how quickly they can do it.
Why It Matters: As AI systems become more resource-intensive, access to infrastructure is emerging as a defining constraint. That shift is influencing where companies invest, how regions compete, and what tradeoffs come with rapid expansion. Having the best models may matter less if the infrastructure to run them can’t keep up.
- Spending is rising alongside compute demands: Meta increased its planned investment from $1.5 billion to $10 billion for a single site, alongside a target of 1 gigawatt of capacity. That level of buildout matches the scale of projected demand for training and inference workloads over the next several years. A facility of this size consumes power on par with a small city, placing it firmly in the category of industrial infrastructure.
- The buildout is global, but conditions vary widely: Nebius is planning a 310-megawatt data center in Finland, adding to a wave of projects across France, Sweden, and Norway. These efforts are moving forward under different conditions than in the U.S., including higher electricity prices and more constrained grid access. Those factors affect both timelines and the total cost of deployment.
- Electricity is becoming the limiting factor: In many regions, access to power is dictating whether projects move forward at all. Data centers are competing for grid capacity, and interconnection timelines can stretch for years. As a result, site selection increasingly depends on proximity to existing energy infrastructure or the ability to develop new generation.
- Control of the AI stack is consolidating: NVIDIA is expanding its role beyond chips through partnerships like its deal with Marvell. These efforts center on tightly integrated systems that combine compute, networking, and storage within a single architecture. As more infrastructure is built around these systems, dependence on a smaller set of providers increases.
- Expansion is creating local pressure points: Large data center projects bring construction activity and long-term investment, but they also increase demand for water, electricity, and land. In some areas, this has already led to concerns about resource allocation and grid strain. Local response and regulatory oversight are becoming more relevant as additional projects are announced.
Go Deeper -> How Big Tech’s New Data Centers Are Reshaping the Midwest – Inc.
Meta boosts investment in West Texas AI data center by over sixfold to $10 billion – CNBC
NVIDIA AI Ecosystem Expands as Marvell Joins Forces Through NVLink Fusion – NVIDIA
