GPU hoarding is straining AI hosting in Europe
In brief: The four largest US tech companies have snapped up billions of dollars' worth of Nvidia GPU chips, with Meta alone owning 350,000 units. Fire co's are hoarding billions worth of Nvidia GPU chips. Meta has 350K of them
Why it matters: Open-weight AI models require large amounts of GPU resources to run efficiently. When most GPU resources are concentrated among a small group of players, it becomes harder for European hosting providers — including Danish ones — to get access to the hardware they need.
Consequences for EU hosting: Open-Weight AI Models Open New Possibilities for European Hosting shows that self-hosting open weights can give full control over data — but only if the underlying infrastructure is available. GPU scarcity can slow this development and make it more expensive for European companies to run AI models locally.
What it means for Danish hosting companies: Danish providers like Liviate, who want to offer AI hosting on Danish soil, face higher hardware prices and longer delivery times. At the same time, it increases the pressure to find alternative solutions — such as optimized models or partnerships with GPU manufacturers.
Conclusion: GPU hoarding is a direct threat to the rapid spread of open-weight AI in Europe. To maintain sovereignty and data privacy, Danish companies need to consider strategic partnerships and investments in local AI hardware.
Bottom line: Liviate helps Danish companies host AI models on Danish soil.