Dark Compute
Dark Compute refers to computing power in data centers that has been installed and paid for but is not being used. The term has become important because companies are pouring billions into chips for artificial intelligence, and part of that capacity remains unused.
Large companies operate halls full of computers, so-called data centers. There, machines sit that answer user requests, deliver videos, or run AI programs. But part of these machines never works, or only rarely. They are installed, connected, and paid for, but run idle. This exact unused computing power is called Dark Compute, meaning roughly 'dark computing capacity.' The image behind it: no light burns in these racks, even though the rent keeps running.
Why unused chips cost billions
A modern AI chip costs several tens of thousands of euros, depending on the model. Data centers buy tens of thousands of them at once. If a quarter of these chips never really work, that is no small loss. It quickly adds up to amounts in the billions that generate no return.
On top of that come ongoing costs. Even a server sitting idle draws power and needs to be cooled. The space in the hall is rented, the network connections are paid for. Economically, Dark Compute behaves like a car that you lease, insure, and leave sitting in the garage.
For investors, this is an important metric. If a company announces it will invest twenty billion in chips, the crucial question is: how much of that will actually be utilized? Analysts therefore look closely at utilization, not just at the investment sum. A high rate of expansion without matching demand is a warning sign.
How idle capacity arises
One main reason is planning ahead of need. Data centers are planned years in advance, because construction, power connections, and chip deliveries take a long time. Companies therefore order capacity for demand they merely expect. If demand falls short of the forecast, the hardware sits unused.
A second reason is technical in nature. For many chips to work together on one task, they need very fast connections and matching software. If either is missing, the chips run on their own and wait for data. Power limits also play a role: some halls are fully equipped but are not allowed to run all devices at the same time.
A third reason is reservation. Large customers rent capacity on a fixed basis to be sure it will be available in an emergency. This reserve is then unavailable to others, but is itself barely used. It’s important to distinguish here: reserve deliberately kept free for outages is normal and sensible. One speaks of Dark Compute mainly when capacity lies idle unplanned and permanently.
The term in quarterly reports and cloud offerings
The term is most often read in reports about cloud providers. Cloud means renting computing power instead of buying your own computers. When Amazon, Microsoft, or Google present their quarterly figures, journalists regularly ask about the utilization of the new AI data centers. Answers to this often move the stock price more strongly than revenue itself.
The effect is also encountered indirectly in everyday life. Because providers would rather sell unused capacity cheaply than not at all, there are heavily discounted offers. In cloud services these are often called spot instances: you get computing time cheaply, but have to give it up if a paying customer needs it. This is exactly how Dark Compute turns back into usable capacity.
A common misconception is that Dark Compute is simply a planning error. In a growing market, some overcapacity is normal, because no one can predict demand exactly. It only becomes critical when the idleness persists for years. Then it is a sign that AI expansion is running faster than actual usage.