NederlandsData centre
A data centre is a large building full of servers: powerful computers that compute and store data day and night. AI models are trained there and run there when you ask a chatbot something.
A data centre is a building full of servers: powerful computers in racks that work day and night. They store data, serve websites and carry out computation. When you ask an AI chatbot a question, it usually goes to a data centre, where an AI model works out the answer and sends it back.
How does it work?
A data centre holds thousands of servers, connected by fast cables. For AI there are many specialised chips in them, which together do enormous amounts of computation at the same time. Two things are always important: power and cooling. Chips get hot, and a data centre needs a constant, reliable power supply.
The work in a data centre falls roughly into two parts. First an AI model is trained, which can take weeks and uses a great deal of computing power. Then the model runs for users, and that happens continuously, every second, for millions of questions. Spreading that work over different chips makes the answers faster. The company d-Matrix showed this with a mix of two kinds of chips (link).
Why it matters
As AI is used more, data centres get bigger and demand more electricity. That affects the energy sector. OpenAI, for instance, promises that its new data centres will use less power when the grid is under strain. The first, in the US state of Georgia, can hand back up to 1,000 megawatts (link). At the same time Google and Meta are adding supply: they help pay for extra power from existing nuclear plants (link).
Demand for computing power itself is growing too. On GitHub, more than five times as many code changes were saved in September as a year earlier, mostly because of AI agents. The platform therefore put in 4 million extra CPU cores, the computing cores of ordinary processors (link). Data centres are thus more than storage space: they help determine how fast AI can develop and where the power for it comes from.
Articles about Data centre
-
Start-up PSI uses AI to help publish five quantum proofs that a computer can check
Physical Superintelligence has used its own AI system to tackle five open questions in quantum information theory. Each proof is also written in Lean, a program that checks proofs step by step. Other researchers can now check them for themselves.
-
Chinese team sends 3.2 petabits per second through a single optical fibre
Chinese researchers sent 3.2 petabits per second through one optical fibre with 24 cores, and the equipment processed everything straight away. That is a record for this kind of real-time measurement and could help connect AI data centres faster in future.
-
OpenAI will now train its most powerful AI models without internet access
OpenAI no longer lets its strongest AI models onto the real internet during training. From now on they practise with a stored copy of the web. At OpenAI's request, the models themselves searched for holes in this new shielding, so the company could close them.
-
Tiny holes that slow down light make a small Japanese chip for optical fibre far more efficient
Japanese researchers built a component that turns electrical signals into light, using tiny holes that slow the light down. It is 0.3 millimetres long, runs on low voltage and sends up to 200 gigabits per second. The next step: losing less light along the way.
-
Start-up generates electricity for the first time with a power plant on the seabed, at an underwater volcano
At a depth of one and a half kilometres, Endurance Energy ran a small geothermal power plant on the heat of an underwater volcano. For a few seconds, it produced more electricity than it used itself. A stronger version follows this winter.
-
Anthropic tests Claude Duet, an AI assistant that comes to you with tasks itself
With Claude Duet, the AI assistant does not wait until you ask: it sends a summary in the morning, keeps an eye on your apps and keeps working when your laptop is off. You approve important steps yourself first. The test version has been available to paying users in the US since 21 October.
-
Nvidia lets AI agents do only what is allowed, and checks the rules in advance with mathematics
Nvidia presents OpenShell: free software that lets AI agents work in a sealed-off space. A mathematical check looks in advance at whether the rules allow more than intended, and according to Nvidia a separate chip can stop an agent within milliseconds.
-
A different order makes AI servers start answering more than three times faster
Engineers from Amazon and Mistral have made the open software vLLM, which many companies use to run AI models, work together more cleverly. With long questions, the first word of the answer appeared more than three times faster in their test. The new method is free to use.
-
Large open AI model from DeepSeek now fits on a computer with 128 gigabytes of memory
DeepSeek's new open AI model is too big for an ordinary computer. The company Unsloth made a heavily reduced version that runs on a powerful laptop or mini computer with 128 gigabytes of memory. This costs some accuracy.
-
AI model Maxwell makes existing software faster while it runs
The Israeli company doubleAI presents Maxwell, an AI that rewrites programs while they run so that they go faster, without touching the source code. According to the company, one customer's computing costs fell by 67 per cent.
-
d-Matrix splits the work of an AI model across two kinds of chips for faster answers
At a major data centre conference, the American chip company d-Matrix showed how an AI model answers faster when ordinary graphics chips and its own chips each do the work they are good at. The company now also supplies complete computing clusters.
-
Chinese company Sugon shows computing cluster with 100,000 home-grown AI chips
China's Sugon is showing its Sugon 8000: a computing cluster that links 100,000 computing cards made in China. The system combines AI with classic supercomputing and, according to Sugon, has already run ten research projects.
-
Stanford builds a grid of more than 600 light traps to link quantum computers made of atoms
Physicists at Stanford built a grid of more than 600 tiny light traps, each of which can catch the light of a single atom. That helps to read out quantum computers made of atoms faster and later to link them together.
-
Meta has a subsea cable built with two light channels per glass fibre: twice as much data across the ocean
Meta is having a subsea cable laid between the United States and France that can carry 1 petabit per second, according to Meta twice as much as the best cables today. The secret: each glass fibre has two light channels instead of one. The cable is due to work in 2029.
-
Working AI server rack lets chips talk through light, with the optics right next to the chip
At a major data centre trade show in the US, three tech companies showed a working server rack in which chips talk through light, with the light components right next to the computing chip. That could make AI data centres much more energy-efficient. Large-scale production is next.
