The way to the stars is long. So far that it would take ages to travel even if you could move at the speed of light. The only thing that helps is a star gate. A Stargate, like in the TV series of the same name, that creates a shortcut to distant galaxies and solar systems. The software company Microsoft and its partner in artificial intelligence, Open AI, are also looking for a shortcut. In order to accelerate the complex calculations for new, increasingly powerful AI systems such as Chat-GPT, the partners plan to build up enormous computing capacity. The icing on the cake is going to be a supercomputer called “Stargate”.
In a five-stage program, the two companies want to spend a staggering sum of $115 billion to build data centers that can handle the demands posed by AI. Like the portal that specializes in tech topics The Information reports, it’s about technology that works faster than previous data centers. We are talking about supercomputers, i.e. high-performance systems that, in conjunction with specialized AI chips, are intended to deliver results more quickly. The program is scheduled to end in 2030, but the first stages are expected to be completed in 2028, including the highlight of the project, the supercomputer “Stargate”. A smaller data center is scheduled to begin operations in 2026. Stargate alone could cost up to $100 billion.
The mega project is likely to be to the liking of Open AI boss Sam Altman, even if even the 115 billion dollars does not come close to his own sky-storming ideas: Altman only caused a stir last February with the demand must invest seven trillion dollars – 7,000 billion – in the development of AI chips and manufacturing systems for them. Which even representatives of the industry considered unrealistic.
Why AI requires so much computing power
But why is AI so particularly computationally intensive? This is because data is not even “eaten” and then evaluated. Rather, the computing systems chew through the data again and again and constantly make the smallest changes to its weighting. At least this is how they often find out the best result. Or they learn to play a game whose rules they didn’t know before. All they need is a goal, such as “win the game with as many points as possible,” and algorithms that help them learn while watching and playing.
Chips that were originally developed for computer games are particularly suitable for this type of calculation. These graphics chips (GPU) consist of many small computing cores, which are not very powerful compared to the main chips (CPU), but can work in parallel to a high degree, i.e. carry out many small computing steps at the same time, not one after the other. Technology that was once used to make even the reflections in the chrome of cars or the fearsome monsters in games appear realistic is now used to accelerate AI calculations.
Nvidia, a design company that only designs chips but does not manufacture them, realized earlier than others how valuable this technology would be and continued to develop not only the corresponding hardware, but also the software. Only the combination of both takes full advantage of the advantages of GPU technology. The company, which is still led by its co-founder Jensen Huang, benefits enormously from the lead that Nvidia has. Based on its market value, it is now one of the most valuable companies in the world.
More expensive than regular data centers
But the competition has woken up. More and more companies with the necessary money for research and development are working on their own AI chips. Microsoft, for example, wants to use its own chip designs and has chosen the faltering chip giant Intel as a manufacturer.
According to reports, Microsoft’s AI data centers will be built in the USA. Because they are designed to be high-performance, they will be significantly more expensive than standard data centers. Just one of Nvidia’s latest AI accelerators, called Blackwell, costs between $30,000 and $40,000. But Microsoft also wants to use other chips, including its own developments.
The huge investment in Stargate is another indication of how hot the battle for supremacy in artificial intelligence is raging. Large companies in particular are investing gigantic sums in this technology. Apparently because they expect it to give them a competitive advantage. Artificial intelligence should also help combat climate change. However, there is also the risk that new data centers – Stargate is unlikely to be the only one – will have the opposite effect due to their enormous energy consumption and prolong the path to climate neutrality.