The chip manufacturer Nvidia is unrivaled in Europe, writes WELT author Benedikt Fuest
Quelle: Jakub Porzycki/picture alliance/NurPhoto; Benedikt Fuest
Overnight, Nvidia gained a record-breaking 160 billion euros in market value. The US group is benefiting from the exploding demand for chips that are crucial for training AI. Meanwhile, Europe is far behind when it comes to AI hardware – and now it has to respond.
Dhe US chip designer Nvidia has been claiming the proverbial role of the shovel seller in the gold rush for about ten years. The group, which once grew with hardware for gaming computers, currently builds the most important chips for the training and application of artificial intelligence and has been able to increase its stock market value significantly since 2016.
But only now does it become clear how big the gold find really is and how important the right shovels are. Nvidia announced the latest business figures, significantly exceeding the analysts’ forecasts. What’s more: CEO Jensen Huang states that demand for his chips has exploded since January. They are essential for training the latest generation of large AI models such as ChatGPTwho are currently revolutionizing the world of work.
Nvidia literally put a good 160 billion euros in after-hours trading overnight to market value. For comparison: the most valuable German DAX company SAP currently has a market capitalization of a good 140 billion euros.
The Californians build the most powerful chips for programs like ChatGPT and can call up corresponding prices for them. A single H100 computing unit costs a good 40,000 euros. It calculates 250 times faster than an ordinary server. Only those who secure enough of it can adapt their algorithms fast enough to keep up with the competition.
A single training run of a large language model currently takes several weeks and costs between two and four million euros. If you calculate faster, you can test new functions faster. So now the race for the most powerful data center has started: around 16,000 of these special chips are required Facebook pack into a supercomputer. Other large US corporations have similar plans.
Another comparison: The fastest AI supercomputer in Europe belongs to the start-up Aleph Alpha and has just 512 computing units. The figures show the problem from the perspective of the Europeans: only a few commercial US providers have sufficient computing power. A common European answer is missing, a kind of CERN for artificial intelligence. This refers to the European Organization for Nuclear Research.
The AI boom requires billions in investments in new suitable ones Hardware. Anyone who cannot keep up falls behind in the race to quickly adapt the algorithms. The large US corporations will continue to expand their initial advantage over all other IT companies.
Europe must act now
Start-ups in Europe, for example, or researchers at European universities who are unable to make such investments will be left behind in the future. You are currently looking for financially strong partners as quickly as possible. They currently have to rent computing capacity from the US giants at high prices. In extreme cases, they only have the choice between selling out and becoming irrelevant.
British Prime Minister Rishi Sunak has already realized all this. He wants to invest £800m in a UK public artificial intelligence supercomputer to give researchers in his country at least a chance to keep up. If Europe wants to be part of this revolution, similar investments in research are essential. Nvidia can be happy, because unfortunately there is no comparable chip manufacturer in Europe.
“Everything on shares” is the daily stock exchange shot from the WELT business editorial team. Every morning from 7 a.m. with our financial journalists. For stock market experts and beginners. Subscribe to the podcast at Spotify, Apple Podcast, Amazon Music and Deezer. Or directly by RSS-Feed.