Microsoft has purchased many graphics cards for AI services. Now it has thousands of graphics cards, but there is a lack of power and infrastructure. A problem that Microsoft shares with many other providers.
The demand for AI apps and services is leading to a significant growth in the need for graphics cards and computing chips that can meet the demand.
However, Satya Nadella, CEO of Microsoft, recently stated in an interview that this is not the biggest issue: because you also need the infrastructure and the energy to actually be able to utilize the computing capacity. A problem he currently has.
Microsoft CEO says: Computing power is not the problem, but the energy supply
Nadella said: Satya Nadella stated in an interview with the Bg2 Pod podcast with OpenAI CEO Sam Altman that while Microsoft currently has a number of AI chips physically stored, it cannot connect them because corresponding data centers, power, and cooling systems are lacking.
He made this statement on YouTube in response to a question from Brad Gerstner, the host of Bg2 Pod, on whether Nadella and Altman agree with Jensen Huang, CEO of Nvidia, who said that there is no chance of an overloading of computing capacity in the next two to three years:
I think you can’t really predict the cycles of supply and demand in this particular case, right? The point is: What does the long-term trend look like? The long-term trend is what Sam (CEO of OpenAI) has said, namely that ultimately, quite frankly, the biggest problem we currently have is not an oversupply of computing power, but the energy supply – that is, the ability to complete the builds quickly enough near the power supply.
If you can’t achieve that, you might have a lot of chips in stock that I can’t connect. That is indeed my problem today. It is not a chip supply problem, but the fact that I do not have warm shells [= pre-equipped server racks with sufficient power supply and cooling] to connect.
Many data centers face the problem of insufficient energy supply
The current crisis is industry-wide relevant: other providers of large data centers (so-called hyperscalers) are also facing similar bottlenecks and are exploring how they can reconcile the growth of AI computing capabilities with infrastructure and energy supply (via kontronn.com).
Nadella speaks of a “Power Density Crisis,” where data centers require far more power per square meter than before, which the networks and buildings cannot supply. The solution requires massive investments in power grids, cooling, and data centers, as well as technological innovations in chips to make them a bit more energy-efficient.
Nvidia earns billions with AI chips and graphics cards, which are needed for data centers. In an interview at the Hong Kong University of Science & Technology, Jensen Huang, the head of graphics card manufacturer Nvidia, explained that today’s AI systems often provide answers that are not optimal. He emphasized that users should constantly question whether an AI answer is “hallucinated” or makes sense: Nvidia earns billions with AI, now the CEO says: It will take years before we can trust AI
