IBM CEO Considers Current AI Frenzy Unsustainable
The arms race between hyperscalers is consuming trillions of US dollars. IBM CEO Arvind Krishna questions how economically viable this is.
(Image: Gorodenkoff/Shutterstock.com)
The spending on AI data centers is disproportionate to the expected profit, summarizes IBM CEO Arvind Krishna in a podcast by The Verge. So-called hyperscalers like OpenAI, Meta, Amazon, Google, and xAI regularly outdo each other with announcements about the computing capacity their new data centers will deliver. New systems consume one gigawatt and more of electrical power.
Krishna urges: "So let's focus on today's costs, because everything in the future is speculative. It costs about $80 billion to fill a data center with one gigawatt of power. That's the number today. If a company wants to provide 20 to 30 gigawatts, that's $1.5 trillion in capital expenditures. [...] You have to use all of that within five years, because then you have to throw it away and refill it."
Nvidia, meanwhile, wants to introduce new, faster AI accelerators annually. After five years, according to current plans, models are outdated and also fully depreciated.
Krishna further elaborates: "When I then look at the total global investments in this area to chase AGI [Artificial General Intelligence], these announcements seem to be 100 gigawatts. That's $8 trillion in capital expenditures. In my opinion, it is impossible to achieve a return on investment with that, because $8 trillion in capital expenditures means you have to generate about $800 billion in profit just for the interest."
The AI Faith
Krishna considers statements from OpenAI CEO Sam Altman, among others, that a company will eventually develop a universal AI and take over the entire market, to be a "faith." The IBM CEO can understand this from their perspective but does not agree. "Some people will make money, some people will lose money." At least, according to him, all the built infrastructure should remain useful when the AI race ends.
Videos by heise
"I want to be clear. I am not convinced, or rather, I consider it very unlikely – we are talking about 0 to 1 percent here – that currently known technologies will lead us to AGI. That is the bigger gap for me. I find the current technologies great. I consider them incredibly useful for businesses. I believe they enable trillions of dollars in productivity gains in companies, to be clear. However, I think AGI will require more technologies than the current path [with large language models]. I believe that the fusion of knowledge with LLMs will be necessary for that."
Empfohlener redaktioneller Inhalt
Mit Ihrer Zustimmung wird hier ein externer Preisvergleich (heise Preisvergleich) geladen.
Ich bin damit einverstanden, dass mir externe Inhalte angezeigt werden. Damit können personenbezogene Daten an Drittplattformen (heise Preisvergleich) übermittelt werden. Mehr dazu in unserer Datenschutzerklärung.
(mma)
