AI companies are spending billions on data centers in a frantic race toward artificial general intelligence (AGI). But Arvind Krishna, CEO of IBM, has a different analytical perspective on the math behind these massive bets, believing there is likely no way for these companies to achieve a rewarding return on investment at current costs.
Article contents:
- Introduction
- Complex trillion-dollar math
- The chip depreciation issue
- Skepticism regarding AGI
- Diverging views with Altman
- Future outlook
- FAQs
Introduction
Spending on data centers is growing at a crazy rate. During recent earnings calls for major companies, words like “capacity” and “AI infrastructure” have been repeated significantly. Google even announced future ambitions to build data centers in space. But the core question remains: will the revenue generated from data centers justify all this massive capital expenditure? In a recent podcast interview, Krishna concluded with a pessimistic assessment that the current economic model may be unsustainable.
Complex trillion-dollar math
Based on today’s costs, Krishna noted that it takes about $80 billion to fill a one-gigawatt data center. He explained: “Well, that’s today’s number. So if you’re going to commit to 20 to 30 gigawatts, that’s a single company, and that means $1.5 trillion in capital expenditures.” Looking at total global commitments in the pursuit of AGI, estimated at around 100 gigawatts, the cost reaches about $8 trillion. Krishna commented: “My view is that there is no way to get a return on that, because $8 trillion in CapEx means you need roughly $800 billion in profit just to pay the interest.”
The chip depreciation issue
Krishna also pointed to another critical factor: the decline in value of AI chips. He said: “You have to fully use them within five years because at that point, you have to throw them away and refill the center.” This rapid consumption of assets means the investment is not a one-time thing, but rather an expensive and continuous race to upgrade, adding tremendous pressure on profit margins.
Skepticism regarding AGI
Krishna clarified that he is not convinced the current set of technologies will get us to AGI, the technological breakthrough that makes AI capable of completing complex tasks better than humans. He estimated the chances of achieving this without an additional technological breakthrough at a very slim 0 to 1%. This view aligns with other leaders who have questioned the current frenzy surrounding AGI, considering that simply scaling existing models will not be enough.
Diverging views with Altman
When it was pointed out that Sam Altman, CEO of OpenAI, believes his company can achieve a return on its massive capital expenditures, Krishna said he disagrees with him. “That’s a belief,” Krishna said. “That’s what some people like to chase. I understand that from their perspective, but that is different from agreeing with them.”
Future outlook
Despite his economic doubts, Krishna, who has led IBM for years, praised the current suite of AI tools. He said: “I think it will unlock trillions of dollars of enterprise productivity.” However, he believes that true access to AGI will require new technologies that go beyond the current path of large language models, perhaps by combining hard knowledge with language models.
FAQs
Question: Why is the IBM CEO skeptical about data center profits?
Answer: Due to the massive cost of construction (trillions of dollars) and the fast obsolescence and wear of electronic chips (about 5 years), making capital recovery very difficult.
Question: What is the cost of building a 1-gigawatt data center?
Answer: Krishna estimates the cost to be around $80 billion at current prices.
Question: Does he believe we will reach AGI soon?
Answer: No, he sees the chance of that happening with current technologies as extremely slim (0-1%), and that we need new scientific breakthroughs.
