IBM CEO questions the industry's massive investment in AGI: Capital expenditures are excessively high and difficult to achieve returns

IBM CEO Arvind Krishna recently expressed serious doubts about the massive capital expenditures by competitors in the pursuit of Artificial General Intelligence (AGI) during an interview. Krishna estimates that the current cost of building a 1-gigawatt data center is about $80 billion; if the entire industry deploys 100 gigawatts of computing power to achieve AGI, total capital expenditures could reach $8 trillion. He pointed out that this means companies would need to generate $800 billion in profit just to cover interest, making it fundamentally impossible to achieve a return on investment from a business logic perspective.

Additionally, Krishna believes that the probability of achieving AGI relying on currently known large language model (LLM) technology is extremely low (only 0% to 1%), unless there are fundamental technological breakthroughs in integrating knowledge with LLMs. Despite his pessimistic view on the short-term realization of AGI, he still acknowledges that generative AI will unleash trillions of dollars in productivity value for enterprises.