IBM CEO: With current costs, investing trillions to build AI data centers is "impossible to make a profit"

IBM CEO Arvind Krishna issued a stern financial warning regarding the current surge in investments in artificial intelligence (AI). He believes that under the current technology and cost structure, the practice of tech companies pouring trillions of dollars into building data centers in pursuit of artificial general intelligence (AGI) is nearly impossible to yield a return on investment. Recently, Krishna expressed this view on the "Decoder" podcast, directly challenging the prevalent industry strategy of "expanding computing power at all costs," which is driving huge demand for AI chips from companies like NVIDIA. He stated clearly, "My point is, you cannot get a return from this." He calculated based on a simple financial model that the global commitment to computing power in pursuit of AGI seems to have reached 100 gigawatts. At today's construction cost of about $8 billion per gigawatt for data centers, total capital expenditure could reach an astonishing $8 trillion. He emphasized that such a scale of investment "means you need about $800 billion in profit to pay the interest," which he believes is an unattainable profit target. He also added a key depreciation factor: "You have to use up all the equipment within five years because by then, you have to throw it away and refill." This viewpoint aligns with investor Michael Burry's recent concerns about depreciation regarding NVIDIA. This cautious assessment stands in stark contrast to companies like Meta and Google, which are continuously ramping up investments in AI infrastructure, as well as OpenAI CEO Sam Altman's call for a significant increase in energy supply to support AI development. Krishna's analysis shifts investors' attention from the distant aspirations of AGI back to the real issues of capital expenditure, depreciation, and profitability. Skepticism About the Path to AGI: Success Probability Only "0-1%" Krishna's financial skepticism is rooted in his profound doubts about whether the current technological path can achieve AGI. When asked about OpenAI CEO Sam Altman's belief that their massive investments could yield returns, Krishna expressed a different view. "That's a belief," Krishna said, "I understand the goal they are pursuing, but that is separate from whether I agree with them." He made it clear that he does not believe the current technology can lead us to AGI and estimates that without further technological breakthroughs, the probability of success is only "0-1%." This skepticism is not an isolated case in the tech world. Salesforce's Marc Benioff, Google Brain founder Andrew Ng, and Mistral AI CEO Arthur Mensch have all expressed doubts about the AGI craze. OpenAI co-founder Ilya Sutskever has also stated that the era of simply expanding computing power is over Affirming the Current Commercial Value of AI Despite a pessimistic outlook on the return on investment for AGI, Krishna is very optimistic about the commercial application prospects of current AI technologies. He emphasizes that his view does not negate the overall value of AI. “I want to be absolutely clear that I believe it will unleash trillions of dollars in productivity in enterprises,” Krishna stated. He believes that achieving AGI requires “more technology than the current large language model (LLM) path” and suggests that integrating hard knowledge with LLMs could be a viable path for the future. However, even so, when asked if this path could achieve AGI, he still replied, “Even then, I would only say ‘maybe.’”