With the rapid development of the artificial intelligence (AI) industry, a surge in data center construction has also emerged. Many tech giants have announced new investment plans, aiming to build new AI infrastructure with investments reaching hundreds of billions of dollars. However, Arvind Krishna, CEO of IBM, has raised doubts about the return on such massive investments in a recent interview.

IBM

According to reports, building a data center with one gigawatt of computing power currently costs around $8 billion, and global commitments for related computing power approach 100 gigawatts, meaning the total investment is approaching $8 trillion. Such a massive capital outlay would require profits of $800 billion just to cover interest, which is almost an unattainable goal.

Krishna emphasized that this estimate is directly related to current hardware, depreciation, and energy costs, rather than long-term projections. He pointed out that the rate at which hardware depreciates is often underestimated by investors, and these data centers usually need to replace most of their hardware every five years, further increasing the pressure on long-term capital expenditures. Recently, some investment institutions have also expressed concerns: as AI performance improves and model sizes grow, the accelerated retirement of older GPUs forces companies to replace hardware at high costs instead of simply expanding scale.

Krishna also mentioned that while it is expected that the next generation of generative AI tools will significantly boost corporate productivity, the relationship between the physical scale and economic viability of current AI infrastructure remains an issue that needs to be resolved. Companies that invest heavily in large data centers and choose to shorten their update cycles must prove that their returns are sufficient to justify unprecedented capital expenditures.

Key points:  

🌐 Whether the return on investing huge amounts of money in AI data centers is feasible has become a focus of the industry.  

💰 IBM's CEO pointed out that the cost of construction and the speed of hardware depreciation are severely underestimated.  

📉 Companies must prove that high expenses can generate enough profits to support future investments.