IBM’s CEO Says GenAI Is Great For Enterprise, But It Will Not Be AGI

December 8, 2025 Timothy Prickett Morgan

You might think that Arvind Krishna, chief executive officer of IBM, just has a bad case of sour grapes because Nvidia, not Big Blue, is the dominant supplier of AI hardware and the main driver of the GenAI boom. But maybe Krishna is just the adult in the room with deep experience with enterprise customers, and that is why he poured some cold water on the exuberant spending forecasts for AI systems spending.

Krisha sat down with Nillay Patel, editor in chief of The Verge on his Decoder podcast last week, and as part of an hour-long conversation, they talked about GenAI investment numbers, and IBM’s chief said that the numbers did not add up.

To be specific, Krishna said that it took around $80 billion to build 1 gigawatt of datacenter capacity and fill it up with compute engines and networks to make an AI supercomputer that could burn that much juice. (We have heard people say it was $50 billion, and this number seems a bit high, but that might include power over the course of the useful life of a GPU or XPU compute engine.) And if one company – and Krishna was referring to OpenAI – commits to somewhere between 20 gigawatts to 30 gigawatts of compute, that rounds up to around $1.5 trillion in capital expenses. And you have to use it up in five years and retrofit it. The total worldwide commitments of the companies that are chasing artificial general intelligence, or AGI, said Krishna is around 100 gigawatts of capacity, which is $8 trillion of capital expenses.

“There is no way you are going to get a return on that is my view,” said Krishna. “Because $8 trillion of capex means you need roughly $800 billion of profit just to pay for the interest.”

Krishna went on to say that he thought it was very low odds – somewhere between 0 percent to 1 percent – that the current set of AI technologies would get us to AGI.

“I think this current set is great,” Krishna continued. “I think that it is incredibly useful for the enterprise. I think that it is going to unlock trillions of dollars of productivity in the enterprise, just to be absolutely clear. That said, I think AGI will require more technologies than the current LLM path.”

The real question is do we really want to create machines that will have the same rights as people? That seems to be the effect – the consequence – of AGI even if it is not the explicit goal of the research being done by the model builders of the world. We have not seen a lot of talk about that issue. And nor do we expect there to be any from companies like IBM that are very much interested in further automating the enterprise with GenAI technologies and not spending much time thinking about AGI.

This Issue Sponsored By

Table of Contents

Content archive

Recent Posts

Subscribe

Pages

Search