‘$1.5 trillion bets’: IBM CEO sounds alarm on AI spending, says current tech has ‘0-1%’ chance of AGI

IBM chief executive Arvind Krishna has issued a stark warning about the financial risks of the global rush to build artificial intelligence infrastructure, cautioning that the pursuit of artificial general intelligence (AGI) could push even the most ambitious technology companies into dangerously expensive territory. 

Speaking on The Decoder podcast, Krishna said the sheer scale of investment now being discussed raises serious questions about sustainability, returns and the realistic prospects of achieving AGI with today’s technologies. 

Soaring cost of AI infrastructure 

Krishna highlighted the extraordinary expense involved in building next-generation data centres. According to him, filling a single one-gigawatt data centre currently costs around $80 billion, a figure he described as “today’s number.” Companies planning infrastructure at the scale of 20-30 gigawatts could therefore be staring at capital outlays of up to $1.5 trillion. 

Adding to the burden is the short lifespan of AI hardware. Krishna noted that advanced AI chips typically need to be replaced within five years, creating a cycle of repeated reinvestment that compounds long-term costs. This, he warned, could severely strain balance sheets as companies race to scale faster than their revenues can justify. 

Can AGI ever deliver the returns? 

Krishna also questioned whether such massive spending can realistically pay off. He estimated that an $8 trillion investment in AGI would require $800 billion in profit just to service interest payments, a calculation that casts doubt on whether the economics of AGI make sense at current cost levels. 

His remarks come as OpenAI, the creator of ChatGPT, announced $1.4 trillion in long-term infrastructure buildout agreements with partners including Nvidia, Broadcom, Oracle and Alphabet. OpenAI CEO Sam Altman has expressed confidence that the company can generate strong returns on its capital investments. Krishna, however, argued that many of today’s infrastructure commitments are based more on belief than on proven outcomes. 

‘Zero to 1%’ chance 

Perhaps Krishna’s strongest statement was his assessment of AGI itself. He said he assigns current technologies only a “zero to 1%” chance of achieving true artificial general intelligence, stressing that large language models, while powerful, do not amount to real intelligence. 

While acknowledging the significant productivity gains AI already delivers in enterprise settings, Krishna emphasised that these advances do not guarantee a path to AGI. He suggested that any real breakthrough would likely require combining language models with systems grounded in “hard knowledge,” though he admitted he remains uncertain whether even that approach would be sufficient. 

Despite such warnings, major technology firms continue to accelerate spending. Alphabet has raised its 2025 capital expenditure outlook to $91-93 billion, while Amazon has increased its estimate to $125 billion. Across the industry, AI infrastructure investment is expected to reach $380 billion this year alone. 

For now, Krishna urged the industry to focus on extracting real productivity and business value from existing AI tools, while recognising that trillion-dollar infrastructure bets may not necessarily deliver the breakthroughs many are hoping for.