Deutsche Bank says AI investors face a major reckoning.

Deutsche Bank says artificial intelligence could theoretically reach superintelligence by 2030, but only if model scaling continues, AI systems learn to improve themselves, and the industry can overcome enormous constraints around power, infrastructure and financing.

The bank's Research Institute said the path depends heavily on whether future AI systems can autonomously improve the technology used to build the next generation of models.

But the physical requirements are staggering. Deutsche Bank estimates that training a hypothetical AGI-like model could require $300 billion to $500 billion of infrastructure and roughly 6 gigawatts of power.

Running an equivalent training cycle in just four months using Nvidia (NVDA) Vera Rubin chips could require approximately 7.8 million chips and 26 gigawatts of facility power, with the infrastructure costing about $900 billion at today's prices. The bank estimates a comparable cluster could fall to roughly $200 billion by 2030 if current improvements in computing efficiency continue.

The industry is already spending aggressively. Alphabet (GOOG), Amazon (AMZN), Meta (META), Microsoft (MSFT) and Oracle (ORCL) are expected to invest almost $800 billion in capital expenditures this year, according to Deutsche Bank, with annual spending potentially exceeding $1 trillion through 2030 and beyond.

Yet the bank stressed that the AI investment case does not require superintelligence. AI adoption by businesses is reportedly advancing roughly three times faster than previous mobile, internet and cloud computing waves, while AI-economy revenue outside China could exceed $200 billion in 2026, up from less than $1 billion in 2022.