Deutsche Bank lays out AI superintelligence scenario by 2030, flags hurdles

Deutsche Bank laid out a scenario in which artificial intelligence reaches superintelligence by 2030 if models learn to improve themselves and computing capacity keeps expanding, while warning that scientific uncertainty, power constraints and costs could derail that path.

In a report, the bank’s Research Institute said the outcome would depend on whether advances in AI scaling continued and systems could autonomously improve the technology used to build future models. It also identified constraints, including electricity supply, data availability, and the time needed to build computing infrastructure.

Alphabet (GOOGL), Amazon (AMZN), Meta Platforms (META), Microsoft (MSFT), and Oracle (ORCL) are set to spend almost $800B on capital expenditure this year, mostly on data centers, and more than $1T annually through 2030 and beyond, the report estimated.

Training a hypothetical AGI-like model would require infrastructure costing $300B to $500B and about 6 gigawatts of power, the bank estimated. Completing such a training run in four months using Nvidia’s (NVDA) Vera Rubin chips would require about 7.8M chips and 26 gigawatts of facility power, at a cost of $900B at today’s prices. The report said a comparable cluster could cost about $200B by 2030 if current trends in computing efficiency continued.

The bank said the investment case for AI did not depend entirely on achieving superintelligence. Business adoption of AI tools was progressing about three times faster than previous waves of mobile, internet, and cloud computing, while revenues for the AI economy outside China would likely exceed $200B in 2026, up from less than $1B in 2022.

However, Deutsche Bank warned that financing for AI infrastructure remained largely dependent on expectations of future returns rather than current revenue. The report also pointed to pressure on hyperscalers’ cash flow as capital spending rises.

Google DeepMind Chief Strategy Officer Jasjeet Sekhon separately warned of the risk of an “AI air pocket” in which spending occurs but the expected revenue fails to materialize, as cited in the report.