Deutsche Bank Research: Superintelligence by 2030 Possible, but Hinges on Two Conditions
On October 12, Deutsche Bank Research released a report titled "AI Boom or Bust: What Would Superintelligence Feel Like, and When Will We Get There?", analyzing the likelihood and economic impact of achieving artificial general intelligence (AGI) and superintelligence by 2030.
The report says that if AI can keep scaling compute and achieve true recursive self-improvement, its capabilities may enter an accelerating growth phase, making superintelligence that surpasses human experts in almost all cognitive tasks by 2030 not impossible. But that depends on unproven technical assumptions and real-world constraints around compute, power, capital and data.
Deutsche Bank points to two key conditions for AGI by 2030: scaling laws must continue to hold, and AI must achieve genuine recursive self-improvement. It estimates that training the most advanced AI models in 2026 requires about 10^26 to 10^27 FLOPs, while an AGI-level model might need three orders of magnitude more. By 2030, the compute and power hardware investment for such training could fall to around $200 billion.
The report also warns of data bottlenecks: adjusted public human text available for training could approach its limit between 2026 and 2032, and as AI-generated content grows, data quality and verification reliability may become new constraints. Even if superintelligence does not arrive, Deutsche Bank believes AI can still generate significant economic value through productivity gains, lower costs and new business models. It projects AI industry revenue, excluding China, to rise from under $1 billion in 2022 to more than $200 billion in 2026, and expects Alphabet, Amazon, Meta, Microsoft and Oracle to spend nearly $800 billion on capital expenditure in 2026, potentially exceeding $1 trillion annually by 2030 and beyond.
