UBS argues AI capabilities are improving at a super exponential pace, while commercial returns are likely to build along a far more linear path. Power supply, grid connections, local permitting and slow enterprise adoption act as speed limits on value capture. The authors see selective opportunities in scarce, hard to replace inputs across the AI and power value chains.
Signal over Noise #14: The AI “super-exponential” meets its speed limits
UBS
Ulrike Hoffmann-Burchardi
Research
9 Pages
Key Takeaways
Capabilities Outrun Earnings: The longest task an AI model can complete at 50% success has climbed from seconds with GPT-2 to over 10 hours today.
Power Queues Lengthen: Median time from grid interconnection request to commercial operation averaged 4.1 years over the last decade, up from 2.5 years previously.
Local Opposition Builds: Local opposition has blocked or delayed 75 data center projects representing roughly $130 billion of planned investment.