AI Scaling Faces Energy and Error Risks as Next-Generation Technologies Seek a Breakthrough

The generative AI race has long pursued better performance by expanding models, computing capacity and data centers, but the marginal gains are declining. As these systems move into legal, financial, compliance and trading decisions, hallucinations can quickly amplify fabricated case law or false alerts. Mohammed Marikar, co-founder of Neem Capital, argues that the industry should pivot toward neurosymbolic and decentralized cognitive architectures that enable traceable reasoning.

Cointelegraph published Marikar's commentary on March 11, 2026. He said global data-center electricity demand will more than double by 2030, with demand in the United States also rising by more than 100% over the same period. A single training run for a frontier model could cost more than $1 billion, while total investment in power grids and infrastructure could reach several trillion dollars. The UK High Court warned lawyers as early as June 2025 not to submit AI-fabricated case law.