AI industry pivots to efficiency and competition amid global investment

Investors see data centre projects that power AI workloads as attractive, but analysts warn success depends on three constraints: availability and cost of GPUs, overall construction expenses, and especially reliable, affordable electricity. Gartner projects global data‑centre electricity use to reach 565 TWh in 2026, a 26 % rise from 2025, and highlights the need for robust power infrastructure.

In a televised AI‑vs‑AI showdown, several large language models were eliminated after poor performance in a World Cup prediction contest, underscoring that AI systems now compete on reliability and real‑world accuracy rather than sheer size.

A Stanford‑led AI Index Report finds the industry’s earlier focus on ever‑larger models is shifting. New open‑source models such as OLMo 3 Think achieve comparable results with far fewer training tokens, showing that data quality and refined inference can outweigh raw compute.

Ant Group’s LingTech unit launched LingBot‑Depth 2.0, a space‑perception model trained on 150 million data points. The model delivers markedly better depth‑completion and object‑recognition performance, especially in challenging glass and mirror scenarios, and is paired with the LingBot‑Vision visual base.

China’s authorities are reportedly discussing restrictions on foreign access to its leading AI systems, including both open‑source and proprietary models, and on funding channels for Chinese AI startups. The move reflects growing geopolitical concern over AI dominance and the desire to protect domestic intellectual property.