CLSA wrote in its latest analysis on global technology outlook, exploring the rapidly maturing landscape for artificial intelligence as the sector enters 2026.
With AI technology evolving, new debates are emerging around efficiency, cost, and the value of tokens generated by language models. Not all tokens are created equal—differentiated by a need for precision, speed, latency, or cost—and the proliferation of efficient large language models (LLMs) is prompting a shift away from the “bigger is better” philosophy.
According to CLSA, AI-specific capital expenditure is projected to rise 25% in 2026, with attention turning to the importance of distribution strength, enterprise AI implementation, and reliable deterministic code.
Competition among AI accelerators is intensifying, with specification “wars” expected to dominate as companies vie for technical features. Token consumption is forecasted to grow three- to fourfold, with growing emphasis on measuring token utility and value.
The role of leading players such as Nvidia and Broadcom may come under scrutiny regarding whether their current profit levels are sustainable, especially as new competitors enter the field and hyperscalers diversify their ASIC partners. Chinese firms’ development of efficient LLMs is also challenging U.S. scaling norms.
The analyst highlights how distribution strategies—such as Google’s AI resurgence—are as critical as technology itself. Meanwhile, company-specific factors, including Anthropic’s spike in revenue per token, underline how business models can diverge sharply.
After several years of experimental pilots, enterprise AI integration is expected to accelerate, especially among IT service providers. Yet, the probabilistic nature of AI remains in tension with regulatory compliance, financial exactitude, and operational demands.
Looking ahead, AI start-ups are expected to achieve significant revenue growth, estimated at 125–150% year-on-year. CLSA expects heightened market volatility but recommends investors concentrate on stocks that are central to AI infrastructure, including high-conviction outperform ratings for TSMC and Nvidia, and maintains a positive outlook on memory stocks.
