Among its collection of 2026 predictions in the AI space, GlobalData is anticipating that 2026 will be the year of “efficiency”, with small language models (SLMs) gaining relevance as enterprises leverage AI for domain and industry-specific use cases. Running SLMs locally or in a self-hosted environment helps address privacy, data security, and regulatory compliance concerns because sensitive information does not need to be sent to cloud servers.
However, SLMs will not replace LLMs for all use cases. Rather, they will complement or displace them for specific tasks where efficiency, speed, privacy, or cost are important considerations. Industries such as financial services and healthcare are using SLMs tuned with proprietary data for specialised automation, showing greater accuracy at lower costs.
Go deeper with GlobalData
Access deeper industry intelligence
Experience unmatched clarity with a single platform that combines unique data, AI, and human expertise.
Adoption of agentic AI will accelerate in the enterprise space
GlobalData’s Agentic AI Forecast estimates that global revenues will enjoy a CAGR of 48% in the period from 2024 to 2029. Sales from agentic AI tools and services are projected to grow from $6.4bn in 2024 to $45.4bn in 2029. This indicates the scale of the impact that agentic AI will have on the overall GenAI market, helping enterprises materialise ROI gains. In 2026, businesses will incorporate the technology into their workflows, and the measurable results will help ease the pressure on technology leaders to demonstrate the financial impact of AI implementations.
However, there remain many questions surrounding the technology. Leveraging context awareness remains the critical piece in agentic AI. Those organisations able to ground AI agents in relevant, dynamic, institutional knowledge will enjoy productivity gains, especially with domain-specific, embedded agentic tools.
GlobalData analyst Beatriz Valle commented: “The rise of agentic AI will throw in stark relief the need to use data of the highest quality, which has been vetted following strict guidelines. Agentic AI leverages contextual decision-making to interpret its environment and understand complex instructions. This is only possible when the AI is trained on large, high-quality, and contextually relevant data covering a wide range of scenarios, outcomes, and user behaviours.”
Small language models will facilitate greater efficiencies for enterprises
SLMs will be deployed for specialised or industry-specific tasks in finance, healthcare, government, legal, and other verticals. This domain focus can deliver better accuracy for targeted use cases than a generic LLM. An increased focus on the environmental impact of AI, as well as financial and logistical considerations around power and electricity requirements, will increasingly lead organisations to a more targeted approach. With AI’s growing carbon footprint, SLMs offer a greener pathway because they consume significantly less energy to train and run.
US Tariffs are shifting - will you react or anticipate?
Don’t let policy changes catch you off guard. Stay proactive with real-time data and expert analysis.
By GlobalDataChina will advance in the AI race
In 2026, large language models, including DeepSeek-V3 and Alibaba’s Qwen 2.5-Max, will be increasingly popular in emerging markets. GlobalData analyst Beatriz Valle continued: “Because of its engineering and industrial prowess, China will be best positioned to win the forthcoming race in physical AI, the intersection of AI with physical systems, enabling machines to interact with the world.”
The presence of AI models embedded in robots, drones, self-driving vehicles, and industrial IoT equipment will grow in 2026. However, chips will remain China’s toughest challenge. The US’s inconsistent approach to export restrictions will continue in 2026. Another hurdle for Chinese companies will be the rise of sovereign AI, which will limit the penetration of Chinese technologies in Western companies, as the geopolitical environment becomes increasingly complex.
