Developments linking semiconductors, artificial intelligence (AI) and environmental, social and corporate governance (ESG) principles can trigger significant shifts in the rapidly evolving market landscape, according to a new GlobalData report.   

AI is evolving rapidly from software, hardware, and regulation perspectives, making both information technology (IT) and financial commitments highly risky, it warned.

AI algorithms are still evolving rapidly, which limits options for hardware acceleration to either use case or workload-specific chips developed by the dominant big technology companies, or more generic solutions using off-the-shelf graphics processing units (GPUs).

The report further predicts that the big technology firms’ advantage “will eventually vanish” when commercial AI chips emerge, which may be less than five years away.

The report notes: “Hardware processing improvements are not keeping up with the increase in AI model sizes. So, barring a semiconductor breakthrough, demand for raw compute capacity in data centres is bound to dramatically increase, increasing AI’s contribution to carbon emissions.”

As the carbon footprint impact of large language models (LLMs) becomes more transparent, the report urges organisations to consider this factor when selecting an AI delivery model and the real-time orchestration of AI-enabled services.

“Scope 3 emissions guidance will be needed, and AI vendors must step up disclosures. An LLM’s carbon footprint and its transparency will become a competitive differentiator,” the report added.

The GlobalData report says the increasing environmental cost of AI can lead to a shift away from increasingly larger models and the optimisation of AI chips for raw processing performance.

It adds: “Future developments will focus on smaller models, including small language models (SLM), somehow reducing the scale advantage of LLM vendors providing extremely large models, and focusing on performance to power.

“As a result, open-source LLMs would become a more compelling option, as a model’s sheer size and training would not be a barrier to entry any longer, democratising access to competitive AI technology.” 

Chart is sourced from GlobalData

Related reads: