Sri Lanka should stop trying to compete on model size and build small, task-specific artificial intelligence systems instead, SLT Group Chairman Dr. Mothilal de Silva has argued, calling the country’s lack of computing capacity its “most critical gap.”

His case, reported by NewsFirst, is that the industry’s founding assumption — that capability scales with parameter count — is breaking down, and that this is an opening rather than a disadvantage for a country of Sri Lanka’s size.

The infrastructure problem, stated plainly

De Silva was direct about what the island does not have: the GPU capacity, hyperscale data centres and computational resources needed to train large language models at the scale seen in the United States, China or Europe.

“This is our most critical gap,” he said.

He set against that several things already in motion — government allocations of billions of rupees toward AI and digital infrastructure, ongoing discussions with global hyperscale operators, and SLT-MOBITEL’s own plans to upgrade its data centres to be AI-ready. He also cited more than 70,000 km of fibre already laid, with fibre connectivity expected to reach every secondary school by the end of the year.

”Small is beautiful”

The strategic argument follows from the constraint. GPT-3 ran to 175 billion parameters and newer frontier models have passed a trillion; but smaller models increasingly match or beat much larger ones on specific tasks at a fraction of the compute and operating cost.

“The future is not only about the biggest models, but the smartest, most efficient ones,” de Silva said.

The applications he named are deliberately local: an assistant that understands Sinhala and Tamil and helps citizens navigate government services; offline-capable smartphone apps letting farmers diagnose crop disease in the field without connectivity; AI diagnostic tools in rural clinics.

Governance framed as the harder half

De Silva put as much weight on regulation as on capability, arguing AI is now “no longer viewed purely as a technological breakthrough but as a governance challenge.”

“Who is accountable when an autonomous agent causes harm? How do we ensure AI does not perpetuate bias? How do we protect citizens’ data while enabling innovation?” he asked. “These are not academic questions. They are governance imperatives.”

Context

The remarks come during Sri Lanka AI Week 2026, the 25–30 September programme run jointly by the Ministry of Digital Economy and SLT-MOBITEL, whose National AI Conference is being held on 29 and 30 September at Monarch Imperial, Colombo. The promotional material accompanying the report carries that event’s branding, though NewsFirst did not state where or when the comments were made.

SLT-MOBITEL has been positioning around this: its first-half results set out fibre, 5G and AI as the group’s growth planks.

No costing, timeline or named partner was attached to the data-centre upgrade in the report.