Nvidia powers 92% of sovereign AI models worldwide, report finds

Nations racing to build independent artificial intelligence systems remain overwhelmingly dependent on a single American chipmaker. A report released on August 5 by Counterpoint Research found that Nvidia ↗ 2.27% graphics processing units power 92.4% of sovereign AI large language models worldwide, underscoring a deep contradiction at the heart of the global push for technological self-reliance.asiae+2

The Paradox of AI Independence

The study, titled the "Sovereign AI LLM Index," analyzed more than 170 AI models developed across 80 regions, identifying over 55 countries in the Asia-Pacific, Middle East, Europe, Africa, Canada, and Latin America — excluding the United States and China — that are actively building domestic AI infrastructure.chosun+1

Despite the stated goal of reducing reliance on U.S. technology giants such as OpenAI and Google, these nations continue to depend on American semiconductor firms to power their efforts. AMD Advanced Micro Devices, Inc. ↘ 1.21% holds a distant second place with 4.1% market share, followed by Cerebras at 1.7%.asiae+1

Mark Einstein, Principal Analyst at Counterpoint Research, noted that while the hosting of sovereign AI is increasingly localized within each country, the sourcing of AI chips has not followed suit. "The majority of domestic LLM developers are being drawn into the NVIDIA CUDA ecosystem and are using NVIDIA AI chips to train their LLMs," he said.asiae

Challengers and Alternatives

Some efforts to diversify are underway. South Korean AI startup Moreh has developed models based on AMD semiconductors, while another Korean firm, Upstage, has adopted AMD's MI355 GPU. In the inference chip market, Korean semiconductor company Rebellions has partnered with SK Telecom to expand its footprint.asiae

Counterpoint Research noted that AMD is targeting niche markets in the European Union and Australia, while Cerebras has made progress in the United Arab Emirates. Qualcomm ↗ 4.66% has partnered with Saudi Arabia's Humane, and Google Alphabet Inc. ↘ 0.88% is supplying its tensor processing units to Israeli AI startups. Japan is using Fujitsu's A64FX ARM processor to train LLMs.asiae

Structural Lock-In

Nvidia's dominance stems not only from its GPU performance but from its integrated ecosystem. The CUDA development platform, compatible exclusively with Nvidia GPUs, has become the industry standard for AI developers. Counterpoint Research described this integrated ecosystem spanning networks and software as Nvidia's "greatest competitive advantage."chosun

The findings add urgency to concerns already voiced in countries like South Korea, where semiconductors now account for 44% of exports and where commentators have warned against betting national prosperity on a single technology sector.chosun