Sovereign AI gets push it needs as control slips away from Silicon Valley

Chinese start-up DeepSeek sent shock waves through global markets in January, with its latest artificial intelligence breakthrough causing a nearly $600 billion rout in Nvidia’s market value. The company has since recouped some of those market losses.

The Chinese group claimed its generative AI chatbot rivals the best efforts of OpenAI, Anthropic and Meta – achieving comparable performance, but with a fraction of the computing power and fewer chips from Nvidia, significantly lowering the costs of development.

DeepSeek's rise is set to accelerate the drive to build national AI infrastructure and models – known as “sovereign AI” – given their perceived greater accessibility as a hedge against technological dependency on other countries.

Already, governments worldwide are pouring billions into supercomputers and proprietary AI models. Nvidia, which became the world’s most valuable company last year on the back of insatiable demand for its AI processors, already derives 10 per cent of its revenue from supplying chips to countries building their own infrastructure.

That figure is likely to climb sharply, as more nations treat AI as a strategic asset. The launch of the DeepSeek-R1 model and its ascent to the top of the Apple’s App Store underscores a broader decentralisation. Control is slipping away from Silicon Valley’s dominant players – notably OpenAI and Anthropic – as more nations and enterprises develop their own AI systems.

Underpinning this shift are the “open” models, not only DeepSeek’s latest release but also Meta’s Llama, which make their code and architecture publicly available, allowing anyone to modify, improve and fine-tune them to meet their specific needs.

Open models will be attractive to countries that fear dependency on foreign-controlled AI infrastructure could become a national security risk. This concern has prompted the US to tighten export controls on advanced computer chips. Similarly, China has introduced its own export restrictions to protect its technological assets.

The result of these measures is a world where nations are less willing to rely on foreign AI, fearing they could one day be cut off from critical technology at the flick of a switch. Enter sovereign AI.

A good example is Stargate, a high-profile AI infrastructure project valued at up to $500 billion, which aims to develop colossal data centres in the US, with companies including OpenAI, SoftBank and Oracle. Trump recently called the project a “declaration of confidence in America”.

China, meanwhile, is urging local companies to buy AI chips from national champion Huawei, to curb reliance on Nvidia, based in California, which remains constrained by US export controls.

What these geopolitical moves show is that the AI arms race is no longer just about who builds the best models – it is about who controls the infrastructure behind them.

But this battle to control the architecture underpinning AI is not confined to the US and China; it has become a multipolar fight. India is investing heavily in the “IndiaAI Mission” to develop large language models, semiconductor manufacturing and cloud solutions.

Japan, too, is ramping up its investment in domestic AI and supercomputing, to bolster national capabilities and reduce reliance on foreign technology.

On top of that, the UAE is establishing itself as a regional powerhouse in AI, with a sharp focus on developing Arabic-language models, such as Falcon. Saudi Arabia has pledged to spend $40 billion on AI models, reports suggest.

The EU has also taken steps to bolster its AI infrastructure with initiatives like LUMI, one of the continent’s most powerful supercomputers, based in Finland. But these efforts are still in their early stages. While the EU arguably leads in progressive AI legislation, its regulatory-heavy approach could prove more of a constraint than a catalyst for sovereign AI development.

Still, these global projects mark a big shift in the thinking about AI. For years, the assumption was that building a world-class generative AI model required vast sums of money, cutting-edge computing infrastructure, gargantuan data sets and deep talent pools.

DeepSeek and other open AI pioneers have upended that thinking. The Chinese company’s new model has demonstrated that high-performance AI can be built at a fraction of the price and time once thought necessary. And crucially, it’s open.

For smaller nations, this changes everything. Suddenly, the barriers to entry have been lowered. Governments that once saw AI self-sufficiency as a pipe dream can now consider the possibility of building their own national AI infrastructure. This is because these open models lower the cost, increase the flexibility and widen access to advanced AI technology.

And that has major implications for multinational companies doing business in these countries. As nations push for sovereign AI and assert tighter control over their digital infrastructure, companies will face new challenges in deploying AI across different regions.

Governments increasingly see AI as a strategic asset, leading to stricter regulations, data localisation requirements and trade restrictions that limit access to key technologies. This can be seen in measures like China’s restrictions on exporting advanced AI models, and America’s ban on selling high-end chips to Chinese firms.

Despite these regulatory hurdles, the AI industry is unlikely to become completely fragmented. Instead, a hybrid model could emerge, where certain elements − such as semiconductor manufacturing and cloud computing − remain part of a global supply chain, because no single country can fully control AI infrastructure.

The production of high-end chips relies on specialised expertise and supply chains spanning the US, Taiwan, South Korea and the Netherlands. Cloud services, dominated by US companies like Amazon, Microsoft and Google, require massive global data centres that few nations can replicate at scale.

While governments push for AI sovereignty, the sheer cost and technical barriers of developing entirely self-sufficient AI ecosystems ensure that some degree of interdependence will persist.

With that said, regional customisation is becoming more important, as governments demand AI systems that reflect local laws, cultural norms and economic policies. In India, AI models must accommodate local languages and data sovereignty laws, while Gulf nations are investing in Arabic-language AI to serve regional markets.

But there is still a fundamental change occurring: more players are entering the AI game, spurred on by DeepSeek’s breakthrough.

And as the lines between economic power, national security and technological supremacy continue to blur, one thing is clear: the fight for AI dominance is only just beginning.

Michael Wade is professor of Strategy and Digital, and the director of the TONOMUS Global Center for Digital and AI Transformation at IMD