Sovereign AI and the emperor’s new clothes – five pillars for AI sovereignty  

This article is part of our Opinions section, where we invite industry professionals to share their views on the most pressing technology questions of our time.

While the 24-hour news cycle struggles to keep up with President Trump’s policy U-turns and tariff drama, a deeper and more enduring concern is taking root – the erosion of trust in the global tech leader. Trump’s “America First” strategy has cast aside all pretence of US free-trade leadership, shed the illusion of the emperor’s clothes and exposed the underlying zero-sum protectionism.

In the words of Singapore’s Prime Minister, Lawrence Wong, “The US… is rejecting the very system it created“, and policymakers are confronted with the bare truth. The frameworks underpinning our global digital economy can be redrawn at any time, and critical digital infrastructure can be dictated by politics rather than principles.

This is the context where policymakers must now reconsider the case for Sovereign AI – the aspiration for a nation to have control and autonomy over the development, deployment, and governance of AI systems within its borders.

The World Economic Forum tries to take a more level-headed approach, pointing out that Sovereign AI “does not necessarily mean digital isolation, but rather a push for strategic resilience,” and that it “can be done in tandem with global cooperation.” But the image of major AI company CEOs lined up behind Trump at his inauguration is hard to shake.

Once upon a time, countries accepted the subscription-based model of the public cloud as a trade-off to accelerate digital transformation. However, the calculus is different for AI – the stakes are much higher. AI will become the defining socio-economic force of our time, shaping the world order and changing the way we organise ourselves in all industry sectors and at all levels of society.

The distrust may have always been there, but Trump’s capriciousness has made the shadows loom larger. Critics have long condemned the harmful influence of major tech players, decrying their use of dark patterns and unchecked market dominance. Yannis Varoukis’ concept of technofeudalism draws a parallel between Big Tech and the feudal overlords of medieval times. We, the users, are serfs – tethered to dominant platforms, producing and surrendering our data in exchange for the blessings of the algorithm and the allure of the infinite scroll. And AI becomes the instrument for the empire.

How will policymakers look to enhance their nation’s AI sovereignty? Examining this challenge through the lens of the AI value chain reveals five possible pillars of action.

Data

Data is the lifeblood of AI, and control over its collection, management and ownership is central to AI sovereignty. Comprehensive data residency and localisation may not be feasible, but policymakers can review their existing data sovereignty and protection frameworks.

They can also explore the use of data quality requirements to ensure fair data representation. Finally, they can explore the use of data sharing frameworks to create a more inclusive data ecosystem that can benefit domestic players.

Infrastructure

Governments will need to assess the current capacity of national compute infrastructure and evaluate whether additional investment is required to meet their AI ambitions. For example, in 2024, the Japanese government provided US$470 million in funds to five companies, including telecom giant KDDI, for GPUs in 2024.

Strategic decisions must account for two distinct types of compute:

• AI Training: High-performance computing resources required to train large-scale models.
• AI Inferencing: Low-latency, distributed compute capabilities needed to deploy and run trained models in real-time, extending from centralised data centres to the network edge and even end-user devices.

Telecom operators will play a pivotal role in this equation, not only as infrastructure providers, but as critical enablers of edge intelligence and distributed compute.

Models

Many AI solutions are built on large, pre-trained models that demand significant financial investment, computational power and specialised talent to develop -resources often concentrated in the hands of a few dominant players.

To reduce over-reliance on a narrow set of proprietary models and providers, policymakers may consider mirroring other types of supply chain diversification requirements. In parallel, regulators can develop standardised contractual clauses to guide small and medium enterprises (SMEs) in safely accessing and leasing AI models or services.

Application

At the application level, governments will develop sector-specific regulations and frameworks to assert greater control over critical areas such as healthcare, finance, transport.

Beyond regulatory control, policymakers also need to consider taking an active role in designing and champion sandbox initiatives and pilot projects. These can help accelerate AI adoption in high-impact sectors while also fostering collaborative partnerships between domestic companies and major international AI players, serving as a platform for knowledge transfer and capacity building.

User

Policymakers can introduce user consent and control requirements, as well as clear appeal and redress mechanisms to empower users without stifling innovation and hindering adoption.

You might also be interested

• What fintechs can learn from the four biggest anti-money laundering (AML) compliance fines so far this year
• “This is not a sale”: what lies behind OpenAI’s plans to boost its for-profit arm
• How to combat the ransomware that struck Marks & Spencer