What AI Sovereignty Actually Means

For governments mapping out their AI strategies, sovereignty is rarely a yes-or-no question. Some nations pursue full self-sufficiency across the entire AI stack. Others, lacking the capacity to build everything from silicon to foundation models, take a layered approach: build capabilities where they can, partner where they can't, and avoid being locked into any single vendor or technical path. That flexibility is the essence of sovereignty as a practical matter: the freedom to choose tools, control data, and deploy applications without constraint.

The infrastructure demands of AI complicate that goal. Training general-purpose models requires enormous, centralized compute. But most real-world AI usage is inference, the work of serving user requests, and that requires infrastructure close to the people and devices making those requests. Distributed edge networks that place compute in proximity to end users are what actually deliver AI to consumers at scale, with the low latency that demanding applications require.

Governments also want the benefits of AI to reach whole economies, not just large enterprises. Traditional infrastructure models financially punish experimentation and small-scale deployment. Serverless computing addresses this, letting developers pay only for what they consume and eliminating the need for expensive reserved capacity. Accessible entry points, including free tiers for testing and building, allow small businesses, research institutions, and non-profits to participate in AI development.

Choice as a Technical Principle

Control over data is often conflated with data locality, but true autonomy comes from visibility and governance. Security tools that give organizations the power to monitor and restrict access to their data can accomplish sovereignty goals without requiring AI workloads to be physically contained within national borders.

Market dynamics matter too. Building applications on closed models creates dependency: switching models later becomes costly, and development becomes subject to a single provider's roadmap. Diversified AI strategies need access to a range of both open source and closed models. Platforms that are model-agnostic allow developers to swap alternatives in and out as better, cheaper, or more specialized models emerge. Interoperability layers that connect multiple AI models, regardless of vendor, through a unified interface further reduce lock-in. Underpinning these approaches are open standards, which prevent dependency, keep markets competitive, and maximize the options available to governments and their developers.

Cloudflare's mission is to help build a better Internet, and we apply that to AI by providing tools for developers to build and deploy Internet and AI applications that are widely, and in many cases freely, available. Our network spans 330 cities in over 125 countries, which supports local developers to build and deploy AI tools and services right where they are through Workers AI, our serverless inference platform, and AI Gateway, which provides unified access to AI models from any provider.

Training Regionally Relevant Models

To date, most AI development has centered on English-centric, and increasingly Chinese-centric, data. That has left many nations without AI tools that fully understand their languages and cultural nuances. Governments in several countries are addressing this gap directly by developing and open-sourcing local language models, lowering the barrier to entry for local organizations to create customized solutions for their own markets.

Cloudflare is bringing such locally developed, open-source models to a global developer audience via Workers AI. Regional initiatives we now support include:

  • India: The national vision of "AI for All" focuses on inclusive growth through AI, with the Bhashini platform enabling Internet access in 22 official languages. Cloudflare now offers @cf/ai4bharat/indictrans2-en-indic-1B, a translation model covering 22 Indic languages including under-resourced ones like Kashmiri, Manipuri, and Sindhi.
  • Japan: The country's strategy aims to make it "the world's most friendly AI nation," prioritizing AI that captures the cultural context of the Japanese language. Cloudflare offers @cf/pfnet/plamo-embedding-1b, a Japanese text embedding model developed by Preferred Networks, Inc. with government compute support through the GENIAC program. It powers RAG applications and semantic search in Japanese.
  • Southeast Asia: Singapore's National AI Strategy 2.0 champions SEA-LION, a family of open-source models tailored to the region's languages, including Bahasa Indonesia, Bahasa Malaysia, Thai, Vietnamese, and Tamil. The @cf/aisingapore/gemma-sea-lion-v4-27b-it model is now available on Workers AI.

These models are offered on the Workers AI platform via the following code snippets:

curl --request POST \
  --url https://api.cloudflare.com/client/v4/accounts/ACCOUNT_ID/ai/run/@cf/ai4bharat/indictrans2-en-indic-1B \
  --header 'Authorization: Bearer TOKEN' \
  --header 'Content-Type: application/json' \
  --data '{
    "text": ["What is your favourite food?", "I like pizza"],
    "target_language": "guj_Gujr"
}'
curl --request POST \
  --url https://api.cloudflare.com/client/v4/accounts/ACCOUNT_ID/ai/run/@cf/pfnet/plamo-embedding-1b \
  --header 'Authorization: Bearer TOKEN' \
  --header 'Content-Type: application/json' \
  --data '{
  	"text": [
            "PLaMo-Embedding-1Bは、Preferred Networks, Inc. によって開発された日本語テキスト埋め込みモデルです。",
            "最近は随分と暖かくなりましたね。"
        ]
}'
curl --request POST \
  --url https://api.cloudflare.com/client/v4/accounts/ACCOUNT_ID/ai/run/@cf/aisingapore/gemma-sea-lion-v4-27b-it \
  --header 'Authorization: Bearer TOKEN' \
  --header 'Content-Type: application/json' \
  --data '{
  "messages": [
    {
      "role": "user",
      "content": "แล้วทำผัดไทยอย่างไร"
    }
  ]
}'

The choice by India, Japan, and Singapore to open-source these language models signals a specific understanding of sovereignty: that autonomy comes from having options, and that practical leverage depends on developers actually being able to use the technology. Supporting such local language models is a commitment to an interoperable and competitive AI ecosystem, one in which governments and developers can solve local problems, generate economic opportunity, and preserve linguistic and cultural heritage. It addresses the full spectrum of developers—from free-tier experimentation to pay-as-you-go scaling in production.