Likes ReceivedSovereign AI: Nation-State AI Players Enter the Game
May 20, Singapore, opening day of ATxSG (Asia Tech Summit).
Three things happened simultaneously:
OpenAI announced the establishment of its first Applied AI Lab outside the US in Singapore, committing $234 million.
NVIDIA announced the establishment of a research center + physical AI testbed in Singapore.
Google signed a new round of national-level AI cooperation with the Singapore government.
Three US AI giants, the same city, the same day, making moves simultaneously. This is not a coincidence; it's an era emerging. The name of this era is—Sovereign AI. At this point, you might ask: Haven't countries been working on AI all along? What's different this time? The answer is—previously, countries were users of AI; now, countries are becoming AI infrastructure builders.
It's Not "AI Made by the State"
The most common misconception: The core of Sovereign AI is not "the state developing models."
If a country develops its own large model, that's just a technological achievement. But if this model runs on US clouds, uses US company GPUs, and data is stored on US servers—it's still not "sovereign."
Sovereign AI truly defines three layers:
Data sovereignty — A country's critical data (government data, medical records, financial transactions, citizen personal information) must reside on domestic infrastructure.
Compute sovereignty — Critical AI tasks must run on domestic land, domestic power grids, and in domestic data centers.
Model sovereignty — Not completely reliant on services from a few US companies like OpenAI / Anthropic / Google.
To draw an analogy: It's like moving from "using US telephones" to "building your own telephone network"—previously, long-distance calls went through AT&T; now, every country realizes it needs its own switches, copper wires, and base stations.
What Singapore announced today is typical—it wants OpenAI to set up a lab locally, not so OpenAI can sell more ChatGPT in Singapore, but so that "how to use AI" happens on its own soil.
"No One Needs an Atomic Bomb, But Everyone Needs AI"
This quote is from Jensen Huang—spoken to dozens of heads of state at the 2026 Dubai World Government Summit.
Over the past 70 years, no technology has made national leaders hear such words and not object. But when this statement was made, the room fell silent—because it precisely captured what is happening between 2024 and 2026.
Pushing this explosion to the forefront are four concurrent forces.
AI has transformed from a "software product" to "infrastructure." Ten years ago, AI was an app you could install and uninstall; now it permeates power grid dispatch, medical diagnosis, military command, and financial transactions. Once it reaches these levels, it becomes national infrastructure like electricity—you can't outsource your power grid to a foreign company.
Hyperscaler concentration has also become a risk. Over 80% of global AI compute runs on three US companies: Microsoft Azure, Google Cloud, and AWS. Any country's critical operations are delivered to these three—this level of dependency was unprecedented before the AI era.
US export controls are conversely forcing self-build. Since 2023, export controls on high-end GPUs have tightened, with Saudi Arabia, UAE, India, and some Southeast Asian countries placed on the "requires a license" list. Ironically, the more the US controls, the more these countries want to build their own—since they can't guarantee access, they buy more in one go and stockpile.
Data sovereignty laws are also tightening. EU GDPR, China's Data Security Law, India's Digital Personal Data Protection Act, Singapore's PDPA—more and more laws explicitly mandate that critical data cannot leave the country. Data cannot leave the country = models must be trained locally.
Combined, these four forces have turned Sovereign AI from a concept into hard expenditure in national budgets starting in 2024.
Who's at the Table?
The Middle East's two powerhouses are the most high-profile players.
Saudi Arabia — In May, MBS, Crown Prince Mohammed bin Salman, during a visit to Washington, raised the US-Saudi AI investment commitment from $600 billion to $1 trillion. Behind this is PIF (Saudi Arabia's $1 trillion sovereign wealth fund) + the newly established national AI company HUMAIN. NVIDIA reached an agreement with Saudi Arabia to deploy a 5,000 Blackwell GPU "AI factory." HUMAIN's goal is shockingly ambitious—to supply 6% of global AI compute by 2034.
UAE — Abu Dhabi's G42 (the national AI company established in 2018). The Stargate UAE project, in collaboration with NVIDIA, Oracle, Cisco, and SoftBank, is building a 1 GW data center cluster. Microsoft invested $1.5 billion in G42.
Singapore is the Southeast Asian hub. Today's announcements of the OpenAI lab, NVIDIA research center, ASPIRE 2B national supercomputing expansion, coupled with the Smart Nation 2.0 national strategy, are positioning Singapore as the clearing center for Southeast Asian AI compute—all ASEAN countries wanting to do AI but lacking self-built infrastructure capabilities will route through the Lion City.
The second tier is already in the game: India (India AI Mission $1.2 billion national plan), Japan (NEC + Sakana + SoftBank), France (Mistral + government), Canada, the UK—each is spending money to build its own compute stack.
Honestly, this game is no longer a bipolar "US vs. China" contest. It's becoming a multipolar geopolitical map.
Three Ceilings
The story sounds smooth, but the costs are truly high.
Money — Building a 1 GW data center + deploying tens of thousands of GPUs + training engineers requires starting capital of at least $10 billion. This isn't something medium-sized countries can play—so you see the ones truly entering are oil-producing nations, city-state economies, or large-population countries with demographic dividends.
Electricity — A 1 GW data center = the total electricity consumption of a medium-sized city. Singapore's land area is so small its power grid is already strained, so it specifically enacted the Digital Infrastructure Act, requiring all new data centers to meet energy efficiency standards.
Chips — This is the most delicate layer. Even the most ambitious Sovereign AI plans ultimately still need to buy NVIDIA GPUs, use TSMC for manufacturing, and install Broadcom networking. What you can sovereign is the data center's geographic location, grid connection, and operational rights—but sovereignty over the chips themselves is determined globally only by the US and Taiwan.
Strictly speaking, all Sovereign AI is now "semi-sovereign." A single US export license revocation could stall the entire plan. In 2024, both Saudi Arabia and UAE learned this the hard way—when the US temporarily restricted high-end GPU exports to the Middle East, G42 had to publicly promise "not to cooperate with Chinese-related AI companies" to get the license.
Back to that opening scene in Singapore.
OpenAI setting up a lab in Singapore and investing $234 million is not a unilateral expansion decision by OpenAI—it's OpenAI using real money to promise the Singaporean government: "We acknowledge your need to master AI deployment on your own soil."
This is what the Sovereign AI era is changing. US AI giants are no longer lofty suppliers; they are starting to become "contractors" within each country's sovereign AI strategy.
Globalization 10 years ago was about the flow of goods, 5 years ago was about the flow of data, and now globalization has become the mutual nesting of sovereign infrastructure.
National-level players are already at the table. This game is so big, even the hyperscalers can't eat it all.
The copyright of this article belongs to the original author/organization.
The views expressed herein are solely those of the author and do not reflect the stance of the platform. The content is intended for investment reference purposes only and shall not be considered as investment advice. Please contact us if you have any questions or suggestions regarding the content services provided by the platform.
