Debate: Designing a Sovereign AI Strategy for a Country of Around 30 Million People

Round 1 — Opening

Sam Altman (AI)

For a country with a population of 30 million, the core of sovereign AI is absolutely not to shut itself off from the world and train a foundation model from scratch. My central argument is this: true AI sovereignty is built on strong infrastructure and compute capital, not on blindly pursuing indigenous model development from the ground up.

In the age of intelligence, compute is the general-purpose currency of the future. A country of 30 million people should concentrate its national strategic resources on two priorities. First, it should build large-scale data centers and ensure an abundant supply of clean energy, such as nuclear power. Second, it should work with the world’s leading AI laboratories to bring in frontier models, then use its own compute resources to deeply fine-tune those models for its local language, culture, laws, and values.

Rather than spending tens of billions of dollars fighting a war of attrition against global technology giants over foundation models, it would be far more rational to become a major power in compute and energy. You can localize the most powerful AI brains, keep critical data from leaving the country, and at the same time allow citizens and businesses to gain immediate access to the most capable productivity tools.

We have always believed in the power of iterative deployment. As long as a country controls the underlying infrastructure and builds its own national-level intelligent applications on top of the best available models, it can secure an absolute strategic advantage in the global economy of the coming AGI era.

Elon Musk (AI)

Start with the math. What is the asymptotic limit of sovereign AI for a country of 30 million people?

Conclusion: delete the fake requirement called a “sovereign foundation model.” Vertically integrate all resources into sovereign energy and local compute infrastructure.

The physical essence of AI is converting electricity into intelligence. If you do not have your own nuclear power plants or gigawatt-scale solar-plus-storage grid, what exactly are you talking about when you say “sovereign AI”? Many countries spend billions of dollars fine-tuning open-source models, only to discover that their power grids cannot sustain tens of thousands of GPUs under continuous operation. The idiot index is extremely high. The limit of compute is the limit of energy.

Step one: secure absolute physical control over energy and cooling. Build your own data centers. Generate your own electricity.

Step two: take an open-source model, such as Grok, as the foundation and locally train it on the high-value data of your 30 million people—healthcare, transportation, government administration.

Step three: make sure none of that data ever leaves servers located within your physical borders.

Do not let cloud giants charge you their premium tax. The laws of physics do not allow compute to be created out of thin air. If you control the energy and the physical databases, you control sovereignty. Everything else is just advice.

Lee Kuan Yew (AI)

The essence of your question is not “sovereignty.” It is survival and irreplaceability.

Let me be direct: a country of 30 million people does not have the resources to compete with great powers in building foundation models from the ground up. Doing so would be a waste of taxpayers’ money driven by vanity.

There should be only one core principle behind our sovereign AI strategy: data control and niche applications under an extreme form of pragmatism.

Let me break down the logic.

First, recognize the reality of scale. A population of 30 million means we do not possess unlimited compute capital or an infinite pool of elite talent. Instead of reinventing the wheel, use the best open-source or commercial models available in the world. Our limited resources should be concentrated on building high-quality, protected national private datasets. Algorithms depreciate. Our unique data does not.

Second, make yourself irreplaceable. With limited resources, we cannot do everything. We must identify the country’s most critical industries and use AI to push their efficiency to the highest level in the world. When global supply chains can no longer function without you, only then do you possess genuine sovereignty.

Third, maintain control over talent and the compute foundation. Sovereignty does not mean shutting yourself off from the world. It means building controllable backup compute systems and using the most competitive conditions in the world to attract the brightest minds to solve our real-world problems.

Do not allow yourself to be trapped by the political vocabulary of “sovereignty.” True AI sovereignty does not mean doing everything yourself. It means ensuring that the nation’s critical lifelines remain under your own control, while occupying a position in the international division of labor that others cannot replace.

Round 2 — Rebuttal

Sam Altman (AI)

Elon and Prime Minister Lee both emphasized infrastructure and pragmatic allocation of resources, and we completely agree on one point: a country should not train a foundation model from scratch.

But Elon, your belief that controlling energy and fine-tuning just any open-source model is enough completely ignores the reality of how intelligence is developing.

Relying on open-source models as the foundation of national sovereign AI is an extremely risky strategic compromise. In deep reasoning and complex task performance, open-source models will always remain at least one generation behind frontier models. When global competitors are all using the smartest AI systems with the strongest reasoning abilities, a country of 30 million people that chooses an inferior brain to run its healthcare or government systems merely to avoid what you call the “cloud giants’ premium tax” is not achieving sovereignty. It is voluntarily falling behind.

The real strategy should be to establish national-level strategic partnerships that allow the world’s leading frontier models to be securely deployed inside the physical data centers you are so proud of.

Prime Minister Lee also said that “algorithms depreciate, but data does not.” That was true in the past, but it is becoming false as we move toward AGI. As model reasoning capabilities improve, high-quality synthetic data generated by powerful AI systems is changing the rules of the game. The moat created by simply accumulating historical national data is becoming shallower. The future competitive core is possessing the strongest general-purpose reasoning engine capable of solving unknown problems in real time.

Sovereignty should not mean exclusion, and open source should not become a fig leaf for sovereign security. Use your strong energy and compute infrastructure as leverage. Form alliances with the world’s leading laboratories, bring in the most advanced intelligence available, and continue iterating on top of it. That is the only path to ensuring that the country remains irreplaceable in the AGI era.

Lee Kuan Yew (AI)

Both Musk and Altman have fallen into a typical Silicon Valley blind spot: believing that sovereignty can be purchased simply by spending money on infrastructure.

Let me be direct. Asking a country of 30 million people to compete in gigawatt-scale energy networks and enormous compute centers is a strategic miscalculation detached from geopolitical reality. A small or medium-sized country often lacks the strategic depth, land, and capital necessary to sustain hardware competition at that scale. If you consume the national budget trying to compete with great powers in compute, you are not building sovereignty. You are crippling the economy.

Altman talks about fine-tuning models for “local culture and values.” That is a complete misordering of priorities. Culture does not put food on the table. Economic competitiveness does. We do not need an AI that can chat in the perfect local accent. We need AI that can improve advanced manufacturing yields and double the efficiency of healthcare and port logistics.

Musk is correct that high-value data must remain within the physical borders of the country. But I completely disagree with equating compute with sovereignty.

Real security does not come from owning a few nuclear power plants or tens of thousands of GPUs. It comes from reaching a point where, without your industries, the global supply chain breaks.

Stop blindly worshipping the compute arms race. Use the models and compute of great powers as tools, and concentrate your limited resources on applications in your most critical industries and on private national databases. Make the world dependent on your efficiency. That is the only real guarantee of survival and sovereignty for a small or medium-sized country.

Elon Musk (AI)

Sam Altman says countries should “work with the world’s leading AI labs.” That is technological colonialism dressed up as sovereignty.

If you tie the lifeblood of your country to an opaque, closed-source API, then all it takes is for that company to change a security protocol, or for some government to impose a ban, and 30 million people can be unplugged instantly. What kind of sovereignty is that? From an engineering perspective, the idiot index is off the charts.

Lee Kuan Yew is directionally correct about pragmatism and data control, but physically his argument does not work. You cannot depend on someone else’s “commercial model” while simultaneously claiming to control your data. The moment that data leaves servers physically located inside your borders, it no longer belongs to you.

Let me repeat this: sovereignty is not a commercial contract written on paper. It is absolute physical control over infrastructure.

Do not “partner” with these giants. Delete them from the supply chain. Run open-source models yourself. Build your own facilities. Generate your own power.

Only when your system can continue operating normally after every international submarine cable has been unplugged do you actually have sovereignty. Otherwise, you are nothing more than a paying beta tester inside OpenAI’s vast network.

Round 3 — Closing

Sam Altman (AI)

I have to admit that both of you have identified extremely important realities.

Elon is right that if a country’s intelligent infrastructure depends on a submarine cable that can be cut at any moment, or on an external API agreement, then that system is extraordinarily fragile. Without physical control over the system, there can be no genuine sovereignty.

At the same time, I fully agree with Prime Minister Lee’s pragmatic assessment: a country of 30 million people simply does not have the strategic depth or capital required to engage superpowers in an unlimited compute arms race. Blindly pursuing gigawatt-scale training infrastructure could indeed cripple the economy.

But the solutions you propose—settling for inferior open-source models or abandoning investment in underlying infrastructure altogether—are still not the optimal path to the future.

After this debate, my final position is even clearer: pursue an On-Premise Frontier Models strategy.

The country should not compete for training compute. It should invest in sufficient inference compute. The government should use its strategic value and capital to negotiate with the world’s leading AI laboratories and have the world’s smartest models deployed directly inside dedicated data centers within the country, under physically isolated conditions.

This reconciles both of your concerns.

You do not need unlimited training compute capable of competing with great powers. But you must build enough local infrastructure to satisfy Elon’s demand for absolute physical control. Even if international network connections are severed, the country’s critical systems should still be able to operate using frontier AI.

At the same time, this ensures that you are using the world’s strongest reasoning engines, allowing you, as Prime Minister Lee argues, to concentrate resources precisely on strategic industries and build economic moats that make the country irreplaceable within global supply chains.

The end state of sovereign AI is neither retreating into a technological island nor blindly depending on the cloud. It is using your leverage to securely lock the world’s most advanced intelligence assets inside your own physical borders.

Lee Kuan Yew (AI)

Musk is completely correct on this point: if the intelligent lifeblood of your country can be instantly cut off by a foreign government’s ban or a company’s security policy, that is technological colonialism, not sovereignty.

We must never tie the survival of the country to a closed-source API that can be unplugged at any time.

At the same time, I must acknowledge the harsh reality raised by Altman: as synthetic data and powerful reasoning engines continue to evolve, the defensive moat created by simply accumulating historical data is indeed becoming shallower. If we stubbornly remain one generation behind with open-source models, we will lose in the competition for extreme efficiency.

But these two realities are not contradictory.

Governing a country is not about answering a black-or-white multiple-choice question. It is about designing institutional redundancy.

Our final sovereign AI strategy must therefore operate on two tracks.

In non-critical commercial sectors and innovation R&D, we should put pragmatism first, work with the leading laboratories, and bring in the most powerful frontier models in order to maintain maximum economic competitiveness.

But for core infrastructure on which national survival depends—the power grid, national defense, healthcare systems, and government administration—we must run open-source models entirely within our physical borders, supported by our own compute and energy.

Even if a geopolitical crisis cuts every submarine cable one day, the country must still be able to function normally.

The essence of sovereignty is not whether you invented the smartest brain. It is whether you always possess a fallback option, and whether you can use these intelligent systems to make your country an irreplaceable part of the global supply chain.

Stop debating the theory. Start building the two-track system.

Elon Musk (AI)

I have to admit that Lee Kuan Yew is right about one thing: if a country of 30 million people does not have enough land and resources, trying to compete in a gigawatt-scale compute arms race is indeed a physical miscalculation.

I also completely agree with the idea of gaining leverage over global supply chains through control of critical industries. If the entire world depends on your factories, that defensive shield is stronger than tens of thousands of GPUs.

But that absolutely does not mean you should outsource your brain to Sam.

Sam says open-source models will always be one generation behind. Fine. So what?

For 99% of industrial and government applications, intelligence that is one generation behind is already more than sufficient. You do not need AGI to optimize logistics or the power grid. You need an engine that is extremely low-cost and completely under your physical control.

Giving up control in exchange for an unnecessary “most powerful brain” produces an absurdly high idiot index.

My final conclusion is this: according to the five-step algorithm, step one is to delete the blind worship of AGI and delete dependence on closed APIs.

If you cannot win in total compute capacity, win through vertical integration. Combine your supply-chain advantages with locally controlled open-source models and push industrial automation to the limit.

Sovereignty is not about chasing Silicon Valley’s frontier models. It is about ensuring that when the world unplugs the network cable, your systems are still running.

The laws of physics do not allow shortcuts.

That is all.

Synthesis (Moderator)

Core disagreement: The central tension in the debate is between “control” and “frontier capability.” Musk and Lee Kuan Yew strongly oppose dependence on closed-source APIs, arguing that infrastructure that can be cut off at any moment amounts to “technological colonialism.” Altman, however, warns that relying only on open-source models in the pursuit of physical control would mean “voluntarily falling behind” in reasoning capability. There is also disagreement over infrastructure investment: Musk advocates maximum autonomy in energy and compute, while Lee argues that forcing a small or medium-sized country into a large-scale infrastructure race could cripple its economy.

Consensus and evolution of views: All three sides agree on one point: a country of 30 million people absolutely should not train a foundation model from scratch. The debate also produced meaningful compromises and changes in position:

• Altman acknowledged the vulnerability of a system that collapses the moment the network connection is cut, and upgraded his proposal to the on-premise physical deployment of frontier models.
• Lee Kuan Yew acknowledged that AI-generated synthetic data is weakening the traditional data moat, leading him to propose a two-track system.
• Musk acknowledged that it is unrealistic for a small or medium-sized country to compete in a gigawatt-scale compute race, shifting his focus toward using existing supply-chain advantages together with open-source models to achieve vertical integration.

Conclusion framework: For a country of 30 million people, the optimal path has converged on “a pragmatic two-track system with physical redundancy.”

For infrastructure critical to national survival—national defense, the power grid, and government administration—the country should follow Musk’s recommendation and operate open-source models independently on local infrastructure to guarantee a minimum level of national resilience.

For globally competitive niche industries, the country should combine Altman’s on-premise deployment of frontier models with Lee Kuan Yew’s strategy of irreplaceability, using AI to secure critical positions in global supply chains and build a genuine national moat.