On October 23, in Astana. At Kazakhstan's national artificial intelligence center, NVIDIA chief executive Jensen Huang sat down for a fireside chat with Deputy Prime Minister Zhaslan Madiyev, who also heads the country's ministry of artificial intelligence and digital development. The conversation was titled "AI Nation."
He began with a small point. Kazakhstan's president had designated the entire country an "AI nation" and, by his account, it was the first country he had visited where a president had made such a declaration. Only then did he turn to how a country might convert what lies underground into something else.
A Cake With Five Layers
From an industrial point of view, he said, AI is a five-layer cake. At the very bottom is energy, then chips and computing infrastructure, then cloud services, then models, and only at the top, applications.
Most people think AI is the fourth layer, the models that write essays and answer questions. He does not see it that way. In his account, that layer rests on the four beneath it, and remove one, and the whole cake cannot stand.
The metaphor is not new; what is new is where he used it. Kazakhstan has sold oil, uranium and coal for decades. Now someone has come to tell it that the same things under its feet, sold differently, might fetch a very different price.
Energy In, Intelligence Out
In the future, he said, you will not only export energy; you can turn energy into intelligence and export intelligence. That sentence is the bone of the whole conversation; everything else grows from it and circles back to it.
He also said AI is manufacturing. It cannot be computed in advance and stored; it has to be generated on the spot every time it is used, and that requires a new kind of factory. He calls it an AI factory, and what NVIDIA sells is the machinery that feeds it.
The analogy is a little convoluted, but the point is not hard. A country that sold electricity used to lose it the moment it left; now the power goes into a server hall and what comes out can be sold abroad, and the machines are still there the next day.
Is Data a Natural Resource?
Someone asked whether Kazakhstan should build its own large model, one that understands Kazakh. He said it absolutely should. He went further and called it a national security question, not a technical one and not a matter of pride.
His reasoning: a country's data carries its culture, its history and its values. The data belongs to the country the way oil belongs to the country. To ship the data out and buy someone else's intelligence back, he argued, does not add up.
He offered an analogy: you have your own power grid, your own telecoms, and in the future your own intelligence grid. Others can supplement it, but they cannot become the only pipe. The argument runs on the same logic as the five-layer cake.
Beyond Robots, Mines
On physical AI, he first drew the boundary: the term does not mean robots. Robots will matter, but they are one piece; the larger part is in mines, underground, in the pipes of factories, in places no one likes to look at twice.
His examples were seismic understanding, mapping mines and wells, geological judgment, materials science, chemical processing. From extraction and ore dressing to turning ore into new materials, every stage can use it — and what is used is often not a walking humanoid but a system that can see and calculate.
One Investment and One Number
The conversation also touched on an investment. NVIDIA put money into Higgsfield, a generative video company headquartered in San Francisco with most of its engineering team in Kazakhstan. His explanation: the first thing and the last thing is that you invest in people, not companies.
Then came the number he gave. The total addressable market for AI, he said, exceeds 50 trillion dollars. That is his own figure, not a third-party estimate. He added that which layer captures the most value is a real question, but all five have to move forward together.
On agents, he raised a practical problem. These programs can read files and call tools; hand them the key and they will open the door themselves. So NVIDIA built something called OpenShell, which holds their permissions at exactly what the job requires.
That sounds remote from ordinary life, but it is not. The assistant on your phone and the support bot at your company run into the same question: how much can it do for you, and how much should it be allowed to touch. How good the lock is decides whether you dare hand over the key.
The Threshold Is Called Ambition
Someone asked about jobs. It is the question every such conversation gets, and the hardest to answer. His answer did not land on a job title. It landed on a word: ambition. The only thing that matters, he said, is whether you have it.
If you have ambition, he said, machines will lift you and let you do more; if you are a farmer with ambition, machines help you farm and you can expand the operation. Without ambition, the same machine is something else: it just lets you do less.
It sounds like a platitude, but it is a conditional. He did not say AI will make everyone better off. He said it amplifies what is already there — the good, and the bad as well.
Someone asked what he would study if he were 18 today. He said he would train himself to be the best AI programmer in the world. Not the best programmer: the best AI programmer. One word sits in between.
His reason: people used to learn how to give a computer instructions; now they learn how to work with something that thinks for itself. The thresholds are different. The first has textbooks; the second does not have a finished one yet.
A Cake on Paper
The same day, NVIDIA's blog carried a different set of facts: Kazakhtelecom has already installed 400 H200 and 480 B300 chips in Astana and Almaty, and plans more than 1,000 more in the Data Center Valley next year. These are things that have happened, not frosting on the cake.
According to the official blog, a cloud provider called Firebird has obtained an export licence for 45,000 chips; the first phase will not be commercially operational until the third quarter of next year, and only after the second phase could installed capacity reach 500 megawatts. The numbers stack up, and every layer needs power.
The Data Center Valley is planned at gigawatt scale. A gigawatt is a unit of power; behind a high-performance chip running year-round sits an entire system of power supply, cooling and maintenance. Where the power comes from, where the people come from, where the customers come from — all of that has to be answered first.
What the conversation left this country was an assessment and several contracts. Whether the cake can actually be baked depends on how fast the meter spins in the years ahead, and on whether anyone is willing to keep standing by the oven.
Disclaimer: This article is based on publicly available information and is for reference only. It does not constitute investment advice.