Vishwamitra Arrow
Issue No #002
I keep hearing the same claim from governments and think tanks: we must achieve AI sovereignty or fall behind. I think that claim is wrong, and the evidence already shows why.
By Santhosh Vishwamitra
Opinion and Analysis
Every few months another country announces a plan to control its own artificial intelligence, Domestic chips, Local models, Export bans or Talent walls. The language is always the same. Sovereignty or Independence or No reliance on rivals.
I have read the Brookings analysis that calls full-stack AI sovereignty structurally infeasible for almost any country. I have read the industry reports that list the new bottlenecks forming right now in design software, advanced packaging, and power. The more I look, the clearer it becomes that the race for complete control is built on a false picture of how the technology actually works.
AI is not a single product you can fence off. It is a stack. Minerals for the chips. Energy to run the data centres. The specialised machines that make the chips. The software that designs them. The models trained on global data. The applications that sit on top. The people who know how to keep the whole thing moving. No single nation owns every layer. The United States does not. China does not. Europe does not. India does not. That is not a temporary gap that more money will close. It is the structure of the technology itself.
Look at the semiconductor layer alone. Industry executives surveyed by KPMG in the latest outlook say they have almost never been more confident about sales driven by AI demand. At the same time they rank tariffs and trade policy as their top concern for the first time. Nearly half put supply-chain flexibility as their number one strategic priority. Deloitte’s 2026 outlook points to new chokepoints that did not exist a few years ago: electronic design automation software, gate-all-around transistors, advanced packaging. A critical technology market worth tens of billions sits underneath a chip market measured in hundreds of billions. That leverage does not disappear because a government writes a sovereignty plan.
China has restricted exports of gallium, germanium and other materials needed for chips. The United States has restricted advanced equipment and high-end chips. Both sides claim they are protecting national security. Both sides are also discovering that the other still finds ways to keep the stack running. Parallel systems are forming, but neither is complete. As Brookings and other stack analyses have put it: no country controls every layer. The cake is too big and the ingredients come from too many places.
I am not arguing that nations should stop building capability. Domestic production of chips, models and talent matters. What I am arguing is that treating full sovereignty as the goal leads to expensive dead ends. It pushes policy toward walls when the real advantage lies in managed connections. Middle powers especially cannot match the scale of the two largest players. Trying to do so alone wastes resources that could be spent on the layers where they can still shape outcomes.
For a country like India the question is not whether it can build a complete AI stack from mines to models. It cannot. The question is which parts of the stack it can influence, which dependencies it can diversify, and which alliances give it room to manoeuvre when one supplier becomes unreliable. The same logic applies to Europe, Japan, South Korea and others that sit between the two poles. Agency does not require isolation. It requires the ability to choose and re-choose partners as conditions change.
The obvious reading is that countries must go it alone. I think we are missing the more useful truth: interdependence is not weakness when it is managed deliberately.
Some will say export controls still slow a rival even if they cannot create pure sovereignty. That is true in the short term. Controls can raise costs and delay specific capabilities. They cannot erase the fact that the underlying knowledge, materials and people continue to move through other channels. The industry already plans around them. New bottlenecks appear in places that were never controlled before. Confidence in demand sits next to rising fragility in supply. That combination is not a temporary phase. It is the new normal of a technology that has outgrown any single border.
I have no interest in soft language about collaboration for its own sake. Power still matters. Access still matters. The countries that treat the AI stack as a shared system they can navigate will keep more real options than the ones that keep promising total independence they cannot deliver. The public statements will continue to speak of sovereignty. The private adjustments will continue to show interdependence. Readers who only follow the statements will keep being surprised. The ones who watch the stack itself will not.
That is the view I hold after looking at the reports and the supply-chain realities. Full AI sovereignty is a slogan that collapses under the weight of the technology. What remains possible is harder and less dramatic: building strength in specific layers, diversifying the rest, and refusing to confuse production numbers or export bans with actual control. That is the only path that still leaves room for choice.
Santhosh Vishwamitra
Vishwamitra Chronicles
Sources & Reference Material
This is an opinion column. Evidence is cited inline in the body of the piece, not listed here. Readers who wish to verify any factual claim can consult the cited material directly.
Follow The Dispatch
Vishwamitra Chronicles Network
Original reporting on science, geopolitics, and history.
Comments
Post a Comment