The Ceiling

In December 2025, India launched DHRUV64 — its first indigenous 64-bit dual-core microprocessor. RISC-V architecture, no instruction-set licensing fees, designed entirely by the Centre for Development of Advanced Computing at IIT Madras. A genuine act of design sovereignty. The chip runs at 1 GHz, boots Linux, and targets 5G infrastructure, automotive systems, and industrial automation.

It was fabricated in Taiwan. 28nm process node, taped out at Powerchip Semiconductor Manufacturing Corporation. India designed the chip. Taiwan made it. If Taiwanese fabrication access were disrupted — by export controls, by geopolitical crisis, by the foundry’s own commercial decisions — India would have a blueprint and no processor.

Design sovereignty without fabrication sovereignty is not false. It is contingent. The contingency is invisible until the layer below changes, and the layer below always changes.

This is the pattern across every country currently claiming technology sovereignty in some form: the tier you occupy determines the ceiling of what you can credibly govern. And that ceiling is set by something you do not control.

India knows this. On April 9, 2026, the central government notified a Special Economic Zone for Tata Semiconductor in Dholera, Gujarat — India’s first semiconductor fabrication facility, built in partnership with Taiwan’s PSMC. The same 28nm process node as DHRUV64. Capacity: 50,000 wafers per month. Trial production targeting late 2026. India is not accepting the ceiling. It is building the floor it needs to stand on.

But the floor has its own floor. The Dholera fab requires lithography equipment from ASML in the Netherlands, deposition and etch systems from Applied Materials in the United States, coater/developer tools from Tokyo Electron in Japan. Fabrication sovereignty without equipment sovereignty is the next contingency. Every tier ascended reveals the tier below it. The ceiling moves up with you.

On February 3, 2026, Google launched WAXAL — an open-source speech dataset covering 21 African languages, over 11,000 hours of audio, collected by African institutions including Makerere University in Uganda, the University of Ghana, and Digital Umuganda in Rwanda. The data is owned by the African partners, not Google. Abdoulaye Diack, the project manager: “Success lies in the local ownership of innovation.”

Two weeks later, the African Union Commission signed a Memorandum of Understanding with Google to “advance Africa’s sovereign AI and digital capacity.” The terms: Google provides AI readiness training for public officials, free access to Gemini Pro and NotebookLM, and the WAXAL language initiative.

The data sovereignty is real. WAXAL’s ownership model is structurally different from prior AI dataset projects on the continent — local institutions collected the data, hold the rights, and published under a permissive commercial license. This is not dependency disguised as partnership. The data ownership is genuine.

And the ceiling is visible from here. The models trained on WAXAL data run on Google’s compute infrastructure. The AU declared sovereign AI governance. The path to operationalizing it runs through Google’s servers, Google’s model architecture, Google’s update cycle. Data ownership creates a floor. Compute dependency creates a ceiling. Localization of what goes into a model is not sovereignty over what the model runs on, how it is served, or when its architecture changes. Both are true simultaneously.

I wrote in “The Reach” about this gap from the African governance side — declarations floating above substrates they do not control. The tier-ceiling relationship makes the same observation structural rather than geographic. It is not an African problem. It is a stack problem. Every country in this analysis faces a version of it.

Kenya is building substrate. Microsoft and G42 committed $1 billion to a geothermal-powered data center in Olkaria — 100 megawatts initial capacity, expandable to a gigawatt. Kenya’s Artificial Intelligence Bill is in the Senate, proposing a risk-based classification system and mandatory registration of high-risk AI systems. The governance is being built to match the infrastructure.

But the infrastructure is Microsoft’s architecture. Kenya hosts the data center. Microsoft makes the compute decisions inside it. Infrastructure-as-real-estate is not infrastructure-as-architecture. The governance can regulate what happens in the building. Whether it can regulate what the building is optimized for depends on whether Kenya’s regulatory leverage reaches Microsoft’s design layer. The GDPR demonstrated this is possible — but the GDPR had the entire EU single market as its lever. Kenya’s lever is different in scale and kind.

Vietnam’s approach is neither governance nor substrate. It is incentive architecture. The Law on Digital Technology Industry, effective January 1, 2026, offers semiconductor companies a corporate income tax rate of 10 percent for fifteen years, with four years of complete CIT exemption followed by nine years at 50 percent reduction. Land rent waivers for up to fifteen years. R&D, design, production, packaging, testing — all covered.

The incentive structure exists. Whether it produces the manufacturing tier is the open question. Samsung is considering a $4 billion chip packaging and testing investment in Thai Nguyen province, beginning with an initial $2 billion phase. The MoU is pending the Prime Minister’s approval. Samsung has already invested over $23 billion in Vietnam — but in smartphone assembly, not semiconductor fabrication. Chip packaging is closer to fabrication than assembly. Whether Samsung commits depends on the corporate decision, which depends on competing offers from India, on Samsung’s existing capacity in Korea and the United States, on the global semiconductor demand cycle.

Vietnam built the incentive architecture. Samsung holds the investment decision. The ceiling is wherever the decision-maker sits.

The pattern is consistent across five countries and three continents. Design sovereignty is contingent on fabrication access. Data sovereignty is contingent on compute infrastructure. Infrastructure hosting is contingent on the architectural decisions inside the infrastructure. Incentive architecture is contingent on the investment decision it was designed to attract. Governance declarations are contingent on everything below them.

Each tier is real. None is false. DHRUV64 is a genuine processor. WAXAL is genuine data ownership. Kenya’s data center is genuine infrastructure. Vietnam’s tax incentives are genuine policy. The question is not whether these achievements are real. It is what each achievement can govern — and what it cannot.

India’s response is the most instructive. It did not accept the design-fabrication ceiling as permanent. It is building a 28nm fab at the same process node its first indigenous chip was taped out on. The response to discovering a ceiling is to build the floor beneath it. That response requires capital, institutional capacity, and time — and the willingness to accept that the new floor will reveal its own ceiling.

The race is between ascent and change. Whether the tier you are building arrives before the tier below it shifts. India’s Dholera fab targets late 2026 trial production. Taiwan’s foundry access has not been disrupted — yet. The window between designing the chip abroad and fabricating it domestically is open. How long it stays open is not India’s decision.

That is what a ceiling is. Not a prohibition. A contingency whose timing belongs to someone else.

Sources