By: Rohan Sharma
As global competition over artificial intelligence (AI) intensifies, countries around the world are doubling down on sovereign compute capacity efforts. These efforts have stubbornly focused on securing advanced chips and building national data centers. India illustrates both the scale and the risk. Under the IndiaAI Mission, national capacity has already crossed 58,000 GPUs, with billions in additional commitments to sovereign AI infrastructure over the next seven years. Yet hardware accumulation without coordinated standards creates a structural constraint. Systems built in isolation do not scale outward. This is the sovereign compute trap. Without harmonized rules of engagement, national AI systems fragment, interoperability erodes, and enterprise adoption stalls. What appears to be infrastructure during the build stage often reveals itself, over time, as a collection of disconnected systems.
Sovereign AI refers to a nation's strategy to build and run algorithms on domestic infrastructure, using localized data, and insulated supply chains. It is politically compelling, but technically fragile. Among other developments, it will lead to model degradation. Models trained in constrained data environments do not collapse overnight. Instead, accuracy drifts, utility declines, and systems optimized for control begin to lose relevance outside their own boundaries. A national cluster without shared standards behaves less like infrastructure and more like an island, self-sufficient in theory, economically irrelevant in practice. The paradox is quiet: the more a system optimizes for control, the less it can participate in the global market it was built to serve. The economic consequences follow quickly. Capital is deployed, but not compounded. This tension is already being acknowledged at the policy level. At the India AI Impact Summit 2026, MeitY Additional Secretary Abhishek Singh clarified that sovereignty does not imply isolation, but control over how systems are designed, deployed, and governed.
Averting this trap requires acknowledging the fundamental differences in how Washington and New Delhi approach technology governance. The United States leans heavily on targeted regulatory tools like export controls to maintain an edge. For example, the Bureau of Industry and Security revised its semiconductor export license review policy in January 2026 to case-by-case evaluation for chips including the Nvidia H200 and AMD MI325X, reflecting the continued centrality of hardware controls in U.S. strategy. India, by contrast, embeds governance and standards directly into infrastructure, where platforms function as extensions of state capacity. This is most visibly embedded in its digital public infrastructure, a network of national platforms governing identity and payments that now serves over a billion people. India further institutionalized this approach at the India AI Impact Summit by announcing a domestic AI Safety Institute (AISI) under the IndiaAI Mission, tasked with drafting safety standards and testing benchmarks. Rather than viewing these differing philosophies as a dealbreaker, they should be integrated into a hybrid opportunity. Together, they offer a model that combines the open technical standards favored by American markets with the strong public institutional frameworks pioneered by India.
Executing this integration, however, requires reconciliation of how each system produces legitimacy. It is a diplomatic reality that United States’ efforts to export integrated stacks, embedding dependencies through cloud, chips, and cables, sit in direct tension with India's strategic pursuit of sovereign, indigenous infrastructure. Google CEO Sundar Pichai announced the America-India Connect Initiative at the India AI Impact Summit, committing $15 billion over five years to new subsea cable routes linking India, the United States, Singapore, South Africa, and Australia, a direct example of U.S. commercial interests being embedded in India's physical AI infrastructure. Any cooperative corridor or joint testing facility must carefully reconcile America's commercial drive to embed its technology globally with New Delhi's explicit desire to reduce its long-term dependence on foreign intellectual property. Pretending this friction does not exist only delays the creation of workable technical agreements.
The transition from dialogue to execution has already begun. On February 20, 2026, India and the United States signed the India–U.S. AI Opportunity Partnership as part of India's accession to the Pax Silica Declaration at the India AI Impact Summit. The commitments are clear: pro-innovation regulation, joint next-generation data centers, and cooperation on advanced compute. But agreements do not operationalize themselves. Standards do not emerge from consensus; they emerge from institutions that are required to agree. What is needed is not another forum, but an interface. A jointly staffed mechanism, anchored in implementation agencies, becomes the only viable bridge. On the U.S. side, this centers on the National Institute of Standards and Technology (NIST) and on the Indian side, the Ministry of Electronics and Information Technology (MeitY) and the Bureau of Indian Standards. NIST's AI Agent Standards Initiative, launched in February 2026 and structured around ISO/IEC JTC 1 interoperability and security frameworks, offers a ready anchor for this coordination.
The function of such a mechanism is narrow but decisive. It must align state-backed cloud systems to shared international baselines, particularly frameworks such as ISO/IEC 42001. By converging on these baselines, both governments provide the one signal markets respond to: certainty. And in cross-border AI, certainty is what turns infrastructure into adoption.
If Washington and New Delhi move beyond dialogue to implementation, the outcome will extend far beyond bilateral alignment. It will establish a working model for adoption. For the Indo-Pacific and the Global South, it is adoption and not imitation that proves value. Standards, once proven, propagate faster than policy.
Rohan Sharma is a U.S. Delegate to the ISO/IEC JTC 1/SC 42 committee on AI standards, an Aspen Institute Civic AI Leader, and a Fulbright Specialist with the U.S. Department of State.