This sector has an unusual dual anchor structure, and it is a tighter, more concentrated version of the dual structure seen in other India industrial primers. On one side, a single private company — Nvidia — sets the technical qualification bar, the architecture cadence, and, functionally, who is allowed to build with its chips at all; no Indian policy instrument can accelerate a certification cycle or a chip allocation decision made outside the country. On the other side, the Government of India runs three concurrent, largely independent instruments that between them shape almost everything domestic demand and domestic manufacturing capability look like. Both matter; neither substitutes for the other.
Nvidia's data-centre segment revenue reached $75.2bn in the quarter ended 26 April 2026 (Q1 FY27), up 92% year-on-year, driven by the Blackwell architecture ramp — the clearest evidence available that demand for the chips themselves remains the binding constraint on this entire industry, not manufacturing capacity anywhere in the chain. At the same time, Nvidia is reported to be narrowing, not widening, the field of manufacturing partners who get to build its newest rack-scale systems, concentrating assembly of its upcoming Vera Rubin racks among a small set of established Taiwanese ODMs and raising the proportion of "customisation not allowed" components in its own reference designs generation over generation (§1). For an India-listed AI-server manufacturer, this means the qualification bar is not fixed — it is actively rising, on a timetable Nvidia alone controls.
India Semiconductor Mission 2.0 pays for capability, not demand. Its $13.5bn outlay incentivises the construction of fabs, compound-semiconductor lines and OSAT/packaging facilities; it does not itself guarantee that global fabless chip designers route work through them, or that the chips those facilities package are the ones going into India's own AI servers. A reader who assumes ISM 2.0 capacity automatically becomes India's own AI-chip supply is reading the scheme more generously than it is designed.
The IndiaAI Mission pays for access, not ownership. Its GPU procurement subsidises compute access for startups, researchers and government users at a fixed hourly rate; it does not itself build India's server-manufacturing capability, though the servers it deploys may well be assembled by India-listed companies covered later in this report. Treat the 100,000-GPU 2026 target as a demand-side figure, not a supply-side one.
IT Hardware PLI 2.0 pays for output, not R&D. Its 5%-of-incremental-sales incentive rewards manufacturing volume of in-scope hardware, including servers, made in India — it says nothing about whether that hardware embeds any India-developed intellectual property, which is a separate, harder question this report returns to in §8.
Taken together, the three schemes are complementary but not additive: ISM 2.0 builds packaging and fabrication capability, IndiaAI Mission creates a captive pool of domestic GPU demand, and PLI 2.0 subsidises the assembly step in between — but no single instrument, and no simple sum of the three, tells a reader how much real, IP-bearing value India is capturing versus how much is pass-through assembly and packaging volume. That distinction is exactly what the company reports in this document are built to surface.
