It's heartening to see wisps of the nascent Indian tech community waking up to the long-term strategic significance of AI capabilities, and the Indian government striving to offer support. I wish the very best to those diving in, and for the rest of this discussion will assume they succeed. What then?
No technology can be sustained on an island of excellence. It really matters what sits below and above in the stack; to push capabilities in one layer you are often enabled only by pushing capabilities in the layer above/below. Otherwise the impedance mismatch causes too much friction (forgive the mixed metaphor). I can offer several examples to buttress the point if need be (cite: DeepSeek/OpenAI, Alan Kay, Joel Spolsky, Apple vertical integration, etc). What does this mean for Indian AI?
Consider the stack below. AI runs atop computers, and one needs to have a good handle on the hardware+software for CPUs/GPUs, networking, datacenters. Progress in AI can easily be knee-capped, even in the short-term, by puncturing or denying developments in any of these components. These turn out to be crucial for modern information/communication technology more broadly -- and it is therefore imperative to build ownership over all of these.
Consider the stack above. Nandan Nilekani is much maligned for advocating a focus on applications of AI technology -- to make India the "use case capital". While this must not be at the cost of technology capabilities, having a robust application ecosystem is necessary for several reasons:
- For actually deriving benefits from the technology. The goal isn't to build technology for its own sake.
- Use-cases then feed back into shaping the underlying technology and its development (eg: AL capabilities are driven for search/retrieval, conversational interfaces, code-generation, controllable video generation, etc). A momentary spark of excellence will otherwise decay in place as the ecosystem around it shifts and adapts.
- Healthy monetization helps ensure that enough capital can be invested into technology development, longer term. It is crucial however, to ensure ownership of the lower layers of the stack -- otherwise any revenue generated will be sucked out by monopoly powers in lower layers (eg: Nvidia, AI capability providers, etc).
In the above, I just sliced out one column to tell the story of the benefits of vertical integration with AI as an example; the same principle applies equally well to all the other technology components in the DAG of dependencies -- be they parents, children or siblings of the AI node. We do have promising developments on some of these components, and it is worth discussing the larger endeavor because it requires sustained and sincere diligence -- in both technology building and judicious capital investment.