Silicon froth: AI chips are riding a massive global wave that could turn without warning

First, the narrative is moving from a speculative buzz around AI models to hard spending on the infrastructure that powers them. Second, big money is flowing into computing power and memory, the choke points of the AI economy.

For India, these shifts augur well. The country is building 12 semiconductor plants (including a full-fledged fab facility), pursuing 2 nanometre and 3nm chip-design alliances and subsidizing AI chips for startups working on Indian language and voice models. India’s AI Kosh repository hosts nearly 10,500 local datasets and about 300 models across 20 sectors, forming a data backbone linked to Aadhaar, UPI, et al.

Progress on chip design is tangible. Various institutes have used government-backed tools to design and produce roughly 150 chips at the 180nm node. New Delhi has cleared 24 semiconductor and system-on-chip design projects across strategic areas—from video surveillance and drone detection to energy metering, microprocessors, satcom services, broadband and internet-of-things.

Some 14 firms have raised over ₹650 crore in venture funds to scale these efforts, while work is on for chip fabrication at more advanced nodes (such as 12nm). India also has supply chain links with the US, EU, Japan, Singapore and the Netherlands.

That is a lot of action. Chip advances are a must for success in electronics overall, not just AI. However, this is not a field for the faint of heart, given how fast its dynamics can shift.

Chip fabrication is capital-intensive, requiring time and scale, apart from technical depth and execution skills. Global leaders like TSMC took decades to achieve dominance. India’s fabs, many of which are being built with foreign partners, should be ready for a long slog too.

How demand and supply match is never easy to foresee. Training and deploying large AI models locally would require enormous processing power and high-bandwidth memory, but the current focus of Indian fabs is on 28-120nm chips that largely serve the automotive, electronics and telecom sectors.

The country has a little over 150 data centres, of which only about 11 cater to AI workloads; cutting-edge AI would call for advanced nodes, specialized accelerators and tightly integrated ecosystems for which they will have to rely on inputs from firms like ASML Holding, Synopsys and Cadence Design Systems.

Besides, fabs are only the base of an AI ecosystem. How well such a base is ultimately used depends on cloud infrastructure, talent, power supply, enterprise adoption and the viability of business models.

Will demand patterns evolve to suit today’s chip thrust? Globally, the surge in AI capital expenditure rests on expectations that may or may not materialize. If they do, India could snap into global supply chains while supporting domestic ambitions.

However, if an AI boom goes bust, capital-intensive parts of the stack would be early casualties; chipmakers that lack buffers of steady demand would suffer. Indian players with diversified bets would fare better against volatility.

As a strategy, an approach that’s hedged for both outcomes would suit us best. In all, our fab pursuits must go alongside rapid chip adoption across sectors, even as we strengthen India’s digital infrastructure, augment our energy capacity and import the top-end chips we dearly need—even if they are subject to a regime of trade nods and quotas.