AI's hidden demand: why mature-node chips matter more than ever
AI is usually told as a leading-edge story. Much of what makes AI work runs on mature-node power, control and connectivity chips, the segment India is building first.
Most of what is written about AI and chips concerns the leading edge: 3 nm accelerators, EUV lithography, advanced packaging and high-bandwidth memory. That focus is understandable. But AI also pulls hard on a part of the industry that gets far less attention, and it is the part India is building first.
AI runs on power, and power runs on mature-node chips
The International Energy Agency projects that electricity demand from data centres will more than double by 2030, to around 945 TWh, with AI the most important driver. Every watt that reaches an AI accelerator passes through a chain of conversion and regulation: from the grid to the rack, from the rack to the board, and from the board to the processor at very low voltage and very high current.
The chips that do this work are power devices, power management ICs, voltage regulators, gate drivers, current sensors and monitoring controllers. Almost none of them are made at the leading edge. They are made on mature silicon nodes, often 40 nm to 180 nm, and increasingly in silicon carbide and gallium nitride.
The rest of the AI system
Power is the clearest example, but not the only one.
- Connectivity and control. Servers, switches and storage contain many microcontrollers, interface chips, clock generators and sensors, mostly on mature nodes.
- Cooling and facilities. Liquid cooling, pumps, fans, chillers and building systems rely on motor controllers and power electronics.
- Electrical infrastructure. New data centres need grid connections, transformers, switchgear and backup systems, all of which use power semiconductors.
- AI at the edge. As AI features spread into cars, phones, appliances and factories, each device needs its own power, sensing and control chips alongside any AI processor.
For every leading-edge AI chip, there are many mature-node chips that make it work.
Why this matters for India
India's first fab, built by Tata Electronics with PSMC at Dholera, is planned on 28 nm to 110 nm nodes for power management ICs, display drivers, microcontrollers and computing logic. Its OSATs package and test a wide range of devices. This is sometimes described as a catch-up position. We see it differently: it places India in a segment where AI is adding demand on top of the existing growth from electric vehicles, renewable energy and electronics manufacturing.
Mature-node manufacturing has its own challenges. Competition is intense, margins are thinner, and customers expect long product lifetimes and very high reliability. Success depends on cost, uptime and yield more than on technology leadership.
What it takes to compete
- Operational excellence. Fast yield ramps, high tool availability and disciplined cost control decide who makes money at mature nodes.
- Specialty processes. Power, high-voltage and analog processes, and compound semiconductors such as SiC and GaN, earn better margins than commodity logic.
- Local supply. Mature-node economics are sensitive to the cost and reliability of materials and services, which makes the local ecosystem even more important.
- Customer proximity. India's own automotive, energy and electronics industries are natural anchor customers, and qualification is easier when the fab is close to them.
Our view
AI is not only a leading-edge story. It is a power story, an infrastructure story and a systems story, and much of that runs on the kind of chips India is building capacity to make. The opportunity is real. Capturing it depends on operating these fabs as well as the best in the world.
Building something in India's chip industry?
Tell us what you are working on. We reply within two working days.