Key facts
- Jefferies says cheaper AI models will fuel, not reduce, infrastructure investment
- Memory chip makers identified as key beneficiaries of the AI infrastructure boom
- Rising AI inference volumes expected to drive demand for high-bandwidth memory
Investment bank Jefferies has published an analysis arguing that cheaper AI models will be a catalyst for a new phase of infrastructure expansion rather than a headwind for the hardware industry, with memory chip makers positioned as standout beneficiaries of the trend.
The argument inverts a common assumption in technology markets: that falling AI model costs would reduce demand for compute and hardware. Jefferies contends the opposite — that as AI inference becomes more affordable, the volume of AI workloads will explode, requiring vastly more infrastructure to support them. This dynamic, sometimes called the Jevons paradox of AI, suggests that efficiency gains in AI models translate into greater, not lesser, total resource consumption.
Memory chips are singled out as a particular area of opportunity. As AI models are deployed at massive scale for inference — the process of running a trained model to generate outputs — the bottleneck increasingly shifts from raw compute to memory bandwidth and capacity. High-bandwidth memory, in particular, is seen as a critical constraint and therefore a growth area as cheaper AI drives higher volumes.
For India, which is actively seeking to build semiconductor and electronics manufacturing capacity, the Jefferies analysis offers a useful pointer: the AI hardware opportunity is not confined to the cutting-edge chip fabrication that remains concentrated in Taiwan and South Korea, but extends meaningfully into memory and supporting components — areas where India's manufacturing ambitions are more immediately realisable.