Micron Technology, one of the world’s leading suppliers of high-bandwidth memory used in datacentre accelerators for artificial intelligence, has seen a marked pullback in its stock even as demand for its products remains intense. The contrast highlights widening concern among investors and industry players about whether the current wave of AI-related spending can be monetised without crippling operating costs.
Price, demand and a wobbling share price
The company recently experienced a sharp correction of around 32% from a June peak, though over a twelve-month window the shares have still risen by nearly 700%. That divergence — spectacular gains followed by a steep fall — has prompted renewed debate over how sustainable spending on AI infrastructure will be as component costs climb.
Memory is central to modern AI workloads. High-bandwidth memory (HBM) increases the speed at which processors can access data, a critical advantage for large-scale machine learning. Shortages in supply, combined with booming demand from hyperscale datacentres, have given suppliers pricing power. But steep component prices are beginning to influence how organisations use AI.
“Building a single gigawatt worth of capacity requires $50 billion worth of capital investment,”
The line above, attributed to Nvidia’s chief executive, has been used in forecasts estimating the scale of capital spending required to meet anticipated demand. A Bloomberg projection cited in market commentary envisages roughly 118 gigawatts of data-centre capacity in the US by 2030 to support the AI boom — a figure that, if realised, implies massive industry investment.
Economic reality checks
Those headline numbers are useful for perspective but underline a key tension: the higher the cost of the underlying hardware, the harder it becomes for firms to justify running expensive AI workloads unless they can generate sufficient revenue from them.
Several companies are already reported to be reining in AI experiments and production workloads because the cost-per-inference or cost-per-hour of cloud compute has climbed as memory and other components have become pricier. That restraint threatens demand growth, and therefore the valuations priced into manufacturers such as Micron.
- Component shortages have pushed HBM prices higher, raising operational costs for AI deployments.
- Rising costs are prompting more conservative use of large models and cloud compute by enterprises.
- Investor sentiment can turn quickly if revenue growth stalls despite a supply-constrained market.
Numbers that matter
| Metric | Value |
|---|---|
| Recent peak-to-trough share fall | ~32% |
| One-year share gain | ~700% |
| Bloomberg US datacentre forecast (2030) | 118 GW |
| Estimated capex per GW (Nvidia CEO) | $50 billion |
| Implied total capex to 2030 | $5.9 trillion |
Those figures are illustrative and depend heavily on assumptions about how much compute capacity is built, the pace of adoption of new memory architectures and whether chipmakers can expand supply without further price shocks.
What investors and customers should watch
For investors, the current episode is a reminder that momentum in hardware markets can reverse rapidly when end users push back on economics. For enterprise IT teams and hyperscalers, the focus now is on three practical questions: can memory suppliers ramp production enough to ease prices; will alternative architectures mitigate memory dependence; and how will cloud providers price compute to keep customers running profitable AI workloads.
Micron sits at the intersection of these issues. It benefits from tight supply and strong demand today, but its fortunes depend on whether rising component costs translate into durable spending or trigger a cautious behaviour change among the very customers it serves. That uncertainty explains why some market participants remain wary about buying the recent dip, despite an otherwise eye-watering run-up in the stock.
Short of a clear sign that component pricing will normalise or that customers will accept persistently higher operating expenses, the story for memory suppliers remains markedly uncertain even as AI continues to reshape demand for datacentre hardware.
