Behind Micron's Long Term Chip Deals for AI Data Centres

Despite memory chips sitting at the heart of modern data centre infrastructure, they have long been treated as commodities that buyers can swap between suppliers.

This keeps prices under pressure and leaves the sector exposed to sharp cycles of oversupply and shortage.

Even with AI workloads increasing demand for high-performance computing, analysts still question how durable the current wave of contracting will be if data centre investment slows or sentiment weakens.

Data centre demand and commodity memory

Memory used in servers and storage systems for the most part follows a familiar pattern.

Electronics firms and data centre operators switch suppliers to secure lower costs, which in turn forces down pricing across the market. Long-term contracts have been attempted before, but they have struggled to fully smooth out volatility.

The arrival of AI training and inference workloads has changed the scale of demand in data centres, yet the underlying structure of the market is sensitive to demand expectations.

If data centre expansion pauses, buyers can return to the spot market quickly, undoing contract stability and exposing suppliers to renewed pricing pressure.

Take-or-pay deals reshape chip economics

To address this cycle, memory suppliers are leaning on longer agreements that bind customers more tightly to future output.

US-based Micron Technology is among those pushing this model further, alongside South Korea’s Samsung and SK Hynix, as hyperscalers and AI developers seek guaranteed access to supply.

Micron Technology said this week that customers such as NVIDIA have committed US$22bn to lock in supplies of memory chips. It played up huge growth in five-year ‘take-or-pay’ deals that required clients to either buy its chips or hand over cash.

Sumit Sadana, Chief Business Officer at Micron Technology, tells Reuters: “Customers have put ‌billions ⁠of dollars on Micron’s balance sheet as a show of confidence and their commitment toward this new business model.”

In practice, this structure links data centre expansion plans more directly to semiconductor manufacturing capacity as large buyers underwrite supply to avoid shortages that could limit AI infrastructure build-outs.

These agreements also give suppliers clearer visibility over future revenue streams, even as demand conditions remain uneven across consumer electronics and enterprise systems.

Pricing power and data centre build-out

Memory pricing has become key to investor attention because it feeds directly into margins for suppliers and costs for data centre operators.

The core questions are whether current pricing strength can persist as new fabrication capacity comes online, and whether demand from AI-focused infrastructure remains strong enough to absorb it.

Although joining the US$1tn valuation club earlier this year, Micron reported an annual loss of US$5.3bn as recently as 2023.

That downturn followed a sharp drop in consumer electronics spending after pandemic-era demand eased, which left memory inventories elevated and pricing under pressure across the sector.

Even with long-term contracts in place, supply constraints are still a factor. New fabrication plants take years to construct and equip, meaning output cannot respond quickly to demand changes from data centre operators.

Micron has indicated that capacity growth will remain limited in the near term, with tighter supply conditions expected to persist until at least 2027.