Memory’s share of hyperscaler capex expected to nearly hit 50% by 2027, per SemiAnalysis and CLSA

According to SemiAnalysis and CLSA research, memory semiconductors are projected to comprise nearly 50% of hyperscaler capital expenditure by 2027, up from current levels. This shift reflects surging demand for AI infrastructure and data center memory, fundamentally altering how major tech companies allocate billions in annual capex.

The dramatic reallocation of hyperscaler spending toward memory represents a structural shift in technology infrastructure investment. As artificial intelligence workloads expand exponentially, data centers require unprecedented volumes of high-bandwidth memory to support training and inference operations. This trend reflects the computational demands of large language models and enterprise AI applications, which are memory-intensive rather than purely compute-bound.

Historically, hyperscalers balanced capex across computing processors, storage, and networking infrastructure. The semiconductor industry has long been compute-centric, with CPU and GPU architectures dominating investment discussions. However, AI's memory bottleneck—where data movement between processors and memory becomes the limiting factor—forces a strategic rebalancing. Memory chips, including HBM (high-bandwidth memory) and advanced DRAM, have become critical infrastructure constraints rather than commodity components.

This capex shift creates cascading market implications. Memory manufacturers like SK Hynix, Samsung, and Micron face unprecedented demand, potentially easing oversupply conditions that plagued the sector in 2023. For hyperscalers, memory costs consume larger portions of capex budgets, pressuring margins and potentially moderating the pace of new data center buildouts unless revenue from AI services justifies continued investment. Equipment suppliers serving the memory manufacturing ecosystem gain visibility into sustained growth cycles.

Investors should monitor whether actual memory spending trajectories match these projections. Supply chain bottlenecks in advanced memory manufacturing could constrain AI infrastructure deployment, creating investment opportunities in specialized memory suppliers. Conversely, if memory demand moderates, overinvestment could trigger cyclical downturns similar to previous semiconductor corrections.

• →Memory capex share for hyperscalers expected to reach nearly 50% by 2027, driven by AI infrastructure demands.
• →Advanced memory semiconductors have shifted from commodity components to critical infrastructure bottlenecks in AI deployments.
• →Memory manufacturers gain revenue visibility while hyperscalers face margin pressure from elevated capex allocations.
• →Supply chain constraints in memory manufacturing could limit AI infrastructure growth and create equipment supplier opportunities.
• →Historical semiconductor cycles suggest potential for oversupply and margin compression once memory capacity catches up with demand.