The driver is structural. AI model training and inference are hitting memory capacity limits worldwide. As models scale and inference deployments serve millions of requests daily, DRAM — not compute — is becoming the bottleneck on infrastructure expansion.
The impact is unevenly distributed across the global AI economy. Cloud providers in the US, Europe, and Asia Pacific face sharply higher server costs. Memory hardware represents a large share of GPU server bills, and a full-year increase of this scale means materially higher infrastructure capex in H2 2026.1 AI startups burning capital on inference workloads face compressed runways. AI SaaS companies with heavy compute costs may miss earnings estimates as hardware expenses outpace revenue growth.1
The winners are concentrated in East Asia. South Korea's SK Hynix and US-listed Micron — the two dominant suppliers of high-bandwidth and standard DRAM for AI servers — are positioned to outperform as prices climb.1 Samsung, which trails in HBM supply, faces a more complex picture. The pricing surge reinforces South Korea's leverage in global AI supply chains.
This creates a strategic divergence. Companies that lock in long-term supply contracts — or vertically integrate memory procurement — gain a durable cost advantage. Those buying at spot prices through H2 2026 absorb the full surge. For hyperscalers in the US, EU, and China, procurement strategy is now a competitive variable.
The broader AI infrastructure buildout — data centers, power capacity, undersea cables — has dominated capital allocation globally. The DRAM signal narrows the next constraint: it is not bandwidth or energy, but specific memory components at scale.
For AI companies planning capacity through the rest of 2026, memory procurement is no longer a line item. It is a strategic decision that will determine which providers can scale inference profitably — and which cannot.
