Amazon Web Services CEO Andy Jassy didn’t mince words on the earnings call. “On memory and storage and the supply chain, I think everybody knows that the cost of components, particularly memory, has skyrocketed. We are in a stage where there is just not enough capacity for the amount of demand,” he said, as reported by The Register. Enterprises scrambling for on-premises servers face stark choices. Lead times stretch. Prices quadruple. Hyperscalers scoop up supply, leaving smaller players high and dry.
AI workloads devour DRAM and high-bandwidth memory, or HBM. Chipmakers like Samsung, SK Hynix, and Micron pivot production lines to these high-margin products. Standard server memory? De-prioritized. Data centers will gobble 70% of high-end chips made in 2026, per TrendForce estimates cited across reports including TechRadar. That’s not hyperbole. It’s arithmetic born from exploding AI server demand.
Jassy sees upside for AWS, which posted $37.6 billion in quarterly revenue. Cloud giants lock in supply through long-term deals. “A meaningful part of those suppliers are prioritizing their very largest customers, which cloud providers are,” he noted. Enterprises with on-prem plans accelerate migrations. Conversations drag for months. Then memory woes tip the scale. AWS isn’t alone. Insight Enterprises’ Peter FitzGibbon observes clients fleeing data centers for Google Cloud. “There’s a couple of things driving it: chip shortages, less access to the gear, better access to the AI models in the cloud,” he told The Register.
Not everyone buys the mass exodus narrative. Omdia analyst Roy Illsley calls it cloud vendor FUD. “Cloud is a strategic choice that customers make, and most on-premises datacenters already have servers with memory, so it just means they delay a refresh or upgrade,” he said. Larger firms wait out delays. Smaller ones might shift. But a flood? Unlikely. Gartner VP Tony Harvey offers a middle ground. Server prices hit four times last year’s levels. Shipments lag. Cloud vendors hold non-GPU prices steady. “When an on-premises server costs 4x what it did a year ago that changes the comparison to cloud,” Harvey explained to The Register.
The crunch extends beyond servers. Meta stretches server lifecycles from six to seven years amid a “significant server supply deficit,” according to the Wall Street Journal. DRAM and hard drives stay tight through 2027. Hard drive makers sold out 2026 output already. Server CPUs face shortages too, with prices potentially up 15%, as noted in earlier Register reporting. Team Group GM Gerry Chen warns 50-60% of HBM and server capacity is locked through 2027 via long-term pacts, per Digitimes. Prices hold firm or climb.
Synopsys CEO Sassine Ghazi pegs the memory pinch lasting through 2027, fueled by AI data center booms that snag chips needed for smartphones too, as covered by CNBC. IDC charts the shift: data centers claimed 50% of DRAM in 2025, up from 32% five years prior, with AI servers eyeing 60% by 2030 (IDC). Hyperscalers like Microsoft, Google, Meta, and Amazon grab silicon wafers, forcing memory firms to chase enterprise grades over consumer fare.
Samsung projects worsening shortages into 2027. Customers book 2027 supply now. Demand fulfillment rates hit record lows as DRAM prices jump 90% quarterly, according to earnings highlighted by Tom’s Hardware. ASML’s CEO flags sold-out 2026 memory capacity, with constraints persisting, echoing posts from industry watchers on X. OpenAI’s massive data center plans pull forward DRAM hoarding, per SHI Insights.
Server configs narrow. Lead times balloon to 36-52 weeks. OEMs shorten quotes, nix price guarantees. IT leaders rethink budgets. Forrester’s Alvin Nguyen sees impacts rippling to storage, networks, desktops—even laptops. A single AI server slurps memory equal to dozens of laptops, Northeastern experts note (Northeastern News). Agentic AI amps needs further: more inference, larger contexts, multi-agent swarms.
Hyperscalers front the queue. Enterprises lag. AWS turns pain into gain. But cracks show. Meta rationes hardware. Even labs like DeepMind gate experiments by compute queues, as X discussions reveal. Intel eyes inference demand for CPUs; ratios shift from 1:8 GPU to 1:1, ballooning memory per node 16-fold. Winners? Memory giants Micron, Samsung, SK Hynix. CPU makers Intel, AMD. Losers? Anyone outside the priority list.
Supply won’t ease soon. New fabs take years. Wafer reallocations favor HBM. AI’s memory wall looms large. Data centers scale out, chaining racks amid bandwidth chokepoints (Fortune). Prices soar. Strategies pivot. Cloud beckons. On-prem holds—for those who can wait.
WebProNews is an iEntry Publication