Singapore's AI Boom Triggers Hardware Cost Crunch as DRAM Prices Surge 95%
Source: The Edge Singapore
The GPU-driven demand powering Singapore's data centre boom is constraining the global supply of standard enterprise hardware. DRAM prices have surged nearly 95% in 2026, and local businesses are feeling the squeeze as memory now accounts for more than half the cost of a new server.
There's an irony at the heart of Singapore's AI buildout: the more infrastructure the country pours into becoming Asia's AI hub, the harder and more expensive it gets for everyday enterprises to buy basic servers. The same GPU-driven demand that powers Singapore's data centre boom is now constraining the global supply of standard enterprise hardware, and local businesses — from SME IT departments to mid-market financial services firms — are feeling the squeeze.
DRAM prices have surged nearly 95% in 2026 alone, according to Broadcom's VCF Division general manager Krish Prasad, who laid out the math in a recent analysis for The Edge Singapore. Memory now accounts for more than half the cost of a new server, up from roughly a third in 2024. The root cause is structural: manufacturers have pivoted production capacity toward High Bandwidth Memory (HBM) for AI accelerators, leaving the commodity DRAM market undersupplied. Analyst estimates suggest these constraints could persist through 2027.
Singapore is on the front line of this supply crunch because it sits at the intersection of two trends. On one side, hyperscalers — Microsoft, Google, AWS — are pouring tens of billions into Singapore-based AI data centres, consuming vast quantities of servers and memory. On the other, Singapore's own enterprise sector — financial services, logistics, government-linked companies — is racing to adopt AI, which means competing for the same constrained hardware pool. The result is longer procurement cycles and thinner margins for any organisation that still thinks in terms of buying more iron to solve performance problems.
The traditional IT procurement playbook — throw hardware at the bottleneck — is becoming financially untenable. Prasad outlines three practical alternatives that align with where Singapore's enterprise sector is already heading: extracting more utilisation from existing servers through virtualisation and software-defined networking, shifting to NVMe-based memory tiering to reduce DRAM dependency, and delaying capital expenditure by modernising infrastructure incrementally rather than forklift-upgrading. These aren't exotic approaches — they're standard private cloud techniques that many Singapore enterprises already use in parts of their stack but haven't applied systematically to their AI workloads.
Why it matters for Singapore: The hardware cost crunch is a stress test for Singapore's AI ambition. If only hyperscalers can afford the infrastructure, the country's vision of broad-based AI adoption — where SMEs, government agencies, and traditional industries all benefit — hits a hard ceiling. The enterprises that adapt fastest to a software-defined, hardware-efficient approach will be the ones that maintain AI momentum while their competitors wait for procurement cycles to normalise. For Singapore's technology leaders, the question is no longer whether to adopt AI, but how to afford the infrastructure underneath it.
