DRAM prices have surged approximately 172% year-over-year by late 2025, with Q1 2026 contract prices projected to jump a record-breaking 90–95% quarter-over-quarter. This unprecedented memory cost escalation — dubbed “RAMmageddon” by industry watchers — is driven primarily by AI infrastructure demand cannibalising conventional DRAM production. For New Zealand organisations, the impact is amplified by a weak NZD and geographic isolation, making cloud alternatives increasingly attractive for those planning hardware refreshes.
AI’s Insatiable Appetite Is Starving the Memory Market
The root cause of today’s DRAM shortage is structural, not cyclical. High Bandwidth Memory (HBM), the specialised RAM required by AI accelerators like NVIDIA’s H200 and Blackwell GPUs, consumes 3–4× the wafer capacity per gigabyte compared to conventional DDR5. AI is projected to consume roughly 20% of global DRAM wafer capacityby 2026, and OpenAI’s Stargate project alone could require up to 40% of global DRAM output. Meanwhile, the “Big Three” manufacturers — Samsung, SK Hynix, and Micron, who collectively control approximately 95% of the DRAM market — are prioritising HBM production because it commands profit margins 5–10× higher than standard RAM.
The numbers tell a stark story. DDR5 consumer pricing has exploded: 32GB DDR5-6000 kits rose from roughly US$80–$120 in mid-2025 to approximately US$410 by December 2025, a 242% increase. DDR4, once the budget option, saw even steeper percentage gains — 32GB DDR4-3200 kits jumped from around US$50 to US$200, a 300% increase — after Samsung and Micron halted most DDR4 production and SK Hynix cut DDR4 output to just 20%.
Server memory has been hit especially hard. TrendForce’s February 2026 forecast revised server DRAM contract prices upward to approximately 90% quarter-over-quarter growth for Q1 2026 — the largest quarterly increase ever recorded. Counterpoint Research projects 64GB DDR5 RDIMM modules could cost twice as much by late 2026compared to early 2025. Micron’s CEO Sanjay Mehrotra stated bluntly that demand will outstrip supply for the foreseeable future. All major server OEMs — Dell, HPE, Lenovo, and HP — have announced or implemented approximately 15% server price increases, with memory-heavy configurations seeing steeper hikes. Relief from new fabrication plants is not expected until late 2027 or 2028.
New Zealand Faces a Compounding Triple Penalty
New Zealand organisations absorb these global price surges through a punishing multiplier effect. The NZD/USD exchange rate hovered around 0.58–0.60 throughout 2025, meaning Kiwi buyers pay approximately 67–72% more in local currency for every US-dollar-denominated piece of hardware. A server priced at US$10,000 translates to roughly NZ$16,700–17,200 before GST. Add the mandatory 15% GST on all imports, and the cost climbs further still.
Geographic isolation compounds the financial pain with logistical friction. Ocean freight reliability to New Zealand deteriorated sharply in 2025, with on-time arrivals dropping to just 13.9% and freight costs rising an estimated 10–15%. The economic backdrop intensifies pressure on IT budgets — New Zealand’s GDP contracted 0.5% in 2024, and Budget 2025 carried the lowest operating allowance (NZ$1.3 billion) in a decade. A server that might have cost NZ$15,000 in 2024 could realistically cost NZ$22,000–25,000+ in 2026 after accounting for DRAM increases, OEM price hikes, currency effects, and freight.
The Shifting Calculus: Cloud vs On-Premises
Rising hardware costs fundamentally alter the economics of the build-versus-buy decision. Memory typically represents 20–30% of a server’s bill of materials for general workloads, rising to 30–50% for memory-intensive applicationslike databases, virtualisation, and in-memory computing. When DRAM prices nearly triple, the capital expenditure case for on-premises infrastructure becomes significantly harder to justify.
Cloud computing offers a structural hedge against hardware volatility. By converting large, unpredictable capital expenditures into smaller, more predictable operational expenses, cloud insulates organisations from buying hardware at peak prices. Critically, cloud providers absorb a significant portion of infrastructure cost increases rather than passing them through directly — analysis suggests providers pass through roughly 33–40% of input cost increases. This absorption is possible because providers negotiate long-term supply contracts and purchase at vastly greater scale than any individual enterprise.
However, cloud is not universally cheaper. For sustained, high-utilisation workloads running 24/7, on-premises infrastructure can deliver significantly lower total cost of ownership over a five-year horizon. The landmark Andreessen Horowitz analysis estimated cloud repatriation typically achieves one-third to one-half the cost of equivalent cloud workloads at scale. Notable repatriation successes include Dropbox (saving US$75 million over two years) and 37signals (projecting US$7 million savings over five years after exiting AWS).
Cloud pricing is also rising — Microsoft raised Azure and Microsoft 365 prices up to 40% in April 2025, and global cloud operating costs rose approximately 19% in 2025. Yet even with these increases, cloud costs are rising more slowly than hardware costs, widening the relative advantage for organisations that would otherwise need to purchase new servers.
Migration and Repatriation: A Nuanced Picture
Cloud repatriation — moving workloads back from public cloud to on-premises infrastructure — is a genuine trend. The Flexera 2025 State of the Cloud Report found that 21% of cloud workloads have been repatriated, and 84% of organisations cite managing cloud spend as their top challenge.
Yet this must be read alongside continued strong cloud growth. Public cloud spending reached US$723.4 billion in 2025, up 21.5% year-over-year. The dominant enterprise strategy is hybrid: Gartner predicts 90% of organisations will adopt hybrid cloud by 2027. Most repatriation is selective — bringing back specific predictable workloads while keeping elastic or compliance-sensitive workloads in the cloud.
Critically, rising hardware costs are extending repatriation breakeven timelines considerably. Organisations now face 24–36+ months to recoup repatriation investments, all assuming sustained high utilisation. For AI workloads specifically, hardware cost increases combined with GPU procurement delays of 6–9 months make repatriation economically unviable in most scenarios.
What NZ Organisations Should Consider
For New Zealand specifically, the cloud ecosystem has matured rapidly: AWS launched its Auckland region in September 2025, Microsoft Azure established a local hyperscale region in 2024, and sovereign NZ providers like ASI Cloud, Catalyst Cloud, and Datacom offer alternatives with stronger data sovereignty guarantees.
Data sovereignty remains a critical differentiator. The US CLOUD Act means American cloud providers can be compelled to hand over data regardless of where it is physically stored — including in NZ data centres. Genuinely sovereign NZ cloud providers offer data that remains exclusively under New Zealand law.
The practical framework for NZ IT decision-makers comes down to workload characteristics:
- Variable, bursty, or growing workloads strongly favour cloud — avoiding the risk of purchasing servers at peak prices with unfavourable exchange rates.
- Stable, predictable, high-utilisation workloads may still justify on-premises investment, though the breakeven timeline has lengthened significantly.
- AI and GPU-intensive workloads strongly favour GPU-as-a-Service models, given the extreme cost and scarcity of AI hardware.
- Compliance-sensitive workloads benefit from sovereign NZ cloud providers that guarantee data residency under NZ jurisdiction.
- Hybrid approaches — placing the right workloads in the right environment — represent the dominant strategy globally, and offer the most balanced approach to cost, performance, and risk management.
Conclusion: Volatility Demands Flexibility
The DRAM crisis reflects a structural reallocation of the global memory supply chain toward AI, with meaningful supply relief unlikely before late 2027. For NZ organisations, the combination of global hardware inflation, unfavourable currency dynamics, and logistics challenges creates a compelling case to scrutinise every planned hardware purchase against cloud and hybrid alternatives. Cloud is not always cheaper — but in a period of unprecedented hardware cost volatility, the OpEx model’s predictability represents a genuine strategic advantage. The organisations best positioned through this turbulence will be those treating infrastructure decisions as a portfolio allocation problem — matching each workload to the deployment model that optimises cost, performance, sovereignty, and risk.
- TrendForce — Memory Price Outlook for 1Q26 (Feb 2026) — The definitive industry source for the 90% QoQ price jump figure. TrendForce is the gold standard for DRAM market data.
- IEEE Spectrum — AI Boom Fuels DRAM Shortage and Price Surge — Authoritative, well-respected technical publication that ties the AI/HBM demand story together clearly.
- The Register — Server Prices Set to Jump 15% — Covers the OEM price hike impact (Dell, HPE, etc.) which is the direct cost transmission mechanism to NZ buyers.
- Flexera — 2025 State of the Cloud Report — Industry benchmark for cloud spend stats, repatriation figures (21%), and the 84% cost management challenge — anchors the cloud vs on-prem section.
- HBS — Cloud Repatriation Trends: Cost, AI and the Push Towards Hybrid — Best single source covering the repatriation nuance and breakeven timeline extensions.