AI Is Quietly Inflating Your IT Budget — And Most Organisations Haven't Planned For It

"I have never seen memory-chip costs rise this fast."

That was Dell's Chief Operating Officer, Jeff Clarke, speaking to investors in late 2025. Within weeks, Dell issued procurement alerts warning enterprise customers of 15 to 20% price increases. Lenovo followed in January 2026. HP shortly after.

If your 2026 IT budget was approved before these supplier notices landed, there's a good chance it no longer reflects market reality. What began as a procurement issue for hardware vendors is rapidly becoming a budgeting challenge for every organisation planning a technology refresh.

Everything is more expensive — and it's structural, not cyclical

AI data centres require GPUs to train and run models, memory chips to feed those GPUs data at speed, and CPUs to orchestrate the workloads. The same manufacturers producing these components for hyperscalers also produce the chips inside enterprise laptops, servers, storage systems and network equipment. When AI companies buy at the volumes they currently are, supply for everyone else contracts—and prices rise across the board.

Memory has been hit hardest. The three companies controlling virtually the entire global DRAM market—Samsung, SK Hynix and Micron—have shifted production toward high-bandwidth memory (HBM) for AI chips. Unlike conventional memory used in business laptops and servers, HBM is designed to feed AI processors with enormous amounts of data at extremely high speeds. It is significantly more complex to manufacture, meaning every unit produced consumes substantially more manufacturing capacity than conventional memory.

The price of a common server memory module used in enterprise servers rose from roughly US$250 to around US$700 between October 2025 and March 2026, according to industry pricing tracked by enterprise hardware distributors.

CPUs followed. Intel and AMD both implemented price increases on selected processor lines, while manufacturers increasingly focused production and investment on higher-margin data centre products, contributing to tighter supply across the wider enterprise market. GPU order backlogs for Nvidia's latest generation stretch well into 2027. Storage and networking are not immune either. NAND flash prices surged in early 2026, while major infrastructure vendors have also signalled price increases across servers and networking equipment.

This is not a single-component problem. It is an across-the-stack cost event—and unlike the COVID-era semiconductor shortage, which gradually eased as manufacturing and logistics recovered, today's pressures are driven by manufacturers deliberately allocating more capacity towards higher-margin AI products. Even as manufacturers expand production, building and ramping semiconductor fabrication capacity takes years rather than months, meaning supply constraints are unlikely to ease quickly.

The risk that doesn't appear in the procurement report

The instinctive response to hardware inflation is rational: extend the lifecycle of existing equipment. If a server refresh now costs 20% more, deferring it by 12 to 18 months looks sensible from a financial perspective.

What often doesn't appear in that calculation is the security exposure that accumulates when hardware runs beyond its intended lifecycle.

Older hardware eventually runs firmware and operating systems that manufacturers stop supporting. End-of-life devices no longer receive security patches. The attack surface grows not because the threat landscape changed, but because the organisation chose—for entirely legitimate financial reasons—to keep hardware in service longer than its security posture was originally designed to support.

Organisations have already seen a version of this with Windows 10 reaching end of support in October 2025. Many accelerated hardware refresh programmes to remain on supported platforms, while others delayed upgrades and temporarily operated unsupported systems—precisely the period when attackers know security patches have stopped arriving.

A lifecycle extension decided in procurement carries security and governance consequences that often land on risk and compliance teams without the connection being made explicit when the original purchasing decision was made.

What this means for Singapore organisations

Cloud providers purchase infrastructure built from many of the same semiconductor supply chains. If higher hardware costs persist, organisations should expect some infrastructure costs to work their way into cloud pricing over time, although the timing and extent will vary by provider and service. Businesses that shifted workloads to the cloud to avoid on-premises hardware costs may therefore find that the savings are not as permanent as expected.

For Singapore SMEs, the Productivity Solutions Grant (PSG) currently co-funds up to 50% of qualifying digitalisation costs. If a hardware refresh was already on the roadmap, running the numbers while current grant support remains available—and before procurement costs increase further, should current market trends continue—is worth the effort.

How to hedge the exposure

Multi-year contracts with preferred vendors remain one of the most direct hedges against price volatility. Locking in 24- to 36-month pricing agreements can insulate the organisation from spot market fluctuations. The trade-off is committing to future purchase volumes, but in a market where prices continue to trend upward, that certainty may be worth more than maximum flexibility.

Device-as-a-Service (DaaS) and leasing models offer another option by transferring much of the hardware price risk to the vendor. The organisation pays a predictable monthly fee while receiving hardware on a scheduled refresh cycle regardless of what the underlying component market does. As an added benefit, devices remain within their supported lifecycle, reducing the security risks associated with ageing equipment.

Cloud reservation pricing also deserves renewed attention. Committing to one- or three-year reserved capacity with cloud providers can lock in today's rates for predictable workloads. What was once viewed primarily as a billing optimisation increasingly becomes a risk management decision in an environment where infrastructure costs may continue to rise.

Finally, organisations should revisit the assumptions behind their technology budgets. A procurement budget based on last year's prices may no longer reflect today's market. Building additional headroom into future hardware refresh plans is becoming less about conservatism and more about recognising structural changes in the technology supply chain.

Equally important, every device being considered for a lifecycle extension should trigger an explicit review of vendor support status. Procurement decisions and risk acceptance decisions are related—but they are not the same thing. Both deserve to be documented.

The hidden cost of the AI boom

The AI boom is not coming for your budget directly.

It's arriving through your next procurement cycle.

For many organisations, the first signs won't be a breakthrough AI application or a new business opportunity. They'll be unexpectedly higher quotations for laptops, servers, storage and networking equipment.

The organisations that recognise this shift early will have time to renegotiate contracts, adjust budgets and reassess refresh strategies before the impact compounds.

Those that don't may discover that the cost of AI isn't only measured in GPUs powering someone else's data centre—but in the technology budget they thought had already been approved.