Ask anyone who has priced a server in the past six months. Quotes that once remained valid for 30 days are now changing within a week—sometimes even 48 hours. Configurations are being re-priced between the proposal stage and the purchase order. This is more than routine market volatility; it reflects a structural shift in the enterprise technology landscape.
The answer lies in memory. DRAM contract prices rose by 90 to 95 per cent in the first quarter of this year, the steepest quarterly increase TrendForce has ever recorded, followed by another 58 to 63 per cent rise in the second quarter. NAND flash prices are now accelerating as well. Dell’s Chief Operating Officer recently noted that DRAM, which cost around US$0.43 per gigabit in mid-2025, has crossed US$2.39. Memory now accounts for roughly one-third of the bill of materials for a new machine, compared to about one-fifth just a year ago. Nearly every OEM has revised prices, and many have done so more than once.
It is worth understanding why this is happening, because it fundamentally changes how enterprises should approach infrastructure decisions. This is not a pandemic-style disruption that will ease once factories resume normal operations. High-bandwidth memory (HBM) for AI accelerators consumes roughly three wafers of manufacturing capacity for every one wafer of conventional DDR5. At the same time, hyperscalers building AI data centres are placing open-ended orders at prevailing market prices. SK Hynix declared its 2026 production effectively sold out before the year had even begun. What remains is what the rest of the market competes for. Analysts have started calling the resulting premium the AI tax, and the description is fitting.
Every enterprise in India purchasing a server, workstation or laptop today is, in effect, helping fund someone else’s AI infrastructure expansion. With significant new fabrication capacity not expected until late 2027, this pricing environment is likely to persist for at least the next eighteen months.
This also challenges the common instinct to simply "move everything to the cloud." Cloud infrastructure depends on the very same memory supply chain. In fact, hyperscalers are the buyers driving much of today's demand, and memory is consuming an unprecedented share of their own data centre investments. Their costs are increasing as well. Whether, and how quickly, those costs are passed on to customers remains the only uncertainty.
The cloud continues to excel where it always has—handling elastic workloads, unpredictable demand, experimentation and rapid capacity expansion. However, much of enterprise computing is neither unpredictable nor temporary. It is steady, consistent and runs around the clock. Paying an elasticity premium for predictable workloads was already difficult to justify before the AI-driven memory surge. It is becoming even harder today.
That is why I continue to believe in hybrid infrastructure—not as a compromise between cloud and on-premises environments, nor as a transitional stage, but as a permanent operating model. Decisions should be made workload by workload. The mature enterprise of the next decade will not define itself by having either a cloud strategy or an on-premises strategy. It will have a workload strategy.
There is, however, another dimension of hybrid that deserves equal attention: hybrid ownership. The question is no longer simply, "Should this workload run in the cloud or on-premises?" Increasingly, it is also, "Should we buy new equipment, rent it, extend the life of existing assets, or deploy certified refurbished infrastructure?"
Viewed through today's pricing environment, the answer often changes. A five-year-old server already on your balance sheet contains memory purchased at 2021 prices. Extending its useful life with the right maintenance and support effectively creates an arbitrage opportunity against today's elevated memory costs. Certified refurbished systems bring yesterday's component economics into today's budgets. Renting transforms a capital expenditure shock into a predictable operating expense while transferring residual-value risk elsewhere.
None of these options were unconventional before. At current DRAM prices, they are simply the questions a disciplined CFO is expected to ask.
It is only fair to acknowledge my own perspective. Comprint has spent more than three decades helping organisations manage the complete IT lifecycle—from procurement and rentals to maintenance, refurbishment and certified recycling. Naturally, I believe these approaches create value. Yet the underlying economics stand regardless of who makes the argument.
Total cost of ownership—considering acquisition costs, maintenance, financing, downtime risk, upgrade cycles and end-of-life management—is increasingly becoming the most meaningful measure of infrastructure investment. The most forward-looking organisations we work with no longer begin by asking, "What should we buy?" Instead, they ask, "What does this workload require, and what will it cost throughout its entire lifecycle?"
Regulation is reinforcing the same trend. India's data protection rules were notified last November, while substantive compliance obligations begin in May 2027. Although the legislation does not mandate blanket data localisation, categories of data that significant data fiduciaries must retain within India are yet to be fully specified. Meanwhile, sector-specific regulators—most notably the Reserve Bank of India for payment data—already require local storage. As a result, enterprises are designing 2027 infrastructure under 2026 uncertainty. In such circumstances, greater control over infrastructure and data location becomes a strategic advantage.
I therefore expect three shifts to define enterprise IT over the next few years. First, workload-level economics will become standard practice, with infrastructure investments evaluated individually rather than as a single IT budget. Second, automatic hardware refresh cycles will gradually give way to lifecycle extension, certified refurbishment and structured asset retirement as mainstream business decisions. Third, compliance will move from being an afterthought to becoming a core architectural consideration, with data location driven as much by regulation as by engineering requirements.
For India's mid-market—the thousands of companies that rarely make technology headlines but power the country's everyday economy—these are not theoretical discussions. They are practical business decisions. Which workloads justify cloud spending at today's rates? Which servers truly require replacement? And which systems can continue delivering reliable performance for another two years with the right support?
After more than three decades in this industry, one lesson has remained consistent through every technology cycle: infrastructure decisions deliver lasting value when they are driven by economics and accountability, not by industry trends or fashion. The recent memory market disruption has simply made that lesson far more expensive to ignore.
Kunal Sancheti is a Director at Comprint Tech Solutions, a Mumbai-headquartered enterprise IT company that has managed the complete technology lifecycle for more than three decades: procurement, rentals, maintenance, refurbishment and certified recycling. Views are personal.
𝐒𝐭𝐚𝐲 𝐢𝐧𝐟𝐨𝐫𝐦𝐞𝐝 𝐰𝐢𝐭𝐡 𝐨𝐮𝐫 𝐥𝐚𝐭𝐞𝐬𝐭 𝐮𝐩𝐝𝐚𝐭𝐞𝐬 𝐛𝐲 𝐣𝐨𝐢𝐧𝐢𝐧𝐠 𝐭𝐡𝐞 WhatsApp Channel now! 👈📲
𝑭𝒐𝒍𝒍𝒐𝒘 𝑶𝒖𝒓 𝑺𝒐𝒄𝒊𝒂𝒍 𝑴𝒆𝒅𝒊𝒂 𝑷𝒂𝒈𝒆𝐬 👉 Facebook, LinkedIn, Twitter, Instagram