The memory industry in 2025 is experiencing one of the most tense moments in its recent history. Spot prices for DRAM chips have reached levels unseen in nearly 25 years, and forecasts from manufacturers and analysts point to an uncomfortable situation for users and hardware producers: shortages will not only persist into 2026 but could extend into 2027 and even 2028.
According to data from investment banks and companies like SK hynix, 16 Gb DDR4 and 16 Gb DDR5 chips are already trading in the range of $28–30, which is approximately 26–28 euros per chip at current exchange rates (1 USD ≈ 0.92 €). These prices are three times higher than in 2022–2023 and double the peaks of the last major bull cycle in 2017–2018, when prices hovered around $10, roughly €9.
In a reference chart prepared by BofA Global Research, the 2025 curve literally breaks the scale: the sharp vertical spike dwarfs previous crisis peaks (2004, 2017, 2018) and places DRAM prices in uncharted territory for an entire generation of PC users and data center managers.
AI is eating the memory… and supply is not catching up
This escalation is driven by multiple factors, but one clear protagonist: Artificial Intelligence. Every new generation of GPUs for training and deploying models requires more, faster, and more efficient memory. This increased demand directly impacts both HBM (High Bandwidth Memory), used in AI accelerators, and conventional DRAM powering servers, workstations, and, cascaded down, consumer PCs and laptops.
A recent report by SK Securities estimates that DRAM demand will grow between 20% and 25% in 2026. The problem is that production will not be able to match that pace. The firm believes that even by late 2026, a significant improvement in supply-demand balance will be unlikely, and inventories at major manufacturers will remain at historic lows.
In other words: the market has entered a favorable supercycle for suppliers, and capacity is insufficient to absorb the AI demand surge without pushing prices higher.
Samsung and SK hynix hit the brakes… and accelerate at the same time
SK Securities’ analysis pinpoints specific supply limitations. In 2025, investment by major players is focused on two fronts:
- Converting existing lines to more advanced nodes, especially targeting high-performance HBM and DDR5.
- Specific projects for HBM3E and HBM4, where the value per chip is significantly higher than in consumer-grade DRAM.
The challenge isn’t just funding—it’s also physical space. Current factories are operating at capacity, and building a new plant is a multi-year project.
- Samsung Electronics will dedicate 2025 to ramping up its new P4 plant, whose actual DRAM volume won’t materialize until the second half of 2026.
- SK hynix follows a similar approach with M15X, a facility focused almost exclusively on HBM, which will start contributing around mid-2026.
- Later projects include the Yongin fab (SK hynix), scheduled for 2027, and Samsung’s upcoming P5, expected no earlier than 2028.
Until these additional capacities come online, supply will remain constrained. And if AI demand maintains its current pace — which seems likely, given larger models and extensive cloud and enterprise deployments — the imbalance may persist beyond 2027.
Prices in euros: what do these figures mean for users
Talking about $28–30 per chip may seem abstract, but the real-world implications are straightforward:
- A 16 Gb (2 GB) DRAM chip at around 27 euros means that a 16 GB module for PCs, which integrates multiple chips along with PCB, controllers, and markup, will have a much higher base cost.
- In practice, analysts expect 32 GB and 64 GB DDR5 modules — now standard in enthusiast PCs and workstations — to remain expensive throughout 2026, with occasional dips but no return to 2022 prices.
- For consoles and laptops, where memory is soldered or integrated into the SoC, the impact translates into less room to lower retail prices or more conservative memory configurations in budget models.
The BofA chart itself shows that the current level far exceeds any other in the last 25 years. Unlike previous cycles driven by isolated disasters or forecasting errors, this time the trigger is a structural shift in how the industry consumes memory.
From spot contracts to long-term agreements
An important aspect of SK Securities’ analysis is the shift in business models. In an environment where prices can double or triple in a quarter, nobody wants to rely solely on spot markets.
Server manufacturers, hyperscalers, and major cloud providers are pushing for long-term supply contracts, not only for HBM but also for:
- Server DRAM
- Modules for workstations and enterprise solutions
- High-end consumer products, where predictability is critical
These contracts serve to:
- Secure volume in a tight market
- Smooth price curves by agreeing on less volatile adjustments or ranges
- Provide memory manufacturers with sufficient demand visibility to justify multimillion-dollar investments in new factories
All of this reinforces the supercycle narrative: high prices, multi-year visibility, and a central role for memory in the AI economy.
A supercycle that eats into macroeconomics… and boosts valuations
Within this context, SK Securities maintains its “Overweight” recommendation on the semiconductor sector, with target prices of 170,000 won for Samsung Electronics and 1,000,000 won for SK hynix. The thesis is clear: the prolonged memory supercycle acts as a driver capable of offsetting even a challenging macro environment.
While other parts of the tech market debate AI bubbles and short-term corrections, the numbers from major memory producers continue climbing, both in sales and margins. Factories are operating near full capacity, the product mix is shifting toward higher-value solutions (HBM, server DDR5), and customers accept more rigid contracts in exchange for supply security.
And the end user? Patience… and realism
The outlook for everyday users is less optimistic. It seems likely that:
- Upgrading PC RAM, buying a laptop with “lots” of memory, or building a home server will cost more, at least through 2025 and 2026.
- Next-gen consoles or mid-cycle hardware revisions will have little room to increase memory capacity without raising prices.
- Data centers and enterprises will need to rethink budgets and prioritize workloads that truly require cutting-edge memory versus those that can operate with more modest configurations.
Meanwhile, system designers will seek to optimize every gigabyte. Expect increased focus on memory compression, intelligent paging, offloading tasks to fast storage, and various tricks to maximize existing hardware.
2025 is not the end of the road, but a warning sign
The key takeaway from the price chart and SK Securities’ analysis is simple: 2025 is not a peak, but a warning that the industry has entered a new phase. The combination of:
- Accelerated transition to HBM3E and HBM4
- Exploding demand driven by generative AI and large models
- Physical limitations on expanding manufacturing capacity
has fundamentally changed the game in the memory economy.
Until new factories from Samsung, SK hynix, and others reach full production — expected no earlier than 2027–2028 — DRAM and HBM shortages will remain the norm, not the exception. Consequently, prices in euros, already painful for large data centers, will continue to impact enthusiastic PC users’ wallets.