AI storage crisis: data flood looms
The rapid development of Artificial Intelligence (AI) is leading to an unprecedented flood of data, pushing global storage capacities to their limits. Hard drives are becoming scarcer, SSDs more expensive, and data centers are expanding at a pace that overwhelms supply chains. This development is leading to a storage crisis that affects not only hyperscalers but also businesses and private users.
Source: Own representation
Introduction
The underlying dynamics are clear: Generative KI und andere datenhungrige Anwendungen treiben eine Daten- und Speicherexplosion voran. At the same time, the production of storage chips and hard drives is hitting physical, financial, and ecological limits. This combination is leading to a storage crisis that impacts everything from hyperscaler data centers to the everyday lives of businesses and private users. (CineD).
Globally, data volumes are growing towards 200 Zettabyte gespeicherter Daten bis 2025, , while AI data centers are buying up ever larger portions of available storage and storage module production. IDC forecasts that the 'Global Datasphere' will reach around 175 Zettabyte anwächst. . For companies, this means that storage is no longer a given 'cheap and unlimited' commodity, but is becoming a scarce factor of production with rising prices, delivery times, and environmental costs. (Tom’s Hardware).
AI-driven storage demand
The amount of data generated worldwide has exploded in recent years. Statista data, analyzed by Rivery, already put the global data volume at 149 Zettabyte in 2024 and expect around 181 zettabytes by the end of 2025. Exploding Topics, based on similar Statista figures, concludes that in 2025 worldwide 181 Zettabyte an Daten generiert werden and approximately 402.74 million terabytes will be generated daily.
A significant portion of this growth is directly or indirectly related to AI. Companies are building data lakes for training data, storing model checkpoints with tens of terabytes per version, and retaining huge log and telemetry data for evaluation and monitoring. (CIO.com). The IDC/Seagate report "Data Age 2025" estimates that to handle this data flood, over 22 Zettabyte an zusätzlicher Speicherkapazität will need to be shipped across all media between 2018 and 2025, with almost 59 percent coming from the HDD industry.
At the same time, the balance of power within the storage industry is shifting. Tom's Hardware describes how AI data centers are absorbing large portions of the global supply of DRAM, NAND flash, and HDD capacity, setting the stage for a „Preiskrise“ über das gesamte Jahrzehnt bereiten. The board of storage manufacturer Adata even speaks of the first situation in their career where DRAM, NAND flash, and hard drives are scarce simultaneously because cloud providers with an AI focus are buying up most of the material directly. Tom’s Hardware).
Data centers at their limit
IDC and Seagate project the global datasphere to reach 175 Zettabyte. by 2025. Cybersecurity Ventures concurrently assumes that by 2025, the world will have 200 Zettabyte an Daten speichert.
Even without AI, this would be a challenge. According to IDC, a large portion of this data will need to be stored, analyzed, and backed up by companies, leading to massive investments in storage infrastructure. (i-scoop.eu). However, the AI wave significantly exacerbates this. A report by IoT Analytics estimates spending on data center equipment and infrastructure in 2024 at 290 Milliarden US-Dollar, , with market size expected to grow to $1 trillion by 2030, largely driven by AI workloads. McKinsey estimates worldwide by 2030 6,7 Billionen US-Dollar Investitionen in Rechenzentren, , of which $5.2 trillion is explicitly for AI-enabled infrastructure.
At the same time, individual providers are planning extreme expansion goals. An internal Google presentation, reported by PC Gamer, describes the goal of doubling AI serving capacity every six months and scaling it by a factor of 1,000 in four to five years. PC Gamer). Tom's Hardware cites estimates that the 250 Gigawatt Rechenkapazität bis 2033 targeted by OpenAI would consume as much electricity as the entire country of India. Meta, according to Blocks & Files, is planning "super massive" AI data centers that are expected to lead to gigantic storage orders for selected suppliers. Blocks & Files).
Thus, the storage crisis due to AI is not just a vague feeling, but the result of measurable plans: exponentially increasing data volumes meet long-term multi-billion dollar programs for data centers that already dominate the supply chains for storage components today. (Yahoo Finanzen).
Source: dinnova.ch
TrendForce data, cited by TechRadar, shows that delivery times for Nearline HDDs of 32 terabytes or more have reached über ein Jahr gestiegen sind, as AI data volumes overwhelm the global storage infrastructure. HWBusters, in an analysis, speaks of an "AI storage crunch" where the demand for high-capacity HDDs from AI applications is causing massive delays and price increases. HWBusters).
Tom's Hardware describes how the bottleneck extends along the entire chain: a report quotes the Adata Chairman saying that AI data centers are consuming hard drives, SSDs, and DRAM „wegfressen“ and inventoried at suppliers have shrunk from the typical two to three months to two to three weeks. TechSpot references a report indicating that data centers are now "hoarding" SSDs because hard drive supply chains are so tight that delivery times for enterprise HDDs have extended to bis zu zwei Jahre ausdehnen können. Network World adds that delivery times for hard drives have increased from a few weeks to über ein Jahr gestiegen sind and prices for enterprise flash are already significantly increasing.
On the flash side, manufacturers are warning of structural bottlenecks: Solidigm expects at least dreijährigen NAND-Knappheit. The CEO of Phison publicly confirms that prices for NAND flash have doubled within six months mehr als verdoppelt haben and shortages are expected until at least the end of 2027. In parallel, Sourceability reports an AI-driven DRAM shortage with sharply rising module prices and a foreseeable further scarcity towards 2026. Sourceability).
The consequences are already noticeable for specialized industries: the trade magazine CineD warns filmmakers that a global storage shortage from 2026 onwards could significantly increase prices for camera media, SSDs, and post-production hardware, and partially delay the release plans for new camera models. CineD).
For those who want to see the interplay of HDDs and SSDs in the context of the AI storage crisis visualized, this YouTube video offers a concise overview of the current situation in data centers:
Source: YouTube
Consequences of the storage crisis
For large cloud providers, the storage crisis primarily leads to rising CAPEX and OPEX: IoT Analytics shows that spending on data center infrastructure in 2024 is already at 290 Milliarden US-Dollar lagen. Inside Towers describes how the global AI wave is increasing the demand for data center capacity so much that operators are competing for suitable locations, power, and network connectivity. Inside Towers).
These costs are not only reflected in balance sheets but also ultimately in prices: network and IT media such as Tom's Hardware and Network World report in unison that SSD and HDD manufacturers are already announcing price surcharges for enterprise models, which are likely to be reflected in cloud storage tariffs, backup solutions, and consumer products in the medium term. Tom’s Hardware, Network World).
For smaller companies, the crisis is more apparent in everyday life: CineD points out that film productions have to recalculate their media and backup budgets because storage cards, SSD RAIDs, and archiving systems are becoming more expensive and harder to obtain. CineD). The same applies to agencies, research institutes, or start-ups that process large amounts of data: rising storage prices can delay projects or force them to keep less data, which directly affects the quality of analyses and AI models. CIO.com).
Private users are indirectly affected: trade media like Tom's Hardware and Yahoo Finance point out that the same DRAM and NAND production used in AI servers is also needed for consumer devices such as laptops, smartphones, and televisions, so shortages on the corporate side lead to steigenden Endkundenpreisen führen können (Yahoo Finanzen).
Ecological aspects
The storage crisis due to AI is inextricably linked to the issue of energy. Goldman Sachs forecasts that the global electricity demand of data centers will increase by up to 165 Prozent im Vergleich zu 2023 ansteigen könnte. by the end of the decade. A study cited by Reuters estimates that data centers in the US will consume up to 12 Prozent des gesamten nationalen Stromverbrauchs ausmachen könnten. by 2028. The International Energy Agency expects data centers to mehr Strom verbrauchen könnten als das heutige Japan.
by 2030. In addition to this energy hunger, there is an often underestimated water consumption: Brookings describes that a typical data center uses around 300.000 Gallonen Wasser pro Tag für Kühlung benötigt. A recent overview of the water balance of AI data centers summarizes that US data centers directly consumed approximately 66 Milliarden Liter Wasser verbrauchten.
in 2023. This ecological perspective increases the pressure on the storage industry: the higher the storage density, the greater the production complexity – for example, with modern 3D NAND structures with over 200 layers – and the higher the resource consumption in the fabs. Blocks & Files). At the same time, providers like Huawei argue that all-flash data centers can be more energy-efficient in the long run than HDD-based systems, which could at least partially mitigate the storage crisis with efficiency gains. Huawei).
For those who want to better understand the physical infrastructure and environmental consequences of data centers, this explanatory YouTube video provides a good introduction to the functionality and growth pressure of data centers in the AI era:
Source: YouTube
Solution approaches
The storage crisis is not a law of nature, but it cannot be solved with a single "technical trick" either. IDC and Seagate have been pointing out for years that a large portion of the data generated nie genutzt wird. . This suggests that technical capacity growth needs to be combined with much stricter data discipline.
On a technical level, manufacturers rely on multi-tiered storage architectures: extremely fast but expensive NVMe SSDs serve as caches for active AI workloads, while cheaper, high-capacity HDDs and QLC SSDs are used for archiving and cold data. Tom’s Hardware). Providers like Seagate position systems such as the Exos 4U100 mit bis zu 3,2 Petabyte explicitly as a response to AI workloads. Hitachi Vantara argues that a consistent transition to all-flash architectures with mature data services can improve not only performance but also energy efficiency and reliability. Hitachi Vantara).
On an organizational level, it's about which data should be permanently retained in the first place: studies like "Data Age 2025" emphasize that companies need to differentiate between regulation-relevant, business-critical, and merely "nice to have" data in order not to unnecessarily tie up storage capacity and energy. Seagate). CIO.com shows that many companies are currently redesigning their storage architectures to provide AI workloads with fast, consistent access times while efficiently handling backup, archiving, and compliance. CIO.com).
For companies, this means specifically: clearly defining data lifecycles, enforcing deletion deadlines, targeted curation of training data instead of storing "everything forever," and designing AI workloads to achieve more with less data – for example, through efficient model architectures, retrieval-augmented generation, or smaller, specialized models. Hitachi).
For those who want to see the economic and infrastructural constraints behind the AI storage crisis presented visually, this video provides a good overview of data centers at the intersection of AI, energy, and storage:
Source: YouTube
The storage crisis due to AI is the logical consequence of three developments: exploding data volumes Cybersecurity Ventures), , aggressive expansion plans for AI data centers, and a storage industry that is reaching physical and financial limits for capacity expansion. Seagate).
For the next few years, higher prices, longer delivery times, and more intense competition for storage resources are likely, both in the enterprise segment and in the consumer sector. (Tom’s Hardware, Tom’s Hardware). At the same time, the ecological dimension – particularly energy and water consumption – is forcing operators and policymakers to think about storage strategies not only economically but also in terms of resource policy. (Goldman Sachs, Brookings).
The good news: the crisis is not a static state, but a pressure that drives innovation and discipline. Those who consistently question their data architecture, wisely combine storage tiers, and intentionally build AI so that it works with less, better-structured data, can not only reduce costs and risks but also contribute to making storage in the AI age a manageable rather than purely scarce factor.