{
  "id": 6567295,
  "title": "AI’s overlooked storage opportunity",
  "url": "https://urgent.news/2026/09/10/ais-overlooked-storage-opportunity",
  "topic": "ai",
  "section": "AI",
  "published": "2026-09-10T09:12:33.000Z",
  "source": {
    "name": "TechRadar",
    "slug": "techradar",
    "url": "https://www.techradar.com/pro/ais-overlooked-storage-opportunity"
  },
  "original_language": "en",
  "account": "The AI infrastructure discussion predominantly revolves around data centers, power consumption, and computational resources necessary for both training and running models, including GPUs and high-performance storage. This focus is understandable given the massive investment figures that dominate the headlines. However, data is another critical resource, often overlooked in these discussions. According to Stanford University's 2025 AI Index Report, dataset sizes for training large language models (LLMs) are doubling every eight months. As models are developed, some data moves swiftly into curation and model-development environments, where rapid access is vital. Yet, a significant portion remains long-term, awaiting relevance to a specific AI use case. This transition phase indicates that data does not require the same level of performance at all stages of the AI pipeline. The key insight is that a dataset needs to be ready when it is needed, not constantly on high-performance infrastructure throughout, which can become prohibitively expensive at scale. The strategic question is not whether AI requires fast storage, but where organizations should retain the massive datasets required in future, before they are ready for processing. This decision is strategic, not merely housekeeping. Organizations should select the appropriate capacity tier, ensuring data is protected and readily recoverable into AI, training, and transformation pipelines, while allocating performance only where the work actually takes place. This approach maximizes the budget. Each organization is unique, located at various stages in their AI journey. Some have already incorporated AI systems into production, while others are still exploring how their existing data could support AI initiatives, ranging from documents and images to operational records. This realization underscores that proprietary data is a competitive advantage, contingent on its accessibility and scalability. Organizations that swiftly identify their data holdings, their locations, and their readiness for use are likely to move faster in the AI race. Hence, data is not merely an input to AI; it fundamentally shapes the models. The more an organization can leverage its own data, the more precise and effective its AI tools become. Traditionally, large datasets were stored in disk-based data lakes until required for further processing. However, as datasets grow larger, costs escalate. Keeping every candidate dataset on always-on, high-performance infrastructure restricts the amount of data an organization can afford to retain. The challenge lies in ensuring all data is accessible when needed, with cost as the deciding factor, not the storage bill. A cost-effective solution is tape storage, a technology that has consistently evolved. Recent LTO generations have enhanced its capacity, throughput, and security while the industry has shifted focus. Modern tape systems function akin to other tiers in the stack, ready to stream data into fast storage when curation or training is ready. Tape's economics improve as its scale increases. At the multi-petabyte scale common in AI programs, tape's cost per terabyte is significantly lower than flash or even HDD infrastructure. Performance and capacity can also be scaled independently, enabling more drives for throughput and additional cartridges for capacity. When considering tape's energy efficiency along with its cost advantages, it emerges as a strategic storage technology to integrate into data centers. This allows organizations to maintain their entire data estate in play, at a predictable cost. Tape's role extends beyond data storage; it provides a secure environment for valuable data, handling encryption and ensuring datasets are immutable. For highly sensitive information, tape sets can be physically moved to secure, offsite locations, rendering them offline and air-gapped, thus safeguarding them from potential threats within the production environment. In building data and AI pipelines, the focus should be on ensuring data is prepared for rapid transfer into the appropriate performance tier upon need. Data serves as the fuel for the models, business decisions, and speed of action. By leveraging tape storage, organizations can affordably retain 'data fuel' at scale, protect irreplaceable information, and employ it on demand as needed. Embracing tape storage in the data infrastructure enables organizations to avoid limitations imposed by online storage costs, positioning them ahead of competitors.",
  "summary": "AI success depends on keeping more data accessible, protected, and affordable at scale.",
  "key_points": [],
  "editors_take": null,
  "illustration": null,
  "coverage": {
    "outlets": 1,
    "also_reported_by": []
  },
  "ai_generated": true,
  "disclaimer": "Summaries, key points and the editor’s take are written by software from other outlets’ reporting and may contain errors — always check the linked original."
}