The development of LLMs, data replication and the extension of data retention periods very significantly increase storage requirements. This requires an increase in investment
in this area.
Storage is the second most important component of an AI infrastructure according to respondents to the latest Seagate survey by Recon Analytics, behind security. A result that should not be surprising from a major supplier of storage media. Note that the companies to which the respondents belong report an annual turnover greater than $10 million and a current storage capacity greater than 50 TB.
The study found that six in ten solution buyers primarily use cloud storage for AI data management and expect storage needs to at least double by 2028. There are three reasons for this . On the one hand, the data retention period is longer, from six months to several years. On the other hand, 73% of panel members use daily or weekly LLM checkpoints to store critical parameters. Finally, 80% of respondents rate data replication for AI as very or moderately important.
Of those surveyed who back up checkpoints daily, or 28% of respondents, nearly one in three keep data for more than 12 months, while 29% of respondents keep it for 6 to 12 months.
Increased use of the cloud to meet needs
The graph below shows the solutions adopted to address the growing storage needs due to the use of AI, based on revenue. In short, the cloud is mainly used by companies whose turnover is between 1 billion dollars and 500 million dollars.
Chart source: Recon analytics
-According to the study, cloud storage is expected to remain the main storage vector for AI, with 65% of data stored in the ‘cloud’ rather than in-house in 2024
and 69% in 2028. Storage needs are expected to double over the next three years in hybrid or on-premises mode.
Most companies that record and save checkpoints do so on a daily or weekly basis on the cloud. This concerns organizations that manage more than 100 PB of data, considering that data replication improves AI results very significantly.
Energy needs are increasing accordingly
However, a quarter of respondents say security is a priority, followed by 18% for storage. Two out of three list storage among their top four infrastructure concerns. Please note that energy consumption due to AI is also growing drastically, alongside the increase in storage capacities, a point which cannot be neglected and isolated from a global environmental and technical issue.
One way to reduce the volume of stored data is to eliminate the very numerous data that are unusable. Data governance work that must be seriously carried out by those concerned.
Related News :