A large enterprise stores petabytes, and the total grows every year. Finished projects, video, logs, backups kept for the record, research output, and object data written by applications that rarely read it again.
Much of that data is cold. It is kept because it might be needed, because a policy says so, or because nobody is sure it can be deleted. Yet it usually sits on the same continuously spinning disk as active data, drawing power and cooling every hour of the year, and it has to be migrated to new hardware every few years.
The alternative many organizations reach for, a cloud archive tier, trades those costs for others: retrieval and egress charges, restore times measured in hours, and dependence on a provider's pricing over decades.
Cold is a pattern, not an age
The first step is knowing which data is actually cold. Age is a weak proxy: some old data is read constantly, and some new data is never read at all. Access history over the past year, by data class, is a far better guide.
The data that matters most here is cold but must stay online: data that is rarely read, but when it is needed, it needs to be retrievable in minutes through the same interface applications already use. That is different from data that can be deleted, and different from data that can sit in a deep archive for days.
An online optical tier for cold data
Scale-out ELS libraries, managed through oRain, give cold data an online home on optical media. Written media needs no power to retain its data. Libraries draw power to retrieve and write, not to keep data spinning.
Objects remain in one namespace and stay accessible through the S3-compatible interface oRain provides, so applications do not need to know which tier an object lives on. Active data stays on performance storage. Capacity grows by adding networked libraries rather than rebuilding the archive, and write-once media is not rewritten on a refresh cycle.
This suits data that is cold but must remain online. Data that needs continuous high-throughput access belongs on performance storage, and data that can be deleted should be deleted.
Energy reporting makes the cost visible
In the European Union, Article 12 of the Energy Efficiency Directive (EU) 2023/1791 requires owners and operators of data centers with at least 500 kW of installed IT power demand to publish energy-performance information every year, starting 15 May 2024. The required information includes energy consumption, power use, renewable energy share, waste heat, water use, and the amount of data stored and processed. Delegated Regulation (EU) 2024/1364 defines the indicators. Data centers used exclusively for defense and civil protection are exempt.
The rules are moving further. In September 2026 the European Commission proposed a common EU rating scheme for data centers and opened consultation on minimum performance standards, with a legislative proposal planned for 2027. These are not yet final requirements, but they indicate the direction.
Germany has gone furthest so far. Section 11 of its Energy Efficiency Act (EnEfG) sets power usage effectiveness limits: existing data centers must reach 1.5 from July 2027 and 1.3 from July 2030, and facilities starting operation from July 2026 must reach 1.2, along with requirements for waste-heat reuse and renewable electricity.
Measure the right thing
Moving cold data off spinning disk reduces IT energy and the cooling that goes with it. That shows up in total energy consumption and in energy per stored terabyte.
It does not, by itself, improve power usage effectiveness. PUE is the ratio of total facility energy to IT energy, as defined in ISO/IEC 30134-2. Lowering IT load reduces both sides of that ratio, and if facility overhead does not fall proportionally, PUE can rise even as total energy falls. Operators facing PUE limits need to address facility efficiency directly; reducing stored-data energy is a separate, complementary gain.
Results also depend on specifics: the access pattern of the data moved, the systems it was moved from, and how often it is recalled. Measure consumption before and after a pilot, and report measured figures rather than vendor estimates. That is also what reporting regimes ultimately expect.
A practical starting point
Pick one large data class. Record its size, annual growth, access frequency over the past year, current storage tier, share of power and cooling, refresh schedule, and any cloud retrieval costs. Agree with the business on the access time it actually needs.
Pilot moving that class to an ELS library. Measure retrieval times against the agreed target, and measure energy before and after. Use those measured results to decide how far to extend the approach, and to report the change accurately.
The goal is not to move everything to optical. It is to stop paying, every hour of every year, to keep data spinning that nobody is reading.