The IT director of a manufacturing company faces the end of life of the server room and two contradictory signals. The board asks why not move everything to the cloud, since “there you only pay for what you use”. The legal department reminds him that some production-system data should not leave the plant. And the three-year growth plan is so uncertain that every number in a hardware request for quote will be a guess.
Between buying hardware and the public cloud there is a third way: the vendor’s infrastructure standing in your own server room, billed by usage. In this article we explain what this model shifts to the vendor and what it does not, how to compare it with buying without promising savings, and what to put in the contract. HPE GreenLake serves as the example, but the questions apply to any offer of this kind.
What is infrastructure as a service, and does it have to mean the public cloud?
Infrastructure as a service is a service model, not the place where the hardware stands. In SP 800-145 “The NIST Definition of Cloud Computing” (September 2011), NIST defines IaaS as the capability to provision “processing, storage, networks, and other fundamental computing resources” on which the customer runs its own software, including operating systems. The customer does not manage the underlying infrastructure, but controls the systems, data and applications.
The same publication describes a private cloud as infrastructure provisioned for a single organisation. It may be owned and managed by the organisation, a third party or both, and may exist on or off the organisation’s premises. Nothing in the definition requires the hardware to sit with a public cloud provider.
It is also worth separating IaaS from a similar-sounding concept. Infrastructure as code is a way of configuring resources through files that an automated tool deploys — we describe it in our compendium on Infrastructure as Code. You will also find the definition of the service model in our glossary under IaaS.
IaaS ≠ IaC in one sentence: infrastructure as a service says how resources are delivered and billed; infrastructure as code says how they are configured.
How does the consumption model work in your own server room?
The vendor installs its infrastructure at your site, and you pay for its use instead of buying it up front. Among the five essential characteristics of cloud computing, NIST lists measured service: resource usage is monitored and reported, providing transparency for both parties. The consumption model brings this characteristic into your own server room.
The second characteristic is rapid elasticity. In the public cloud it comes from the provider’s scale. In your own server room it comes from a buffer — hardware installed in reserve, which you start paying for only when you use it. HPE writes that every GreenLake Flex Solution includes a capacity buffer and that it manages capacity together with the customer, adding resources before they are needed. HPE does not publish the size of the buffer — the contract sets it.
The model differs from leasing in that the payment is not fixed. With a lease you pay the same regardless of utilisation. In the consumption model you pay more when you use more, but usually within a minimum commitment below which the bill does not fall.
What distinguishes it from the public cloud is where the data is and who has physical access to it. The hardware stands in your server room, colocation facility or at the edge, so the data does not leave the specified location. We compare the classic cloud models in the article on public, private and hybrid cloud.
What does HPE GreenLake cover today: platform, Flex Solutions and private cloud?
Three things worth distinguishing, because each answers a different question. The first is the platform. HPE describes GreenLake as the cloud that delivers unified management of a hybrid environment — compute, storage, networking, private cloud and edge. The platform includes core services, such as identity and access management and APIs, and additional ones, for example cost optimisation (FinOps).
The second is GreenLake Flex Solutions, the actual consumption model. HPE groups it by workload: ProLiant servers and the Synergy platform, SAP and databases, virtualisation and hyperconverged infrastructure, container platforms, data protection, AI and analytics, and edge solutions. You either manage operations yourself through the platform or choose an option in which HPE takes over part of them.
The third is private cloud. HPE Private Cloud PC7000 is a private cloud with pay-per-use billing, managed for the customer, in its data centre, colocation facility or at the edge. HPE also offers it in a variant whose management does not require an internet connection, aimed at regulated industries. PC3000 is the self-managed version, with VMware and HVM hypervisors under one console.
We keep an overview of the offer’s components and their uses on the HPE GreenLake page. According to HPE, the model works well for workloads that are hard to move to the public cloud and for companies that design a hybrid environment from the outset.
Who is responsible for what: dividing duties between the company and the vendor?
The vendor takes over the hardware and capacity, while the company keeps the data, the configuration and the access. This division follows directly from the IaaS definition in NIST SP 800-145: the customer does not manage the underlying infrastructure, but controls operating systems, storage and applications. The consumption model does not change that, even if the hardware stands in your own server room.
The vendor’s side usually covers: delivering the hardware, maintaining the capacity buffer, planning its expansion and — depending on the contract — service and firmware updates. With a managed-for-you option such as PC7000, the vendor’s scope also covers the private cloud layer.
The company’s side covers: data and its backups, configuration of guest systems, application patching, account and permission management, network segmentation and the physical security of the room. It also keeps the legal responsibility for the data it processes, which no contract transfers in full.
The most common mistake is to assume that “since the vendor manages it, it manages everything”. Every duty the contract leaves unstated has, in practice, no owner. That is why it is worth writing the division out in a matrix before signing, not after the first incident.
When is paying for usage cheaper than buying, and when is it more expensive?
When buying would mean paying up front for capacity that sits idle for a long time. With a purchase, hardware is sized for the end of its life, not for day one. If growth is uncertain, the company buys a reserve just in case and pays for it from the first day — or buys too little and adds more in a hurry.
The consumption model shifts that risk to the vendor. The vendor does not do this for free, though: the risk is priced into the unit rate and the minimum commitment. With a stable, well-understood workload that uses most of the purchased capacity throughout the period, buying is therefore often cheaper.
A fair comparison uses the same horizon for both options. On the purchase side you count the hardware price, service support, expansions during the period, the team’s time spent on maintenance and the cost of unused capacity. On the consumption side — the minimum commitment, forecast usage above it, managed services and the costs of ending the contract.
The key variable is not the unit price, but the predictability of usage. If a company can say with confidence how much capacity it will need in three years, the consumption model’s advantage shrinks. If it cannot, the advantage grows. HPE prepares a cost assessment for GreenLake Flex Solutions individually, with a representative, for the specific workloads — and that is what you compare with buying, not general claims.
How do you read the contract: capacity buffer, minimum commitment and end of term?
From the end, because the exit terms decide the real cost of the whole model. The first question: what happens when the term ends — renewal, hardware replacement, return, data migration — and how much each of these paths costs. The second: what the minimum commitment looks like and whether it changes during the term.
The third question concerns metering: in which units usage is measured, how often, and where the customer can verify it. The fourth concerns the capacity buffer: how large it is, how quickly the vendor replenishes it after it is used, and how much usage above the buffer costs before more hardware arrives.
The fifth question is often skipped, yet it is key for security: what happens to the storage media when the hardware is returned. The contract should state who removes the data, by what method and what confirmation the customer receives. The reference point is NIST SP 800-88 Rev. 2 “Guidelines for Media Sanitization” (September 2025), which describes how to build a media sanitisation programme matched to the sensitivity of the data.
Questions for the contract before signing: end-of-term terms and their cost; minimum commitment; metering units and frequency; buffer size and replenishment time; price of usage above the buffer; media sanitisation on return; scope and logging of the vendor’s remote access.
Does the as-a-service model help or hinder NIS2 requirements?
It can help, but only if the contract settles what the Directive requires. Article 21(2)(d) of NIS2 requires supply chain security, including relationships with direct suppliers and service providers. Article 21(3) also requires taking into account the vulnerabilities specific to each direct supplier and the quality of its products and cybersecurity practices.
A vendor that manages hardware in your server room and has remote access to it is exactly such a direct supplier. Its processes, service accounts and remote support channels fall within the scope of the risk assessment. We describe how to carry out this assessment under Polish rules in the article on high-risk suppliers under the Polish cybersecurity act, and explain the concept of the IT supply chain in our glossary.
The model can also help. Regular hardware refresh and firmware updates carried out by the vendor — if the contract covers them — reduce the risk of running on outdated, unpatched infrastructure. Business continuity, required by Article 21(2)(c), still depends, however, on backups and recovery procedures on the company’s side.
What does security look like when the hardware is at your site but the vendor manages it?
As in any environment with two privileged parties: you need to know who has access, from where and when. The first point is the account on the management platform. Resources are ordered and configuration is changed through it, so it requires strong authentication, a limited list of administrators and access reviews.
The second point is the vendor’s remote access. It is worth establishing which route it takes, whether it is logged and whether the company can temporarily disable it. For environments where management over the internet is unacceptable, HPE offers a PC7000 variant with management disconnected from the public network.
The third point is physical service access to the server room: who lets the technician in, who supervises them and what happens to replaced drives. The fourth is separating backups from the environment the vendor manages. A backup in the same administrative domain as production does not protect against a takeover of that domain.
Each of these points can be solved, and none is a reason to reject the model. They must, however, be solved before deployment rather than assumed to have been solved by the vendor.
How do you compare buying, the public cloud and infrastructure as a service?
Along the same criteria, because each option wins in a different place. The table does not name a winner — it shows which question decides the choice in a particular company.
| Criterion | Buying hardware | Public cloud (IaaS) | On-premises infrastructure as a service |
|---|---|---|---|
| Where the data physically is | in your own server room | in the cloud provider’s data centres | in your own server room, colocation or at the edge |
| Who bears the over-provisioning risk | the company | the cloud provider | the vendor, priced into the unit rate and minimum |
| Payment model | up front, plus service | by usage | by usage above the minimum commitment |
| Speed of adding capacity | pace of purchase and delivery | immediate, within account limits | from the buffer, then hardware delivery |
| Hardware refresh | the company’s decision and budget | on the provider’s side | per contract |
| Control over hardware configuration | full | none | limited, depending on the contract |
| Supply chain assessment (NIS2) | hardware and service suppliers | the cloud provider | a vendor present in your server room |
| Cost of exit | low, the hardware belongs to the company | data transfer and migration | end-of-contract terms |
| Best suited to | stable, predictable workload | variable workload without a location requirement | variable workload with a data location requirement |
The last row is the most important. The on-premises consumption model has an edge in one fairly narrow but common case: when growth is unpredictable and the data has to stay on site. We give a broader view of choosing between your own server room and the cloud in the guide on cloud or on-premise servers.
Where should you start assessing whether this model fits your company?
With data on your own usage, not with an offer. The first step is an inventory of workloads: which systems must stay on site, which can go to the public cloud and which can be retired. The second is measuring actual utilisation and growth over recent periods, because that shows how predictable usage is.
The third step is calculating three options over the same horizon: buying, the public cloud and the consumption model. The fourth is going through the contract using the questions from the box above, before comparing prices. The fifth is writing out the responsibility matrix and checking that every duty has an owner.
We help with the assessment on the technical and contractual side: we compare the consumption model with buying the same infrastructure, review the terms on the division of duties and security, and design the target environment as part of our server infrastructure deployments and virtualisation services. We set the GreenLake offer against the alternatives rather than assume it is the best; we describe the vendor’s portfolio on the Hewlett Packard Enterprise page.
When NOT to choose the consumption model: the workload is stable and well understood for the whole period; the company has a team that maintains its own hardware efficiently; full control over hardware configuration is essential; the exit terms of the contract are unclear.
Sources cited in the text: NIST SP 800-145 “The NIST Definition of Cloud Computing” (September 2011); NIST SP 800-88 Rev. 2 “Guidelines for Media Sanitization” (September 2025); Directive (EU) 2022/2555 (NIS2), Article 21(2)(c) and (d), Article 21(3); HPE pages: GreenLake, GreenLake Flex Solutions, GreenLake — compute, HPE Private Cloud (hpe.com, as of September 2026).
Frequently asked questions
Does infrastructure as a service always mean the public cloud?
No. In SP 800-145 (September 2011), NIST defines IaaS as a service model — providing processing, storage and networks on which the customer runs its own systems. The same publication notes that a private cloud may exist on or off the organisation's premises and may be managed by the organisation, a third party or both.
How does infrastructure as a service differ from infrastructure as code?
Infrastructure as a service (IaaS) is a model for delivering and billing resources. Infrastructure as code (IaC) is a way of configuring them — describing servers and networks in files that an automated tool deploys. One does not require the other, although they often go together.
How does the consumption model differ from hardware leasing?
With a lease you pay a fixed instalment regardless of how much hardware you use. In a consumption model such as HPE GreenLake Flex Solutions, you pay for the infrastructure you use, and the vendor keeps a capacity buffer for unexpected needs. The minimum commitment and the metering units are set by the contract.
Does HPE GreenLake ensure NIS2 compliance?
Not automatically. Article 21(2)(d) of the NIS2 Directive requires supply chain security management, and a vendor managing hardware in your server room becomes part of that chain. Compliance depends on the division of responsibilities, the configuration and the contract terms.
What happens to the drives when the contract ends?
It depends on the contract and has to be settled before signing. Good practice is to specify who removes data from the media before the hardware is returned, by what method, and what confirmation the customer receives. The reference point is NIST SP 800-88 Rev. 2 "Guidelines for Media Sanitization" (September 2025).
