A director of infrastructure has a data center lease ending in 30 months, a storage array nearing end of support, and a CFO asking why one business unit's cloud bill doubled. What she has rarely heard from sellers is a clear account of what actually changes for her organization if more of its computing moves to the cloud: who will be allowed to spend, who will be accountable when something breaks, how the bill will behave, and what it would cost to leave.

That is the question this article answers for the seller. Cloud computing is usually explained as "someone else's servers." That description is accurate and almost useless in a deal, because buyers do not purchase servers when they buy cloud. They purchase a different way of running technology: a different operating model, with different controls, risks, and economics. The seller who can explain that shift in plain language earns the right to sit with the architect and the CFO at the same table.

The most widely cited definition comes from the US National Institute of Standards and Technology. NIST Special Publication 800-145 (Mell and Grance, September 2011) defines cloud computing as a model for enabling convenient, on-demand network access to a shared pool of configurable computing resources (networks, servers, storage, applications and services) that can be rapidly provisioned and released with minimal management effort or service provider interaction [1]. It is fifteen years old and still the reference point for regulators, auditors and procurement teams.

NIST names five essential characteristics [1]:

NIST also defines three service models, which describe how much of the technology stack the provider runs. In Infrastructure as a Service (IaaS) the customer rents compute, storage and networking and runs its own operating systems and applications. In Platform as a Service (PaaS) the customer deploys its own applications on tools and runtimes the provider manages. In Software as a Service (SaaS) the customer uses the provider's finished application. And there are four deployment models: public cloud (open to the general public, on the provider's premises), private cloud (for one organization, on or off its premises), community cloud (shared by organizations with common concerns), and hybrid cloud (two or more distinct clouds bound together so data and applications can move between them) [1].

The definition does not say "off premises" or "cheaper." A private cloud in the customer's own building can meet the definition; a rack of rented servers that takes six weeks and a purchase order to expand does not. Cloud is defined by how resources are obtained, shared, scaled and measured, not by where they sit.

Each of the five characteristics changes who in the buying organization holds a decision that used to belong to someone else. That is why cloud deals stall in places sellers do not expect.

Self-service moves the power to spend from a central procurement and infrastructure gate to whoever holds an account. That is the source of the speed engineering leaders want, and of the uncontrolled spend finance leaders fear. Resource pooling moves part of the security job to the provider. AWS's own documentation describes this as the provider being responsible for "security of the cloud" (the hardware, software, networking and facilities that run its services) while "security in the cloud" remains the customer's, with the customer's share determined by which services it selects [2]. For a virtual server, the customer still patches the guest operating system and configures the firewall; for a managed storage or database service, the customer still classifies its data, sets encryption options and manages access permissions [2]. Every provider draws this line in its own documentation, and the seller should check where it falls for the services in scope; the line moves with the service model, but it never disappears.

Elasticity moves capacity planning from a multi-year forecast to a near real-time decision, which changes the infrastructure team's job. Measured service turns a capital purchase that depreciates on a schedule into a variable operating cost that moves with usage, which changes the CFO's job. And because a hybrid environment is the norm rather than the exception (Gartner predicted in November 2024 that 90% of organizations will adopt a hybrid cloud approach through 2027 [4]), the buyer is rarely deciding "cloud or not." The buyer is deciding which workloads run under which operating model, and who governs the seams between them.

This is where the general selling principle applies directly. The Sales Agent article on Value Creation in B2B Selling: What the Evidence Says It Actually Means argues that value lives inside the customer's operation, not inside the product, and that a value claim must pass four tests: Anchor, Alternative, Arithmetic and Advocate. Cloud features are not value. "Rapid elasticity" becomes value only when it is anchored to a metric the customer already tracks (for example, weeks to provision an environment, or revenue lost during a seasonal peak), stated against the real alternative, written as arithmetic the buyer can edit, and carried by someone inside the account. As What Is Selling? An Evidence-Based Answer for 2026 argues, buyers are informed but overwhelmed; the seller's job is to show which of the five shifts matters to this buyer.

The market is large, concentrated and growing faster again. Synergy Research Group reported that enterprise spending on cloud infrastructure services (IaaS, PaaS and hosted private cloud) reached $143.4 billion in Q2 2026, up 43% year on year, with trailing twelve-month revenue of $500 billion [3]. Amazon, Microsoft and Google held 28%, 20% and 15% of that market respectively, and Synergy attributed the acceleration primarily to generative AI, with GenAI-specific cloud services growing 165% year on year (analyst data, as of July 2026) [3]. Gartner's broader measure of end-user public cloud spending, which includes SaaS, was forecast in November 2024 at $723.4 billion for 2025, up from $595.7 billion in 2024 (analyst forecast) [4]. The two figures measure different things and should never be quoted interchangeably.

Waste is real and has recently risen. Flexera's 2026 State of the Cloud Report estimates that 29% of IaaS and PaaS spend is wasted, reversing a five-year downward trend, and that 85% of respondents say managing cloud costs is their top challenge, ahead of security at 82% (vendor data, summarized by Flexera in September 2026; the summary does not state the sample size) [5]. The same summary reports that 73% of organizations operate hybrid estates and that fewer than half use any one commitment discount per provider [5]. Treat these as self-reported estimates from a vendor that sells cost tools; the direction still matters, because the variable bill is the part of the operating model buyers trust least.

How customers are supported after adoption changes how much they use. The most useful peer-reviewed evidence for sellers comes from Retana, Forman, Narasimhan, Niculescu and Wu in MIS Quarterly (2018), using usage data from 22,179 firms on one public cloud infrastructure provider between March 2009 and August 2012 [6]. Buyers who received full support (education, training and personalized guidance, not just basic support) used the service 34.85% more than other users and ran a larger share of their servers in parallel, a sign of more efficient, cloud-appropriate architecture. Buyers who later dropped back to basic support still used 15.01% more than buyers who never had full support, which the authors read as evidence of lasting buyer learning [6]. The limits matter: it is one provider, an early period of the market, and an observational design. Still, it fits the definition: an operating model has to be learned.

Outsourcing infrastructure does not outsource outage risk. Uptime Institute's 2026 outage analysis reports that third-party IT and data center service providers, including cloud and internet companies, telecommunications and colocation firms, accounted for about two-thirds of publicly reported outages across the nine years it has tracked them [7]. In its 2025 survey, 57% of respondents said their most recent major outage cost more than $100,000, and one in five reported costs above $1 million [7].

Selling servers instead of the operating model. A proposal that compares the price of on-premises hardware with the price of equivalent virtual machines invites a commodity comparison and ignores the real decisions: governance of self-service, the security split, and the behavior of a variable bill.

Promising savings the operating model cannot deliver by itself. Elasticity saves money only if someone turns capacity off. Measured service saves money only if someone reads the meter. A seller who promises a percentage reduction without naming who will do that work is making a claim the buyer's finance team will later use as evidence against the seller.

Treating shared responsibility as the provider's responsibility. Buyers sometimes assume that moving to cloud moves security and compliance obligations with it. The provider documentation says otherwise [2]. When the security team discovers this late, the deal stalls in a review the seller never planned for.

Ignoring exit costs. Data egress charges (fees for moving data out of a provider's network) are a common hidden cost and a common source of lock-in fear. The market has started to move: Google Cloud announced in January 2024 that customers leaving its platform could migrate data out with free network data transfer [8], and the EU Data Act, in force since 11 January 2024 [10], provides for switching charges, specifically including data egress charges, to be abolished three years after entry into force, with only reduced charges permitted until then [9]. A seller who addresses exit honestly, including what still costs money in staff time and re-architecture, removes an objection; a seller who avoids it confirms the fear.

Single-threading the deal. Because each characteristic moves a decision to a different person, a deal sold only to infrastructure meets the CFO, CISO and procurement later, on their terms.

The Operating Model Map is a reusable way to turn the NIST definition into a discovery and qualification tool. For each of the five characteristics, ask what changes, who now owns it, and what proof that person will need.

A deal where the seller can name the owner and the open question in every row is a deal the seller understands. A row with no owner is the place the deal will stall.

For account executives: stop leading with services and start leading with the operating model shift. Ask which of the five characteristics matters most to this buyer's compelling event, and build the business case around that one. Use the Four Tests from the value creation article on every savings claim before it leaves your hands.

For sales engineers and architects: own the Protect and Plan rows. Map the provider's shared responsibility model to the customer's control framework early, and separate workloads that benefit from elasticity from those that do not. Being candid that a steady workload may stay on premises builds the credibility you will need when you recommend moving the variable ones.

For partners: the Retana evidence is your strongest argument. Customers who receive education and guided support after adoption use more of the service and architect it better [6]. Position enablement, governance and FinOps services as part of the operating model the customer is buying, not as an optional add-on.

For sales leaders: inspect deals against the map. A forecast that names the owners of Pay and Protect is more reliable than one that names only the champion. Coach teams to raise exit costs and shared responsibility early.

Cloud computing is a way of obtaining and governing technology. Sellers who help the buyer assign each new responsibility sell more durably than sellers who explain the servers.