Module 1: What HCI Is and Why It Exists

The Promise

By the end of this module you will:

1. Explain HCI without using a single marketing word. Out loud, on a customer's whiteboard, in two minutes.
2. Pass every HCI-fundamentals question on NCA. This module covers roughly 12% of the NCA blueprint, the part that builds the entire foundation for everything else on the exam.
3. Walk into a discovery call with three questions ready that determine whether the customer should be looking at HCI in the first place.
4. Recognize when HCI is the wrong answer, and say so to a customer. That posture buys you trust the rest of the deal runs on.

Here's the thing about HCI as a category. Everyone you talk to has an opinion. Most of those opinions are five years old. Your job, before you can sell or design anything specific to Nutanix, is to know the category cold. Until you do, every Nutanix-specific concept will land softly. After you do, the rest of this curriculum compounds.

This module is the ground.

Foundation: What You Already Know

You manage (or have managed) a vSphere environment. Right now, on a whiteboard, you can sketch the physical and logical layers without thinking:

1. A rack (or many) of ESXi hosts. Compute.
2. A storage array. FC or iSCSI. NetApp, Pure, Dell, EMC. Storage.
3. A network fabric tying them together. Top-of-rack switches, often a separate storage network. Network.

Three tiers. Three vendors, often. Three teams: compute, storage, network. Three refresh cycles. Three support contracts. Three sets of firmware. Three places a problem can be hiding when a VM goes slow at 3am.

This three-tier model has been the dominant pattern in enterprise virtualization for two decades. It is not stupid. It exists because, historically, building a really good storage array required different silicon, different software, different expertise than building a really good compute server. Specialization paid off.

The Foundation question is: what changed?

Two things changed.

First, x86 servers got obscenely fast and obscenely cheap. The CPU in a modern dual-socket server has more capability than a 2010-era midrange storage array's controller, and there are 64 of them in a rack. The hardware that used to be "specialized storage controller" is now "what you have lying around."

Second, software got better at distributed coordination. Distributed file systems, consensus protocols, software-defined networking. The kind of thing Google was doing in 2005 became table stakes for infrastructure software by 2015.

Hold those two facts in your head. They are why HCI exists, and they are the answer when a customer asks you why this is a category and not a fad.

Core Content

What HCI Actually Is (Plain English Version)

Hyperconverged infrastructure is a software architecture in which compute, storage, and (sometimes) networking are delivered by software running on the same physical servers, with no separate storage array.

Read that again. There is no SAN. There is no array. There are servers, each with local disks, and software that takes those local disks across all servers and presents them as a single shared pool.

That's it.

The compute layer is your hypervisor (ESXi, AHV, Hyper-V). The storage layer is software running on each host that pools the local disks. The pool is shared across the cluster. Every VM sees one logical "datastore" (or its equivalent) regardless of which host it runs on.

The Cycle, Frame Two: The Procurement Angle

You have lived the three-tier procurement cycle. Compute refresh in year one. Storage refresh in year three. They never align. Your storage vendor sells you a controller upgrade that requires a forklift migration. Your network team needs to certify firmware against both the new compute and the existing storage. The Visio diagrams take six months to update.

HCI collapses this. One platform. One refresh cadence. You add capacity by adding nodes, which give you more compute and more storage at the same time, in the same proportion, from the same vendor, with one upgrade procedure (LCM, in Nutanix-land, Module 4).

This is the procurement pitch. It is not wrong. It is also not the most interesting part.

The Cycle, Frame Three: The Architectural Angle

In a three-tier world, compute and storage have independent fate. A storage array failure takes down storage but not compute. A compute host failure takes down compute but not storage. Your HA model treats them as separate things to protect.

In an HCI world, compute and storage share fate at the node level. A node failure is both a compute failure and a (partial) storage failure simultaneously. The software has to handle this. It does, by replicating data across nodes so that any single node loss is recoverable. But the architectural assumption is fundamentally different.

This is why HCI is interesting and also why HCI is, in some scenarios, the wrong answer. We will return to this.

Diagram: Three-Tier vs HCI

Scale-Up vs Scale-Out

The next concept the reader needs is the shape of growth.

Three-tier storage scales up. When the array fills, you buy bigger controllers and more shelves. Eventually you hit the controller's ceiling and you replace the array. There is a top end.

HCI scales out. To grow capacity or performance, you add a node. Adding a node adds CPU, RAM, and storage proportionally. The cluster gets bigger, not denser. There is no controller to outgrow because there is no controller.

This sounds like a small distinction. It is not. It changes how you forecast capacity, how you justify spend, and how you negotiate with finance. In the three-tier world, you spend big once every three to five years. In the HCI world, you spend smaller amounts more frequently and you tie growth directly to demand.

Building the Fence: What HCI Is Not

Confusion in this category is rampant because every vendor uses the word "converged" or "hyperconverged" to mean something slightly different. Build the fence:

HCI is not "converged infrastructure" (CI). Converged infrastructure (VCE/VxBlock, FlexPod, UCS-with-NetApp reference architectures) is prepackaged three-tier. Still a separate array, still separate compute, still separate fabric. The vendor integrated procurement and support. Every part is still distinguishable. HCI removes the array.

HCI is not "all-in-one appliances" of the SMB variety. A $5,000 box that runs a hypervisor with local disks is just a server. HCI is specifically a clustered, distributed, software-defined storage architecture. Single-node "HCI" claims are mostly marketing.

HCI is not the same as software-defined storage (SDS) on its own. SDS is a component. Pure SDS solutions (Ceph, GlusterFS, ScaleIO, StorPool) provide a distributed storage layer but don't necessarily integrate with a hypervisor management plane. HCI is SDS plus virtualization plus management, sold as one operational unit.

HCI is not composable infrastructure. Composable (HPE Synergy and similar) lets you assemble logical servers from disaggregated pools of compute, storage, and fabric over a high-speed backplane. It is, in a sense, the architectural opposite: where HCI fuses the layers together at the node, composable splits them apart at the rack. Both claim flexibility. They go in opposite directions.

HCI is not cloud. Running an HCI cluster in your datacenter does not give you AWS. You still own the hardware, the lifecycle, the capacity planning. HCI does, however, present a shared substrate that makes it easier to consume like a cloud (single API, declarative provisioning), which is why most "private cloud" implementations sit on HCI today.

Where HCI Is the Wrong Answer

This is the section nobody at a vendor conference will give you. Read it twice. As a BlueAlly SA, this is the section that buys you customer trust. When you say "actually, for that workload, HCI is the wrong answer," the customer hears that you're not selling. After that, when you do recommend Nutanix for the rest of the workloads, the recommendation lands with weight.

1. Workloads with extreme storage-to-compute imbalance. Backup repositories that need petabytes of cheap, slow storage and almost no compute. Massive media archives. These workloads still belong on dedicated, dense storage. Putting them on HCI means buying a lot of CPUs you will not use just to get the disks attached. (BlueAlly play: in these conversations, position Nutanix Files / Mine for backup or partner with a storage-heavy partner. Don't force HCI where it doesn't fit.)

2. Workloads with hard, predictable, latency-sensitive performance contracts. A Tier-0 OLTP database with a 99.999% latency SLA at the single-digit-millisecond level may still benefit from a dedicated all-flash array. HCI has gotten dramatically better here, but for the genuine top of the performance pyramid, dedicated arrays still win on tail latency. (Note: NVMe-heavy modern Nutanix nodes close most of this gap. Have specifics ready: AOS 7.5 + all-NVMe routinely delivers sub-millisecond latency for OLTP. Don't concede more than the data forces you to.)

3. Environments where storage and compute teams are politically separate and will remain so. This is not a technology problem; it is an org problem. HCI assumes one team owns the platform. If your customer's storage team will fight to the death to keep their kingdom, HCI causes damage that exceeds the technical wins. (Discovery question: "Who would own the platform on day one, and is that a fight or a foregone conclusion?")

4. Very small environments. Two or three VMs do not need distributed storage. A pair of clustered servers with shared storage (or local disks and a backup) is simpler and cheaper. HCI pays off when you have at least three nodes, ideally more. (There is a 2-node ROBO Nutanix offering, but the economics only make sense above a certain scale.)

5. Cases where the workload genuinely belongs in cloud. If the question is "should this run on HCI in our datacenter or in AWS," the answer might be AWS, or NC2 (Nutanix on AWS bare metal, Module 8). HCI is a compelling on-prem story; it is not always a compelling on-prem-vs-cloud story.

Why HCI Wins Where It Wins

The other side, with equal honesty.

Operational simplicity is not a hand-wave. When you halve the number of vendors, halve the number of upgrade cycles, halve the number of escalation paths, you get back team capacity that you can apply elsewhere. This compounds. Real customers report 50-70% reduction in infrastructure operational hours after migration. (Source: standard Nutanix customer-evidence anecdotes; verify against your own customer base before quoting specific percentages.)

The economics of scale-out match the economics of demand. Capacity grows in the same shape as need. You do not pre-pay for a forklift you might not use.

Software-driven means automation-friendly. Distributed storage exposing a clean API (Nutanix v4 REST) is dramatically easier to automate than a SAN with a vendor-specific CLI and a CIM provider. Infrastructure-as-code is the natural fit.

One control plane is genuinely better than three. Once you have used a single management console that knows about VMs, storage, and capacity simultaneously, going back to "open vCenter, open the array UI, open Solarwinds, correlate manually" is painful.

The Honest Verdict on HCI as a Category

HCI is not a religion. It is an architectural pattern that fits a wide and growing band of enterprise workloads and does not fit a narrow but real set of others. Treat it as a tool. Match the tool to the job.

For the BlueAlly SA in 2026, the relevant question is not "is HCI good." The relevant question is: which HCI, for which workloads, under what licensing model, and given Broadcom's VMware pricing, what is the migration path and what is the real cost to the customer over five years. We will spend the rest of this curriculum answering all four.

Lab Exercise: Stand Up Your First CE Cluster

Goal of this lab: Get Nutanix CE running so you can touch every concept in this curriculum yourself. This is the foundation lab. Every subsequent module assumes you have a working CE environment.

Two paths:

Path A, Nested on VMware Workstation or ESXi (recommended for first-time setup)

1. Download CE 2.1 from nutanix.com/products/community-edition (requires free MyNutanix account)
2. Provision a VM: 4 vCPU, 32 GB RAM recommended (the documented absolute minimum is 16 GB, but CE running Prism Central is realistically a 32 GB+ workload). Expose hardware-assisted virtualization, VMXNET3 NIC, vSwitch promiscuous + MAC changes + forged transmits all enabled
3. Add disks: 1× 32GB (boot), 1× 200GB SSD-backed (hot tier), 1× 500GB (cold tier)
4. Boot from the CE ISO. Walk through the installer.
5. After install, connect to Prism Element on https://<cluster-IP>:9440
6. Default creds: admin / nutanix/4u, change immediately

Path B, Bare metal (best long-term lab, more work up front)

• Used Intel-based desktop with VT-x, ≥32 GB RAM, ≥1 SSD + 1 HDD, 1 GbE NIC
• Same install process; install directly to USB or SATA-DOM boot device

What to do once it's up:

1. Log into Prism Element. Click around. The dashboard is your first introduction to the Prism aesthetic.
2. Find the Cluster Details page. Note the Cluster Virtual IP, AOS version, AHV version.
3. Open the Hardware view. See "your one node" represented.
4. Open Storage > Storage Containers. Note the default container.
5. Open VM > Create VM. Don't create one yet, just look at the form. Compare it mentally to vCenter's New-VM wizard.

Bookmark this: every module from here on assumes you can reproduce its concept in your CE lab. If you skip the lab, the rest of this curriculum is entertainment, not training.

Practice Questions

Ten questions. Six are NCA-style knowledge or NCP-MCI-style scenario MCQs. Two are short-answer / open-ended in the style of NCX-MCI design defense. Read each, answer in your head, then read the explanation.

What You Now Have

You can now define HCI without saying "hyperconverged" twice and calling it done. You know it is fundamentally a software architecture, not a hardware product. You know it eliminates the separate storage array by distributing storage across the same nodes that run compute. You know it scales out, not up, and you know what that implies for procurement and capacity planning.

You can now name five scenarios where HCI is the wrong answer. That alone puts you ahead of most people who pitch it. It also gives you a sales posture customers will trust.

You have the fence: HCI is not CI, not just SDS, not composable, not cloud, not all-in-one. When the next customer's incumbent vendor uses one of those words interchangeably, you will catch it.

You have the SA-chair language for the most common HCI objection ("we tried it five years ago") and for the most common opening positioning question.

You have ten practice questions worth of HCI-category discrimination. Eight test recognition and scenario judgment in NCA and NCP-MCI style. Two are open-ended NCX-MCI-style design defenses, covering staged-migration recommendation under organizational constraint and architectural defense against a fault-domain critique. The first eight earn you points on NCA. The last two are practice for the kind of architectural reasoning the NCX-MCI panel actually evaluates, and for the kind of conversation that wins or loses customer trust in front of a sharp infrastructure architect.

You are now ready to look at a specific HCI implementation, Nutanix, and see what choices it made. That is Module 2.