NetApp Sellers & Partners

Joint Educational Series

Field Brief · The Deployment Decision

Where Should Your AI Actually Run?

Public, private and national AI all promise control. Most companies already use more than one, often without a plan for which job belongs where.

What you'll learn

The two questions that sort every AI job: how sensitive the data is, and how often it runs

Which everyday jobs belong in the public cloud, and which belong behind private walls

Why public, private and national AI compare like renting versus owning

The core argument

Using public AI is like renting an apartment. Owning your AI is like buying a house. Some countries are building a whole city — and even they rent a few things from someone else.

Public AI

Rent an apartment

Fast to move into. The landlord sets the rules, and the walls are thin.

Private AI

Buy a house

More work up front. Every wall is yours, and nobody can look in.

Sovereign AI

Build a city

A nation's own stack — though most still lean on a foreign partner somewhere.

Key facts

The average enterprise already runs 4.2 AI models in production.

Renting AI and owning it cost very differently over time.

“Sovereign AI” often still depends on foreign infrastructure.

Workloads can be sorted by which belongs in public cloud, private walls, or both.

By the numbers

67/22/11%

Enterprise AI split: public / hybrid / on-premises

4.2

AI models the average enterprise runs, up from 1.9

~70%

Of “sovereign” AI projects still rely on a foreign partner

The decision may already be made for you

While the architecture is debated, staff are quietly picking public AI on their own — and taking company data with them.

66%

Of office professionals have used AI at work believing it was against policy

34%

Have entered customer data into public AI tools

4×

Rise in shadow-AI detections in a single year

PagerDuty 2026 Shadow AI Survey (Wakefield Research, 1,250 office professionals, companies over $500M revenue); Verizon 2026 DBIR, which now ranks shadow AI the third most common non-malicious insider action.

Renting vs. owning your AI house

Most companies are renting when they could be buying. The AIPod Mini makes owning your own AI house just as fast as renting an apartment.

NetApp | Iterate.aiWhere Should Your AI Actually Run? · 01 / 02

Which job belongs where

This is rarely all-or-nothing. Most companies should run both, and sort the work by two questions: how sensitive is the data, and how often does the job run. Sensitive or constant belongs inside. Neither, and the cloud is fine.

The job	Belongs in	Why
Drafting public marketing copy	Public cloud	No private data, occasional use.
Summarising patient or client records	Private walls	Regulated data that cannot leave.
Answering staff questions on policy	Private walls	Internal only, and runs all day.
Translating a public web page	Public cloud	Nothing confidential in it.
Reviewing contracts for risk	Private walls	Your terms are your position.
Brainstorming with a general model	Either	Fine, provided nothing internal is pasted in.
Agents running against your files	Private walls	Constant use, and reads everything.

The pattern: anything touching your own records, or running constantly, belongs inside.

Four questions that place any workload

How sensitive is the data? Health records, financials and trade secrets point inside by default.

What does the regulator demand? Healthcare, finance and government often have the decision made for them.

How often does it run? Twice a month is cheap in the cloud. All day, every day, is where metered pricing turns painful.

How strategic is it? A source of competitive advantage deserves the investment private AI requires.

The list most companies never made

The question was never public or private. It is which work belongs where — and most companies have never made that list.