Every business is being told to “use AI” — but almost nobody explains the first real decision: where should that AI actually run? On computers you own, on a cloud service like ChatGPT, or a mix of both? Get this wrong and you either overspend on hardware you didn’t need, or quietly hand a cloud provider data you were supposed to protect. This guide explains the options in plain English, then a two-minute quiz gives you a recommendation based on your situation.
The three options, in plain English
Cloud AI: you rent it
ChatGPT, Claude, Gemini, Microsoft Copilot — AI that runs on someone else’s computers and reaches you through the internet. You pay monthly or per use. It’s the fastest way to start: sign up, type, done. The trade-off is that your prompts and documents are processed on the provider’s servers, under the provider’s terms.
Local AI: you own it
The same kind of AI, running on a computer in your office. Nothing you type or upload ever leaves your building. There are no per-use fees and it works even when your internet doesn’t. The trade-off: you buy the hardware up front, and someone has to look after it. Open-weight models have gotten good enough that this is now a serious option for small businesses, not just enterprises.
Hybrid: you split the work
Sensitive work stays on your own hardware; everything else goes to the cloud. A law office might analyze case files with a local model while the marketing person uses ChatGPT for newsletters. Most businesses that get serious about AI land here eventually.
The six factors that actually decide it
| Factor | Points to Cloud | Points to Local |
|---|---|---|
| Data sensitivity | Public or low-stakes info | Client records, health, legal, financial data |
| Cost shape | Small monthly fees, no upfront spend | One purchase, then near-zero per-use cost |
| Internet dependence | Reliable connection everywhere you work | Field sites, workshops, spotty rural internet |
| IT capability | Nobody to maintain a server | In-house staff or an IT partner who can |
| Workload | Writing, research, general questions | Confidential document analysis, custom automation |
| Usage volume | Occasional use by a few people | Constant use by many — per-use fees add up |
No single factor decides it — it’s the combination. Which is exactly what the quiz below weighs for you.
Find your fit in two minutes
Your answers never leave your browser — nothing is sent or stored.
What each result means in practice
If you got Cloud AI
Start this week. Pick one tool (a business-tier plan, not the free consumer version — the data terms are better), give your team two or three concrete tasks to try, and revisit in a month. Your main job is writing a short internal policy: what kinds of information are never allowed to be pasted into it. Example: a five-person marketing agency drafting proposals, blog posts, and research summaries — cloud covers all of it for a few hundred dollars a month.
If you got Local AI
You’re protecting something — regulated data, client confidentiality, or a budget that can’t absorb open-ended per-use fees. A single GPU workstation runs capable open-weight models entirely offline. Example: a medical billing office that wants AI summaries of patient correspondence without any of it leaving the building. This is the path where having a partner matters most — sizing the hardware to the models you actually need is where budgets go wrong.
If you got Hybrid AI
Draw one line: which data must stay in-house? That goes to a local model. Everything else gets the convenience of cloud tools. Example: an accounting firm runs client-file analysis locally during tax season, while staff use cloud AI for emails and scheduling. Start with the cloud half this month and add the local half once the sensitive use case is proven.
Not sure? Ask us — or ask Lenny
The quiz is a starting point, not an engagement letter. If you want a human read on your situation, get in touch — or open GPT_Lenny (the chat bubble in the corner) and describe your setup.
Frequently asked questions
Is local AI cheaper than cloud AI?
It depends on usage. Cloud AI is cheaper to start (no hardware) and stays cheaper at light usage. Local AI costs more upfront but has no per-use fees, so at heavy daily usage it usually becomes the cheaper option within one to two years.
Is ChatGPT safe for confidential business data?
Consumer ChatGPT may use your conversations to improve its models unless you opt out. Business and API tiers offer stronger contractual protections, but your data still leaves your network. For regulated data (health, legal, financial), most businesses choose local or hybrid AI instead.
What hardware do you need to run AI locally?
A workstation or small server with a modern GPU. Capable small-business setups typically run $2,000–$10,000 depending on the size of the models you want. Smaller models handle document analysis and internal chat well on modest hardware.
What is hybrid AI?
A setup where sensitive work runs on your own hardware while general-purpose tasks use cloud services. Example: a local model analyzes client files, while staff use cloud AI for marketing copy. You get cloud capability without exposing protected data.
Do I need an IT team to run local AI?
Not a full team, but someone has to maintain the machine and update the software. Many small businesses have their existing IT provider manage it, or work with a partner like Lenox Systems to set it up and keep it running.