Also called: AI tech stack · Last checked: 7 November 2026
In short: An AI stack is the set of layers a business uses to run AI: the computing power, the data, the models and the apps people use. Each layer can come from a different supplier or be built in-house.
Table of Contents
How it works
- Infrastructure. At the bottom sit the chips, servers and cloud services, such as AWS or Microsoft Azure, that provide computing power.
- Data and models. Above that sit your data stores and the AI models, bought from suppliers or built yourself.
- Apps. At the top are the tools staff and customers use, linked to the layers below by software connections.
Where you’ll meet it
ChatGPT, Claude and Microsoft Copilot sit in the app layer. LangChain is a common tool for linking models to company data. Cloud bills, model licences and app seats all show up as separate stack costs.
What it means for your business
- Map what you have. List the tools you already pay for at each layer before buying more.
- Keep the model swappable. Put a layer between your apps and the model, so you can change suppliers when a better or cheaper one appears.
- Your data layer matters most. Models are easy to replace, while clean, well-organised data is what makes AI useful to you.
Watch out for
- Tool sprawl. Teams buying their own AI tools can create overlapping costs and security gaps.
- Hidden joins. Every link between layers needs upkeep, and a supplier change at one layer can break another.
Example
JPMorgan Chase. The bank built LLM Suite, an internal AI portal that sits between staff and the models. American Banker described it as a layer through which models are “swapped in and out”, so the bank is not tied to one supplier. By May 2025 about 200,000 employees used it, rising to about 250,000 with access by September, and CNBC reported it used models from OpenAI and Anthropic.
AI stack vs hyperscaler
| AI stack | Hyperscaler | |
|---|---|---|
| What it is | Your full set of AI layers | One supplier of cloud services |
| Who decides it | Your business | The provider sets its own offer |
| How they relate | May include a hyperscaler | Supplies the bottom layers |
Frequently asked questions
Does a small business need its own AI stack?
Not a custom one. Most small firms use off-the-shelf apps, but listing which supplier covers each layer helps you spot gaps, overlaps and data risks.
Which layer of the AI stack should we choose first?
Start with the app layer for a specific task, then check that the data it needs is ready. Infrastructure choices matter only once you build your own tools.
How often should we review our AI stack?
Once or twice a year. AI models and prices change fast, so regular reviews help you drop unused tools and switch to better value options.