The AI Stack: Need to Have vs. Nice to Have
Five layers, two of them your job. Applications and models are where you build, buy and compete. The rest is someone else's balance sheet.
← All Frameworks
Which layers of the AI stack should a business actually focus on?
The AI Stack splits AI into five layers: applications, models, infrastructure, chips and energy. Two of them are Need to Have: applications and models, where a B2B company builds, buys and competes. The other three are Nice to Have. Unless you run a utility, a chip company or a hyperscaler, you rent them, and their cost is already in your cloud bill.
The five-layer model was built to explain infrastructure economics to investors and hyperscalers writing data center checks. It was not built for operators. Most B2B leadership teams spend their AI planning time in the wrong layer of the stack.
Within the top two, the moat has moved. Model quality is converging and switching costs are falling, so treating model selection as your differentiator in 2026 is like treating your choice of cloud provider as one in 2016. The value now comes from applications that complete work inside your workflow, on your data, with limited human correction. You do not need to bake the whole cake. You need to own the top two layers before someone with more capital decides to own them for you.
How to use it
- Audit where AI planning time goes.
Look at the last quarter of AI meetings and decks. Count how much time went to chips, energy and data centers versus applications and workflows.
- Treat models as a reversible choice.
Pick the model that fits each use case today and keep switching costs low. It matters operationally but it is not a moat.
- Invest in the application layer.
Spend on what an agent needs to see, touch and decide inside your workflow before it can act without oversight: your data, your process, your customer trust.
- Score investments on ownership, not the model.
For founders and investors: ask what the application does that a general-purpose assistant cannot. Proprietary data, workflow lock-in and earned trust are the moat. A wrapper is not.
Questions
What are the five layers of the AI stack?
Applications, models, infrastructure, chips and energy, from top to bottom.
Which two layers matter most to an operator?
Applications and models. They are where a B2B company builds, buys and competes. Infrastructure, chips and energy are rented and priced into your cloud bill.
Are AI models a competitive advantage?
Rarely on their own. Model quality is converging and switching costs are falling. The advantage now sits in applications that own a workflow, run on your data and have earned customer trust.
Related: Which One Do I Adopt? · Five Paths, Five Different Disciplines · Two Clocks, Not One
The AI Stack: Need to Have vs. Nice to Have is an Uphoff Advisory Framework created by Tony Uphoff, five-time B2B CEO and founder of Uphoff Advisory, first published in Uphoff on Media on July 29, 2026. Read the full piece: Two Layers, Not Five