AI and the Stack of Decisions

We talk about AI as if it were a single thing. A product you buy, a technology you implement, a button you press. But behind every AI application lies a stack of choices that have already been made. Some by you. Most of them for you.
That was the point Jensen Huang made when he, in January at Davos, called AI the largest infrastructure build-out in human history and described it as a five-layer cake: energy, chips, cloud, models, and applications. The model isn't perfect, but it is good to think with. It makes something visible that we otherwise overlook: that the bottom four layers exist, and that each of them in particular holds a decision.
My latest video essay on Nowable is an Explainer, where I go through all five layers and the most important tensions we find in each of them. Along the way, I touch upon some central concepts within AI that are worth knowing, such as Energy Intelligence, Model Saturation, Harnessing, and World Models.

What becomes clear when you use the model as a framework for understanding is that AI is a large stack of commercial, geopolitical, cultural, and human decisions that each layer influences. Should the next megawatt go to a local AI model or to the heat pumps in the neighborhood? Should you build on CUDA or bet on open standards? Own your infrastructure or rent it from a hyperscaler subject to American legislation? Run a cheap open model that you control yourself, or an expensive closed one where you at least know where it comes from? Build your own workflows or weave yourself into a vendor's ecosystem that you later cannot get out of?
Answer those 5 questions and you have made five decisions. But the crucial point is that they are not independent of each other. When you install an application from one of the major providers, you have simultaneously made the decision without necessarily having noticed it. You have chosen a chip ecosystem, a cloud, a model, and an energy footprint. If you take a piece of the cake, all the layers come with it.
Most of this has so far taken place in secret. Very few of us have given the bottom four layers a thought when choosing an app. That is how it was with software for twenty years. It isn't anymore. AI has turned the bottom layers into something that carries geopolitical and economic risk, and thus into something a management team is forced to address.

In Build Resilience into Your AI Strategy, I wrote that the goal is not to control the entire stack. Virtually no one can do that, not even in Europe. The goal is resilience: knowing which decisions you have left to others, and whether you can roll them back if a layer fails. Huang's model is the tool for exactly that exercise. It works at a corporate level when a board of directors needs to map its dependencies, and it works at a national level when Europe needs to decide where we want to place our bets: on energy, on open models, on a European cloud, on efficiency. The EU has just announced a European technological sovereignty package, which shows the path we have chosen. The US and China have their own strategies.
These three regions are currently playing three different games across the 5 layers, and it is valuable to know which game you are a part of.
Awareness of all these choices can feel paralyzing. But it is healthier than the alternative, which is to let the decisions make themselves. You don't need to control every layer. You just need to know that it's there, and what you have said yes to.
Six questions I encourage everyone to ask themselves:

1. Which decisions have we actually made ourselves, and which are made for us because we chose a specific application?
2. If a layer fails, a chip shortage, a price hike, a closed API, a sanction, how long can we then continue, and what is the path forward?
3. What is our strategy regarding open versus closed models?
4. What is good enough to run locally with us, and what requires the large models in the cloud?
5. How much of the work to translate the technology into concrete workflows do we build ourselves, and how much do we place into someone else's ecosystem?
6. How important are independence and sovereignty to our business strategy, and how far down the stack do our ambitions reach?

None of the questions have a single right answer. But an organization that can answer them has done the most important thing: turned its decisions into something it has actively chosen.