A colleague asked me to share with him how the AI stack works and who the leaders are. I ended up doing about 30 hours of research for him and I may as well share it with everyone. I’m going to try to keep it updated quarterly, if not monthly. I basically looked at who consumed the biggest portion of the news cycle and who had super high public valuations. It doesn’t mean the software is great. It just means people use it or they are doing something outside the box.
I broke AI down into eight layers. I’ll admit I had AI help me organize my work to make it clearer than the ramblings of notes I jotted down.
Here are what each of those layers are.
Layer 1 (Silicon and Hardware): This is the physical foundation. The specialized chips that make AI computation possible. Training and running AI models requires massive parallel processing power that traditional processors can't deliver, so purpose-built AI accelerators are essential.
Layer 2 (Cloud and Compute Infrastructure): This layer provides the data centers and cloud platforms that give businesses on-demand access to AI computing power. Most organizations can't afford to build their own AI infrastructure, so cloud providers rent out the GPU capacity needed to train and run models.
Layer 3 (Data Infrastructure): AI models are only as good as the data they can access, and this layer handles storing, organizing, and retrieving that data. Vector databases in particular have become critical because they let AI models quickly search through an organization's own information to generate more accurate, relevant answers.
Layer 4 (Foundation Models): These are the large-scale AI models that understand and generate language, code, and images. They serve as the base intelligence that everything else in the stack builds on top of. Applications, agents, and tools all depend on these models to do the actual reasoning.
Layer 5 (Model Development and MLOps): This layer provides the tools developers use to find, fine-tune, evaluate, and deploy AI models. It's essentially the workshop where raw foundation models get customized and made production-ready for specific business needs.
Layer 6 (Orchestration and Agent Frameworks): This layer provides the software that lets AI agents plan multi-step tasks, use external tools, and take actions autonomously. It's what turns a basic AI model from something that just answers questions into something that can actually do work across multiple systems.
Layer 7 (AI Safety and Governance): Every AI deployment creates new security risks (prompt injection, data poisoning, model theft), and this layer builds the defenses. It's the fastest-growing segment because enterprises are realizing AI security needs to be built in from day one, not bolted on later.
Layer 8 (Applications and AI Agents): This is the layer most people actually interact with. The AI-powered tools embedded in the software they already use. It's where AI delivers direct business value, and where the best profit margins live, making it the most strategically important layer for most enterprises to own.
This Is a Living Document
The AI landscape changes fast. Companies that weren't on anyone's radar six months ago can become category leaders overnight. DeepSeek is a good example. So is Cursor.
I built this chart to help a colleague, but hopefully other executives and decision-makers can use it to see the full picture of how AI fits together. But I know I didn't get everything right. If you think a company is missing, miscategorized, or doesn't belong, I want to hear from you.
Reach out to me at chris@labattsimon.com, visit labattsimon.com, or connect with me on LinkedIn. I'll keep updating this as things evolve.