Applications
The products end users actually pay for — chat, code, search, creative, vertical workflows.
What this layer does
The application layer is where AI revenue is born. Every dollar flowing through the rest of the stack — into GPUs, fabs, transformers, copper — ultimately comes from a person or business paying for an AI product here. Apps split into two business models: consumer subscriptions (ChatGPT Plus, Claude Pro) and B2B (per-seat SaaS or usage-based API consumption layered into an existing workflow).
This layer is overwhelmingly private. The interesting public proxies are the incumbents embedding AI into existing distribution — Microsoft into Office, Google into Workspace, Adobe into Creative Cloud, Salesforce into the CRM — and the chance that a wave of AI-native apps replaces them.
Sub-categories
General-purpose chat & search. The bundles people sign up for personally.
The category with the clearest paid traction. Code is structured, errors are verifiable, and devs already pay for tools.
Document-heavy, high-billable-hour workflows where copilots compress associate work.
Ambient scribes, clinical reasoning, prior-auth, evidence retrieval. Real cost-saving ROI for hospitals.
Voice agents and chat agents handling tier-1 customer interactions. Sold on labor displacement.
RAG over corporate data — the unsexy enterprise gold mine.
Generative image, video, audio, music. Consumer-facing with viral growth loops.
Multi-step task execution, browser/computer use. Mostly still pre-revenue.
The default way most enterprises will buy AI — bolted into software they already pay for. The public-market expression of the application layer.