In the late 90s, we memorized the OSI model with:
"All People Seem To Need Data Processing"
Application · Presentation · Session · Transport · Network · Data Link · Physical
Today, we are living through
"All People Seem To Need
AI Processing."
The AI industry is building a layered technology stack that mirrors the OSI model with striking precision. Tracking where the trillions of dollars are flowing across each layer reveals what is happening today — and who will capture the ultimate value tomorrow.
"The parallel between the OSI model and the AI technology stack lies in their shared, layered approach to managing complex systems — moving from raw physical hardware at the bottom to end-user applications at the top. While OSI structures network communication, the AI stack structures the creation, training, and deployment of intelligent models.
The AI Stack as the New OSI Model
Seven layers. Trillions of dollars. One question: which layer captures the ultimate value? Click any layer to explore the investment thesis.
Delivering measurable ROI to users and businesses
Translating intelligence into autonomous actions
Managing intelligence and reasoning sessions
Reliable delivery and orchestration of compute power
Processing and routing computation (GPUs, TPUs, ASICs)
Physical compute facilities and real estate
Power generation, cooling, and grid infrastructure
Investment by Layer
Annual capital flow (indicative scale)
Note: Chips shown as market cap ($5T), others as annual capex/VC. Scale is illustrative.
The Players & Their Stack Positions
Every major player is racing to own more of the stack. Select a company to see their investment footprint across all seven layers.
Microsoft
Stack investment intensity (0–100 scale)
The Profitability Paradox
The layers receiving the most capital investment are not necessarily the ones generating the highest margins. The companies that laid the fiber optic cables of the 2000s eventually went bankrupt or saw their services commoditized into "dumb pipes." The durable, high-margin value was ultimately captured by the application layer.
Gross Margin by Layer
Approximate operating margins (negative = burning cash)
70%+ gross margins on chips. The Cisco of the 90s — but unlike Cisco, NVIDIA is aggressively moving up the stack into software and services to avoid the commodity trap.
OpenAI burned $8–9B in 2025 to generate $20B ARR. Anthropic is similarly loss-making. The pure model layer is commoditizing fast. DeepSeek proved this.
Amazon, Google, and Microsoft are seeing margins compressed by $700B+ in annual capex. The infrastructure is necessary but increasingly commodity.
Mature SaaS applications earn 70–80% gross margins. The AI application layer is where the internet's value ultimately concentrated. History is about to repeat.
The Winner Will Come
From Nowhere.
The AI industry is currently trapped in the lower layers of the stack. We are constrained by power grids, data center real estate, and chip yields. Consequently, that is where the trillions are flowing today.
But infrastructure is a means to an end.
The ultimate winner of the AI era has likely not even been founded yet — or is currently a small startup flying under the radar. Just as the massive infrastructure investments of the 90s paved the way for Google, Amazon, and Meta to dominate the next two decades, today's $700 billion infrastructure build-outs are paving the way for a new apex predator.
This new giant will come from nowhere, capture the imagination and market share at Layer 7 (Applications), and become the most dominant force to reckon with from 2026 to 2036. They will not win by building better chips or larger data centers; they will win by taking the commoditized intelligence provided by Layers 1 through 6 and turning it into an indispensable, everyday utility that changes how humanity lives and works.
The picks and shovels are being sold. The gold rush is about to begin.
References
Analysis current as of March 2026. Investment figures are indicative.