Module 2: The AI Industry Itself — Who's Actually Making Money
Industries of the Future
That's the first lesson of the AI industry: the infrastructure layer is where the money concentrates first.
The AI Stack — Where Value Lives
Think of the AI industry as a stack with distinct layers, each with different economics:
Layer 1: Compute & chips. NVIDIA controls roughly 80% of the AI training chip market. Their H100 GPUs sell for $25,000-$40,000 each, and data centres can't buy them fast enough. AMD is fighting for second place. Custom chips from Google (TPUs), Amazon (Trainium), and startups like Cerebras and Groq are trying to break the monopoly. This layer prints money.
Layer 2: Cloud & infrastructure. Amazon Web Services, Microsoft Azure, and Google Cloud are the landlords of AI. They rent compute to everyone else. Microsoft alone committed over $80 billion to AI infrastructure in 2025. This layer is capital-intensive but enormously profitable at scale.
Layer 3: Foundation models. OpenAI, Anthropic, Google DeepMind, Meta AI, Mistral. These companies build the base models everyone else fine-tunes or builds on. OpenAI reportedly hit $3.4 billion in annualised revenue by late 2024. But this layer has a problem: training costs are astronomical and it's unclear how many foundation model companies the market can sustain.
Layer 4: Application layer. This is the largest layer by number of companies, but the most competitive. Thousands of startups building tools for writing, coding, customer service, design, legal, healthcare — you name it. Most will fail. The ones that win will own specific verticals.
Layer 5: Services & integration. Consulting firms (Accenture, Deloitte, McKinsey) and system integrators helping enterprises actually deploy AI. Accenture committed $3 billion to its AI practice. This is where a lot of jobs are, even if it's less glamorous than building models.
Order the AI stack layers from bottom (infrastructure) to top (user-facing):
The Economics Nobody Talks About
Here's the uncomfortable truth about the AI industry: most AI startups aren't profitable. The cost of compute is brutal. Training a frontier model can cost over $100 million. Even running inference (actually serving users) is expensive.
Jasper AI raised $125 million at a $1.5 billion valuation in 2023, then laid off staff as ChatGPT ate their market. Many "AI wrapper" companies — those building thin layers on top of OpenAI's API — are discovering that their moat is paper-thin.
Who actually makes money right now:
- Chip companies (NVIDIA)
- Cloud providers (AWS, Azure, GCP)
- Foundation model companies with scale (OpenAI, arguably)
- Vertical AI companies with deep domain integration (Veeva for pharma, Palantir for government)
- Services companies helping enterprises adopt AI
Who's struggling:
- Horizontal AI tools with no differentiation
- Companies whose entire product is "ChatGPT but for X"
- Startups competing on features that foundation models keep absorbing
Where the Jobs Are
The AI industry itself employs across a surprising range of roles:
Technical: ML engineers, data scientists, infrastructure engineers, AI safety researchers. Compensation is extraordinary — senior ML engineers at top labs earn $400K-$1M+ total comp.
Non-technical but essential: Product managers, technical writers, AI trainers (RLHF), policy specialists, partnerships/BD, sales engineers. These roles are growing faster than pure technical roles because every AI company needs people who understand the technology and can interface with customers.
The hidden opportunity: AI companies desperately need people with domain expertise. A nurse who understands AI can be more valuable to a healthtech startup than another ML engineer. A teacher who gets AI can shape education products in ways pure technologists can't.
Why is a nurse who understands AI valuable to a healthtech startup?
My Take: Where to Bet
The infrastructure layer is already dominated. Unless you're an elite chip designer or cloud architect, that ship has sailed for most people.
The foundation model layer is a winner-take-most game. Exciting but extremely high-risk as a career bet.
The real opportunity for most people is the application and services layers. The world needs millions of people who can apply AI to specific domains. That's where skills, domain knowledge, and human judgment create lasting value.
---
Pick one layer of the AI stack that interests you most. Spend 30 minutes researching:
1. Identify the top 5 companies in that layer
2. Look at their careers pages — what roles are open?
3. Read their most recent blog posts or press releases — what are they building?
4. Find one person on LinkedIn who works at one of these companies in a non-engineering role. Look at their career path. How did they get there?
Write a one-paragraph summary: "The [layer] of the AI stack is interesting to me because... The skills that matter here are... My realistic entry point would be..."
---
- 1The AI industry has five distinct layers with very different economics and opportunities
- 2Infrastructure (chips, cloud) is where money concentrates; applications and services are where most jobs are
- 3Many AI startups aren't profitable — the "wrapper" business model is fragile
- 4You don't need an ML background — domain expertise plus AI fluency opens doors across the industry
- 5The application and services layers offer the broadest career opportunities for most people