The AI Startup Landscape: Where Value Is Being Created

Every pitch deck now mentions AI. But not all AI startups are creating equal value. Understanding where durable value is being created versus where it’s thin wrappers helps navigate this landscape—whether you’re building, joining, or investing.

Here’s how to understand the AI startup landscape.

The AI Startup Stack

Layer Analysis

ai_stack_layers:
  infrastructure:
    examples: GPU clouds, training infrastructure
    players: CoreWeave, Lambda Labs, Together
    moat: Capital, relationships, scale
    value: High if executed well

  foundation_models:
    examples: OpenAI, Anthropic, Cohere
    moat: Research talent, compute, data
    value: Extremely high, few winners

  model_tooling:
    examples: Weights & Biases, Hugging Face, Replicate
    moat: Developer adoption, ecosystem
    value: High if category leader

  application_infrastructure:
    examples: LangChain, Pinecone, Weaviate
    moat: Integration, developer experience
    value: Emerging, consolidation coming

  applications:
    examples: Jasper, Copy.ai, Character.ai
    moat: Distribution, workflow, data
    value: Varies widely

The Thin Wrapper Problem

thin_wrapper_risk:
  definition: Application that's primarily API calls to foundation model

  characteristics:
    - Little unique technology
    - Easy to replicate
    - Low switching cost for users
    - Dependent on API pricing and availability

  examples:
    - Simple chatbot wrapping GPT
    - Content generator with basic prompts
    - "GPT for X" without unique data/workflow

  risk_factors:
    - OpenAI adds feature directly
    - Competitors can build same in weeks
    - API price changes destroy margin
    - No compounding advantage

Durable Value Creation

Where Moats Exist

durable_ai_moats:
  proprietary_data:
    description: Data that improves AI uniquely
    examples:
      - User interactions that train better models
      - Proprietary datasets in specific domains
      - Feedback loops that compound

  workflow_integration:
    description: Deep embedding in user workflows
    examples:
      - AI in existing enterprise tools
      - End-to-end process automation
      - Systems of record integration

  domain_expertise:
    description: AI specialized for specific industries
    examples:
      - Legal document analysis
      - Medical imaging interpretation
      - Financial compliance automation

  distribution:
    description: Access to customers at scale
    examples:
      - Existing platform adding AI
      - Partnership with industry leader
      - Community-driven adoption

Questions to Ask

evaluation_questions:
  technology:
    - What's unique about the technology?
    - Can this be replicated with API calls?
    - Is there a data flywheel?

  market:
    - Is this a vitamin or painkiller?
    - How big is the actual problem?
    - Will customers pay meaningfully?

  competition:
    - What if OpenAI builds this?
    - What if incumbent adds this feature?
    - What's the defensibility?

  team:
    - Do they have relevant domain expertise?
    - Do they understand AI limitations?
    - Can they execute beyond technology?

Market Segments

Where Opportunity Exists

ai_market_segments:
  enterprise_productivity:
    opportunity: Large, proven willingness to pay
    competition: Intense, Microsoft and Google
    path_to_value: Workflow integration, security

  vertical_saas:
    opportunity: Industry-specific AI applications
    competition: Less from giants
    path_to_value: Domain expertise, data

  developer_tools:
    opportunity: Accelerate AI development
    competition: Growing, consolidating
    path_to_value: Developer adoption, ecosystem

  consumer:
    opportunity: Large if product-market fit
    competition: Winner-take-most
    path_to_value: Engagement, monetization

  infrastructure:
    opportunity: Picks and shovels
    competition: Requires scale
    path_to_value: Reliability, cost

What I’d Build

High-Conviction Areas

building_opportunities:
  ai_for_regulated_industries:
    why: Compliance requirements create moats
    examples: Healthcare, finance, legal
    requirement: Domain expertise essential

  ai_infrastructure:
    why: Growing market, technical moat
    examples: Evaluation, monitoring, cost optimization
    requirement: Engineering excellence

  vertical_workflow_tools:
    why: Deep integration creates switching cost
    examples: AI for specific professional workflows
    requirement: User understanding + AI capability

  data_advantage_applications:
    why: Proprietary data creates compounding moat
    examples: Applications where user data improves product
    requirement: Product that generates valuable data

Key Takeaways

Not all AI startups are equal—moats matter
Thin wrappers risk being commoditized or copied
Durable value: proprietary data, workflow integration, domain expertise
Ask: “What if OpenAI builds this?”
Vertical applications often have stronger moats than horizontal
Infrastructure layer creates value if you can execute
Distribution matters as much as technology
Be skeptical of “GPT for X” without unique insight
Domain expertise + AI capability is powerful combination

The AI opportunity is real. The question is where durable value will be created.