AI Market Segmentation

Three AI market segments with different dynamics: foundation models, tooling, and applied AI

Bret Taylor
Inside the expert network training every frontier AI model

AI Market Segmentation

"I think there's three segments of the AI market that will end up fairly meaningful markets." - Bret Taylor

What It Is

A strategic framework for understanding where to compete in the AI market, identifying three distinct segments with fundamentally different dynamics, barriers to entry, and success factors. Each segment has different implications for startups, different relationships with the others, and different long-term value capture potential.

Bret Taylor developed this perspective from his unique vantage point as co-founder/CEO of Sierra (an applied AI company), chairman of OpenAI (a foundation model company), and former co-CEO of Salesforce (an enterprise software company).

The Three Segments

1. Foundation Models (Infrastructure Layer)

What it is: Creating frontier/foundation models (GPT, Claude, Gemini, etc.)

Market dynamics:

  • Capital-intensive: Entirely a function of CapEx
  • Consolidating to hyperscalers and well-funded labs
  • Models depreciate quickly as assets
  • Requires massive scale to generate returns

Startup viability: Low

  • Most standalone AI labs have been acquired (Inflection, Adept, Character)
  • No viable startup business model due to CapEx requirements
  • "No entrepreneur should probably build a frontier model"

Exception: Only viable with ability to raise billions (like Elon Musk with xAI)

2. Tooling (Picks and Shovels Layer)

What it is: Services that help others build AI—data labeling, eval tools, data platforms, specialized models (like ElevenLabs for voice)

Market dynamics:

  • Classic "picks and shovels" opportunity
  • Analogous to Confluent, Databricks, Snowflake in cloud era
  • Risk: Very close to the infrastructure layer

Startup viability: Medium

  • Real meaningful companies are possible
  • Key risk: Foundation model companies can easily add competing features
  • "If or when... one of these large infrastructure providers introduces a competitor, why will people continue to choose you?"

Warning: At risk from foundation model developer days releasing competing capabilities

3. Applied AI / Agents (Application Layer)

What it is: Products that use AI to achieve business outcomes—customer service agents (Sierra), legal AI (Harvey), content marketing, supply chain analysis

Market dynamics:

  • "Agent is the new app"—the dominant product form factor
  • Higher margin than foundation models
  • Less technical, more product-focused over time
  • Resembles SaaS market structure

Startup viability: High

  • Most exciting opportunity for entrepreneurs
  • Value comes from understanding business problems deeply
  • Will "pay taxes" to model providers, but capture significant value
  • Long tail opportunity: Not just big markets, but many specialized verticals

Future trajectory: "What database do you use?" became irrelevant for SaaS—similarly, "how you deal with the models" will become irrelevant for agents. Focus shifts to workflows and business outcomes.

How to Apply It

For founders:

  1. Avoid foundation models unless you have access to billions in capital
  2. Approach tooling carefully - Build sustainable differentiation beyond what model providers can easily replicate
  3. Focus on applied AI - The best opportunity is building agents for specific business problems

For applied AI success:

  • Deep domain expertise matters more than AI expertise over time
  • Business model can be outcome-based (resolution rate, sales commissions)
  • Technology commoditizes; understanding the customer problem compounds
  • Look for long-tail opportunities, not just massive markets

Questions to ask:

  • What business outcome am I delivering?
  • Can a foundation model company easily replicate this?
  • Do I understand the domain deeply enough to create lasting value?

The Agent Opportunity

The applied AI / agent segment is where Bret sees the most opportunity because:

  1. Productivity gains are real and measurable - Agents do jobs autonomously, not just make workers slightly more efficient
  2. Outcomes-based pricing aligns incentives - Pay for results, not tokens or seats
  3. Domain expertise creates moats - Understanding healthcare, legal, or CX deeply is harder to replicate than technical AI capabilities
  4. Long tail of opportunities - Just like SaaS has thousands of companies beyond the top 5, agents will serve many verticals

Source

  • Guest: Bret Taylor
  • Episode: "Inside the expert network training every frontier AI model"
  • Key Discussion: (00:52:36) - Overview of the three AI market segments
  • YouTube: Watch on YouTube

Related Frameworks

Back to Product Strategy