From Building Models to Stealing Markets

Happy Monday!

Last week, two seemingly unrelated news stories revealed the next phase of AI competition. Apple entered talks to acquire Perplexity, and OpenAI launched direct competition to Google Workspace and Microsoft Office with new collaboration features.

These aren't isolated moves, but instead are early battles in what will define the next decade of technology competition. The AI infrastructure era is ending, and the integration wars have begun.

After analyzing these developments alongside OpenAI's expansion into biology research and IBM's partnership with Lockheed Martin to eliminate data silos, it’s clear that the infrastructure → integration path will define the next wave of AI progress.

From Picks and Shovels to Vertical Dominance

For the past three years, AI companies have operated like infrastructure providers by selling compute, APIs, and raw model capabilities to anyone who would pay. OpenAI provided the GPT API, Anthropic offered Claude, and hundreds of startups built applications on top.

This commoditization was inevitable. As model capabilities converge and open-source alternatives close the gap, the value has shifted from having the best model to having the best integration with existing workflows.

Apple's potential Perplexity acquisition isn't just about buying search technology. It's about controlling the discovery layer of information consumption. OpenAI's productivity suite isn't about competing with Microsoft on features alone, but about owning the collaborative workflow where knowledge work happens.

The infrastructure play is over. The integration wars have begun.

The Meta Trend: From Horizontal Tools to Vertical Dominance

This wave occurring across AI companies follows a predictable pattern: start with horizontal infrastructure, then move to vertical integration once commoditization pressures mount.

This mirrors the evolution of cloud computing. Amazon started with basic compute and storage (EC2, S3), then moved up the stack to databases, analytics, and industry-specific solutions. The companies that stayed in infrastructure became utilities. The companies that climbed the value chain became platforms.

AI is following the same trajectory, but faster. The difference is that AI integration creates data flywheels that infrastructure provision never could.

Pattern Recognition: Three Integration Strategies

Pattern #1: The Application Layer Takeover

OpenAI's move into productivity applications represents a direct assault on Microsoft and Google's enterprise dominance. It goes beyond building better spreadsheets or document editors; this strategic move is about capturing the context where knowledge work happens.

Every collaboration session, every document edit, every meeting summary becomes training data for OpenAI's models. Traditional productivity companies offer software; AI-native companies offer software that gets smarter with every interaction.

Pattern #2: The Discovery Interface Capture

Apple's interest in Perplexity signals a recognition that search and discovery are becoming AI-mediated experiences. But the real value isn't in replacing Google, it's in controlling the layer where users express intent and receive answers.

Perplexity's conversational search interface captures not just what users want to know, but how they think about problems. For Apple, this integration could transform Siri from a voice command interface into an intelligent research partner embedded across all Apple devices.

Pattern #3: Industry-Specific Embedding

OpenAI's expansion into biology research and IBM's partnership with Lockheed Martin represent the most valuable integration strategy: becoming essential to industry-specific workflows.

When AI companies embed themselves in scientific research pipelines or defense procurement processes, they go beyond just providing tools. These companies become infrastructure for entire industries. The switching costs become enormous, and the data advantages compound over time.

Contrarian Take: The Data Moat Revolution

The real competition isn't about building better models, it's about securing better data sources.

Traditional software companies like Microsoft and Google have user data, but it's largely behavioral (clicks, usage patterns, demographics). AI-integrated workflows capture something far more valuable: human reasoning patterns, decision-making processes, and domain expertise.

When OpenAI processes your collaborative documents, it learns how teams think through problems. When Perplexity handles your research queries, it understands how experts approach new topics. When AI systems manage scientific workflows, they absorb the methodologies that drive discovery.

This creates a new type of competitive moat. Traditional network effects rely on user connections. AI integration effects rely on reasoning pattern accumulation. The more complex workflows an AI company processes, the better it becomes at handling similar workflows for other users.

Traditional tech companies optimized for user engagement. AI companies are optimizing for cognitive integration.

The Bigger Picture: AI as the New Operating System

The integration trend points toward AI becoming the orchestration layer for all digital workflows. Instead of users switching between applications, AI agents will coordinate across tools, platforms, and data sources on users' behalf.

This creates a winner-take-most dynamic similar to mobile operating systems. The companies that establish integration dominance will become the platforms that other software builds upon. Those that remain in infrastructure will become invisible utilities.

The question isn't which company will build the best AI model. It's which company will become the integration layer that other businesses can't live without.

In motion,
Justin Wright

1. Introducing 11ai: the voice-first AI assistant that takes action (ElevenLabs)

2. Prompt engineering bible (Anthropic)

3. Anthropic wins a major fair use victory for AI (The Verge)

4. Introducing Gemma 3n: The developer guide (Google)