The past week exposed a fundamental strategic split in how tech giants are weaponising artificial intelligence. OpenAI and Google both unveiled multimodal models—capable of processing video, audio, images, and text simultaneously—within 24 hours of each other. But their approach to dominance could not be more different.
OpenAI operates as a pure-play AI company. It builds models and ships APIs; others build products on top. This gives it narrative control and keeps it at the front of the innovation curve. By announcing GPT-4O one day before Google's I/O conference, OpenAI ensured that Google's legitimate advances—the Gemini 1.5 Pro, its own multimodal model—read as reactive rather than pioneering. Marketing matters.
Google, by contrast, is an advertising company with 2 billion Gmail users, dominant search, Maps, and Android. It cannot afford to lose customers to ChatGPT or any other AI native. So it is embedding AI into existing products, not shipping standalone models hoping developers adopt them. The AI Overviews in Search, the visual search across Google Photos, the intelligent trip planner that reads your Gmail calendar and restaurant preferences—these are not sexy headlines. But they are defensible moats. A user who gets a better search experience, a simpler way to find that saved photo, or an itinerary that actually knows their constraints has less reason to switch platforms.
Microsoft occupies a third position: workplace dominance. Its Copilot Plus PCs bring neural processing units to Windows, allowing models to run locally without cloud latency or privacy leakage. The Recall feature—automatically screenshotting your work and letting you query it semantically—is workplace automation that plays directly to Microsoft's strength in Enterprise.
What unites Google and Microsoft is ruthless focus. Neither is pretending to be an AI company. Both are using AI to entrench positions in markets where they already win. OpenAI's bet is that best-in-class models eventually become indispensable enough that everyone builds around them. History suggests both strategies can coexist, but the distribution advantage of billions of installed users is not something that can be engineered in a lab.