The Quiet Shift From Products to Platforms in AI

Something curious is happening in the AI industry, visible only if you squint at the pattern rather than the individual announcements. Over the past week, we've seen OpenAI unveil GPT-6 Astra not as a model but as infrastructure for business, launch an Agents API as a managed cloud service, introduce ChatGPT for Financial Services with built-in data integration, and release a Data agent that connects to company systems. Google responded with WeatherNext 3 deployed across Search, Maps, and Gemini simultaneously.

These aren't product launches. They're platform moves.

The distinction matters more than it seems. A product is something you buy, deploy, and control. A platform is something you plug into, where the vendor controls the infrastructure, data flows, and—critically—the upgrade path. When OpenAI talks about serving "22 million requests per second" for ChatGPT's storage infrastructure, they're not describing a tool. They're describing utility-scale computing that rivals AWS or Azure in ambition if not yet in scope.

This platformization of AI follows a familiar playbook from the cloud computing wars, but it's moving faster and with higher stakes. Consider the Agents API announcement: rather than offering tools for developers to build their own agent systems, OpenAI is providing "orchestration capabilities" and "persistent sessions" as a service. You don't own the intelligence—you rent access to it, through their infrastructure, following their terms.

The financial services offering makes the strategy even clearer. By bundling "built-in financial data" with GPT-6 Astra, OpenAI isn't just providing a language model. They're creating a walled garden where the data, the intelligence, and the delivery mechanism all flow through a single vendor. It's the same approach behind Google's integration of WeatherNext 3 across its entire ecosystem: why offer a weather AI you could deploy anywhere when you can bake it into every Google product?

For users, this platform shift brings obvious benefits. The Data agent that lets you "connect company data" and "build interactive dashboards" using natural language genuinely democratizes capabilities that once required data science teams. The natural language voice experiences in GPT-Live-1 work because OpenAI handles the complex infrastructure.

But it also creates new dependencies. When MIT researchers use GPT-5.6 Sol to run quantum computing experiments, or Playco relies on GPT-6 Astra to prototype games, they're building on foundations they don't control. The vendor can change pricing, restrict access, modify capabilities, or simply shut down features—and customers have limited recourse.

The most telling detail might be that storage infrastructure serving "1 billion ChatGPT users." That's not a product ecosystem—it's a population. And populations don't migrate easily.

We're watching the AI industry repeat the shift that turned software from something you installed to something you subscribed to, except faster and with more fundamental capabilities at stake. The question isn't whether platforms will dominate AI delivery—that ship has sailed. The question is whether we'll build regulatory frameworks, interoperability standards, and competitive alternatives before a handful of platform owners control access to intelligence itself.

Right now, we're too busy celebrating each new model announcement to notice we're giving away the keys to the infrastructure underneath.