Nvidia Starts It: AI’s Operating System Moment

— Personal reflection: What happens when AI becomes a feature of the operating system instead of a destination application?

Everyone thinks Anthropic´s biggest threat is OpenAI. Or that OpenAI´s biggest threat is Anthropic.

I think both statements are wrong.

The biggest threat to frontier AI labs might be the same force that has reshaped every major technology cycle over the last 40 years:

The operating system.

For the past two years, we’ve been asking the wrong question.

Who will win?OpenAI? Anthropic? Google? Meta?

Every week, the discussion sounds the same. A new benchmark. A new model. A new leaderboard.

The assumption behind all these debates is that the most important layer in AI is the model itself.

I’m no longer sure that’s true.

Nvidia started it

This week, Nvidia unveiled RTX Spark, a new platform designed to transform Windows into what Jensen Huang describes as:

“An agentic AI operating system.”

Most people focused on the hardware. I think they’re missing the bigger story. I saw the first credible glimpse of a world where AI stops being an application.

To understand why that matters, it’s worth looking backward.

The history of technology is filled with examples of layers becoming commodities. The companies building those layers are often incredibly innovative. Sometimes they become enormously successful. But once a capability becomes native to the platform, the economics change forever.

Browsers, databases, and web servers, for instance, once considered strategic products, eventually became features of larger ecosystems.

The same pattern appears over and over again. Innovation happens at the edge. Value gets absorbed by the platform.

And here is the important part:

And eventually, users stop thinking about the underlying technology altogether.

The model is the product

Today, OpenAI, Anthropic, Google, and others sit at the center of the AI experience.

• You open an app.
• You open a browser.
• You talk to ChatGPT.
• You talk to Claude.

People identify themselves as ChatGPT users. Claude users. Gemini users.

Get Alejo Lovallo’s stories in your inbox

Remember me for faster sign in

Today, everybody sees the model. The model is the product.

However, let´s ask a few more questions: Does someone wake up thinking of:

1. Networking stack behind a Zoom call?
2. Database that powers their favorite application?

But that’s only true because operating systems still don’t understand AI. It´s just a temporary artifact of where the technology currently sits.

Because if intelligence becomes native to the operating system, users may stop interacting with models directly altogether (At least for a lot of use cases).

How (I believe) the relationship between users and Ai companies will change?

Imagine a future where your laptop, phone, car and every device you own comes with a built-in AI agent that can see your screen, access your files, understand your preferences and execute actions across applications.

At that point, ChatGPT and Claude stop being destinations. They become infrastructure.

And infrastructure historically captures far less value than the platform built on top of it.

This isn’t a new story.

• Microsoft didn’t win because Word was the best application. It won because Windows became the distribution layer.
• Google didn’t become dominant because of a single product. It owned Android.

Distribution beats applications more often than founders like to admit.

Intelligence won’t live behind a website. It will live behind the interface itself. And once that happens, something interesting occurs.

The relationship between users and AI companies changes.

Final words

NVIDIA’s announcement matters because it hints at a future where AI is no longer cloud-first.

Not because of what was launched. But because of the direction it points toward.

• The hardware gets powerful enough.
• The models get smaller.
• Inference gets cheaper.

And eventually, AI becomes a native capability, not a destination website.

Ironically, this would be one of the best outcomes for the ecosystem.

A truly commoditized intelligence layer would lower barriers to entry, reduce dependence on a handful of providers and allowfounders to focus on building products instead of wrapping APIs.

In that world, the winners may not be the companies with the biggest models.

The winners may be the companies that own the interface between humans and intelligence.

NVIDIA may have started that transition this week.

The rest of the industry just hasn’t realized it yet.