What Anthropic's talks with Samsung about a custom chip mean

If you've followed AI at all, you've heard that everything runs on Nvidia chips. Anthropic, the company behind Claude, wants more options — and one of them might be a chip it helps design itself. According to a report from The Information, Anthropic has had early conversations with Samsung about building a custom AI chip, possibly using Samsung's most advanced manufacturing process. Nothing is decided yet, but it's worth understanding why a company like Anthropic would even want this.

Today, Anthropic doesn't rely on a single chipmaker. It uses Nvidia's GPUs, Google's TPUs and Amazon's Trainium chips — three different suppliers. Spreading things out like this is smart: if one supplier gets expensive, slow to deliver, or overloaded, you have alternatives. A Samsung chip would simply be a fourth choice on that list.

That last part matters, so don't misread the headline: Anthropic is not walking away from Nvidia. It's adding an option, not swapping one out.

There are two big reasons. First, cost. Running AI models for millions of people is enormously expensive, and most of that cost is the chips. If you design a chip that does exactly your kind of work, you can make each answer cheaper to produce. Second, independence — relying on one supplier for something this critical is risky, so having your own chip gives you more control.

Anthropic isn't alone here. OpenAI has been working with Broadcom on its own chip, and Amazon built Trainium for the same reason. It's becoming normal for big AI companies to design their own hardware rather than just buying it off the shelf.

In the short term, nothing changes — you won't notice a difference in Claude. This is a report about early talks, and building a chip takes years, so don't expect a product soon. Over the longer term, if labs like Anthropic get their chip costs down, that can help keep AI tools affordable and reliable as more of your apps start leaning on them. Think of this as a behind-the-scenes move to make AI cheaper to run, not a feature you'll flip on next week.

It helps to understand why this is happening across the whole industry right now, not just at Anthropic. Running AI for millions of people is astonishingly expensive, and the single biggest cost is the specialised chips that do the number-crunching — most of which come from one company, Nvidia. When your biggest expense and your biggest supplier are the same thing, you're exposed: if prices rise or supply gets tight, you have little room to move. Designing your own chip is a way to take back some control, a bit like a big supermarket launching its own-brand products so it isn't at the mercy of one manufacturer. What would this mean for you in the long run? Mostly, it's about keeping AI affordable and reliable. If companies like Anthropic can lower what it costs them to run Claude, that relief can eventually reach you — as cheaper subscriptions, more generous usage, or simply tools that don't get quietly restricted to save money. You won't notice a Samsung chip the way you'd notice a new feature, but this kind of behind-the-scenes plumbing is a big part of what decides whether the AI tools you rely on stay cheap and dependable as more of the world starts using them. For now, though, it's just early talk — so nothing changes today.