Remember when everyone thought Nvidia's chip dominance was basically untouchable — like, Coca-Cola-owns-Christmas untouchable? Well, apparently two of AI's biggest spenders didn't get that memo. Reports have confirmed Anthropic is in early-stage talks with Samsung Electronics to build a custom AI chip using Samsung's advanced 2-nanometer process (translation: really, really small, really, really powerful). This comes just weeks after OpenAI unveiled its own custom chip project with Broadcom. So within the space of a month, two of the industry's loudest names have quietly started shopping for their own hardware. Coincidence? Sure. And Batman's just a guy who really likes bats.

Here's the thing about Nvidia right now: it doesn't just sell chips, it controls the table everyone else has to play at. The company holds roughly 74% of the AI chip market, which is less "market leader" and more "the only casino in town, and you don't get to choose the odds." Every major AI lab — Anthropic included — has spent the last few years paying Nvidia's prices, waiting on Nvidia's supply queues, and building roadmaps around Nvidia's release schedule. That's a lot of control to hand one supplier when your entire business runs on compute.

So the Samsung talks make a lot of sense, even in their early, nothing's-signed-yet stage. A custom chip built specifically for Anthropic's workloads, on Samsung's cutting-edge 2-nanometer process, would give the company a hardware option that isn't dependent on Nvidia's pricing or its production calendar. It's not about ditching Nvidia entirely (nobody's burning that bridge just yet) — it's about not being stuck as a customer with zero leverage. And Anthropic isn't alone in reaching that conclusion. OpenAI's custom chip project with Broadcom, unveiled just weeks prior, tells you this isn't one company's paranoia. It's an industry-wide realisation that if compute is the oil of the AI age, maybe you don't want to be permanently renting the rig.

Now, if you're running a business that has nothing to do with silicon wafers or nanometers, you might be tempted to scroll past this one. Don't. Compute costs sit underneath almost everything in the AI economy right now — the tools you use, the software you pay for, the AI features quietly baked into your CRM. When the companies building the foundation models start diversifying their chip supply, it's a signal that the cost and availability of AI compute could shift meaningfully over the next few years. More competition among chipmakers, in theory, means better pricing and fewer supply bottlenecks trickling down to everyone building on top of these models — including you. It's also a tell on where the smart money thinks this industry is heading: not a one-supplier world, but a multi-supplier one, the way cloud computing eventually split between a handful of major players instead of staying a single company's fiefdom. If you're an SME leaning on AI tools to punch above your weight, keep an eye on this space. Cheaper, more available compute is good news for your margins, even if you'll never see a chip up close.

So no, Anthropic and Samsung haven't signed anything yet, and OpenAI's chip isn't shipping tomorrow. But the message is loud and clear: the AI industry's biggest names are done being just customers. They want a seat at the table they're currently just paying rent to sit at. Nvidia's still holding most of the chips (pun very much intended) — but for the first time in a while, other players are quietly building their own deck.

— The Business Index Team