Anthropic quietly joins the race to build its own chips
4 mins read

Anthropic quietly joins the race to build its own chips

Anthropic has opened early talks with Samsung Electronics to manufacture a custom AI chip, according to a report from Bloomberg.

Read more Latest Hendrick Automotive expansion in Charlotte covers cars ‘people dream of’

The Claude developer has never built its own silicon before. It has relied entirely on chips rented from Amazon, Google, and Nvidia, and that dependence is now colliding with the soaring cost of running its largest models.

The conversations are still preliminary. Anthropic has not decided what the chip will do, how it will fit into a server, or how powerful it needs to be, according to TechCrunch.

Samsung declined to comment on the discussions when TechCrunch reached out.

Anthropic currently depends on Amazon’s Trainium chips, Google’s Tensor Processing Units, and Nvidia’s graphics processors to train and run its models.

A diversified hardware stack built on those three suppliers will remain central to its compute strategy, the company told TechCrunch. Nothing about the Samsung talks changes that today.

Nvidia controls about 74% of the global AI chip market, according to The Information. That level of concentration gives one company outsized influence over pricing across the industry.

Custom silicon, designed around a lab’s own model architecture, offers one of the only ways around that math.

Anthropic is not moving first. OpenAI unveiled its own custom chip last month, an inference processor called Jalapeno built with Broadcom.

Anthropic’s Samsung talks surfacing weeks later suggest the industry is quietly hedging against Nvidia dependence.

For investors, the read-through lands on the supply side, not on Anthropic itself. Anthropic remains privately held, so there is no direct way to buy into its chip strategy.

The companies that stand to gain or lose are the ones building the hardware underneath it.

Yuichiro Chino / Getty Images

Samsung is not a random choice for Anthropic. The company was one of three memory chipmakers, alongside SK Hynix and Micron, that invested in Anthropic’s $65 billion funding round in May, according to Forbes.

Samsung is the only one of those three investors that also operates its own chip foundries.

Related: Anthropic restores access to Mythos 5 for select organizations

Anthropic is specifically evaluating Samsung’s two-nanometer manufacturing process and its advanced chip packaging facilities, according to The Information.

Read more Charlotte Water customers urged to conserve after line break floods a busy road

Winning a marquee AI client would give Samsung a showcase customer as it works to close the gap with Taiwan Semiconductor Manufacturing, the industry’s dominant foundry.

Samsung Electronics trades in the U.S. as an over-the-counter stock under the ticker SSNLF.

The shares are thinly traded compared to a listed peer like TSM, but the company’s foundry division is the part of the business investors watch most closely for signs it can finally challenge TSMC’s lead.

A few additional data points fill out the picture:

Samsung has historically struggled to match TSMC’s production yields at the most advanced process nodes, a gap analysts have raised repeatedly.

A signed customer is not a shipped chip, and Anthropic has not committed to using Samsung for anything. The company is still deciding whether to proceed at all, according to Bloomberg.

More AI:

Anthropic is also talking to Microsoft and the U.K. startup Fractile about potential chips, The Information reported.

That points to a competitive process rather than a single partner. Losing this deal would cost Samsung little beyond a missed opportunity, while winning it would validate a foundry business still trying to prove it belongs in the same conversation as TSMC.

Anthropic and OpenAI are chasing the same goal from different directions, and neither is abandoning Nvidia to get there.

What is changing is the assumption that AI labs must accept whatever pricing and supply terms Nvidia sets. Microsoft has its own Maia chips, Amazon has Trainium, and Google has TPUs.

The newest AI labs are now trying to build that same leverage for themselves. For Samsung, the stakes reach beyond one contract.

A foundry win with a major AI lab would be the clearest signal yet that its advanced manufacturing can compete with TSMC on merit, not just on price.

Related: White House latest verdict flips script on Anthropic

The Arena Media Brands, LLC THESTREET is a registered trademark of TheStreet, Inc.

Read more 3 Great 4th of July Action Movies to Watch on Prime Video, Disney+ and More

Leave a Reply

Your email address will not be published. Required fields are marked *