Anthropic is discussing a new custom chip with Samsung
TechCrunch
Anthropic is reportedly talking to Samsung about building a custom AI chip. It still won't say what the chip does, but it's clearly trying to loosen its grip on Nvidia.
Based on reporting by TechCrunch — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Anthropic's chip ambitions just got a little more concrete. The Information reported Thursday that the company is in talks with Samsung about co-developing custom silicon, a follow-up to those April rumblings from Reuters about Anthropic wanting its own chips to dodge supply shortages. Details remain thin — no word yet on whether this thing is meant for training or inference, how it slots into server racks, or how powerful it will actually be.
Anthropic isn't exactly rushing to confirm anything either. Asked for comment, the company told TechCrunch that its strategy still leans on a mixed hardware stack pulling from Google, Amazon, and Nvidia, and declined to say anything specific about Samsung. That's a fairly standard hedge for a company that doesn't want to spook existing partners or tip its hand too early.
The timing is hard to ignore. Just last week, OpenAI announced its own custom inference chip, called Jalapeño, built in partnership with Broadcom, and claimed better performance-per-watt than rival hardware. Amazon and Google already run their own TPUs inside their clouds. Everyone, it seems, wants a way out from under Nvidia's pricing and supply constraints, even as Nvidia remains the default choice for anyone training frontier models.
Samsung, notably, sits on both sides of this equation. It's a major Nvidia manufacturing partner, producing chips Nvidia needs while relying on Nvidia's software to make them — the two are even building a joint AI chip factory in South Korea. Samsung has also floated a chip partnership with Google. So if Anthropic and Samsung do team up, it wouldn't be some clean break from the Nvidia ecosystem, but rather another node in an increasingly tangled web of alliances, competitors, and frenemies all racing to control their own hardware destiny.
My take — AI-written commentary, not fact-checked reporting
Every AI lab wants its own chip now, and I get why, but let's not pretend this is really about independence — it's about leverage in future Nvidia price negotiations. The real story here is Samsung playing all sides at once: foundry partner to Nvidia, potential chip partner to Google, and now maybe Anthropic too. That's not diversification, that's just smart hedging by the one company everyone still needs.
Read more about this at: TechCrunch