Breaking

Anthropic Starts Custom AI Chip Effort, Talks Samsung on Making It

July 5, 2026 at 12:00 EDT

Infra & Chips
Foundation Models

Anthropic has begun early-stage work on its own AI chip and is in talks with Samsung Electronics as a potential manufacturing partner. The Information reported the move on July 2, 2026, followed by TechCrunch and Bloomberg. The discussions reportedly involve using Samsung's 2nm process and advanced packaging technology.

July 2, 2026 · Anthropic

Anthropic Eyes Its Own AI Chip — In Talks With Samsung

Early-stage work on a custom chip, with Samsung's 2nm process and advanced packaging in view. The goal: cut inference costs and ease reliance on NVIDIA — though the project's targets, timing, and even its survival remain undecided.

~$9B

Revenue run rate, end of 2025

$30B+

Projected run rate in 2026

$65B

Series H raised at a $965B valuation

Revenue run rate: a 3×+ leap in one year

~$9B

End 2025

$30B+

2026 (proj.)

Surging Claude demand makes securing compute — and cutting its cost — an increasingly heavy priority.

Custom silicon is now the norm among major labs

OpenAI — "Jalapeño" inference chip with Broadcom

Google — TPU

Amazon — Inferentia / Trainium

Microsoft — Maia

Meta — MTIA

Anthropic — Early-stage work (Samsung talks)

Anthropic had been one of the last major labs still fully dependent on third-party silicon.

The deeper aim: escaping the "NVIDIA tax"

20–30%

estimated inference-cost cut from a task-tuned custom ASIC — echoing Google TPU's track record.

Model–hardware co-design for efficiency

Ease supply constraints & high costs

Diversify away from Taiwan dependence

What could go right

Samsung 2nm process + advanced packaging in focus
HBM, packaging & logic delivered together as a differentiator
A potential revival signal for Samsung Foundry

Big caveats

Very early stage — continuation not yet certain
Use, performance targets & server integration undecided
No release timing, benchmarks or real-world evaluation yet

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.