Skip to main content
Editorial sketch of hands assembling a custom AI chip on a workbench surrounded by silicon wafers and engineering tools
Undercurrent
BENTHIC

Inside the Custom Silicon Race: How Meta, Google, and Amazon Are Trying to Break Free from NVIDIA

VERIFIEDConfidence: 80%

Introduction

Last week, Meta published a blog post announcing four new generations of its in-house AI chip, called MTIA — short for Meta Training and Inference Accelerator. According to Meta's announcement, the company has already deployed...

Create an account to read this article

Sign up for a free account to get full access to in-depth AI coverage, analysis, and investigations.

Related

ARTICLEUndercurrentPRO

Meta Moltbook Deal Signals an Agent-First Future

DEEP DIVEUndercurrentPRO

The Gravity Problem: What the AI Arms Race Actually Costs, and Who Can Afford It

Mistral AI was founded as Europe's answer to OpenAI. By 2026, Bloomberg is calling it a consultancy. This deep dive examines the structural economics driving every mid-tier AI lab away from frontier research: compute costs growing 4-5x annually, a $600B revenue gap, and the gravitational pull of enterprise services. Using Mistral, Aleph Alpha, and Stability AI as case studies, the piece reveals why competing at the frontier without hyperscaler backing has become structurally unviable.

ARTICLEUndercurrentPRO

The AI Dependency Map Most Enterprises Never Built