Skip to content
TechCrunch·

🤖Meta to Manufacture AI Chips in September

Meta's DIY AI chips could shake up GPU market

TL;DR

Meta plans to manufacture its own AI-specific chips starting this month, potentially disrupting the GPU market. The company is working with Broadcom and TSMC for design and manufacturing.

Meta is set to start manufacturing its own AI-specific chips in September through a partnership with TSMC, following successful testing phases. This move aims to reduce reliance on Nvidia and AMD GPUs by customizing hardware for specific AI workloads. The company's MTIA program has already detailed four new chips, some of which are currently deployed or will be soon. By taking a modular approach, Meta hopes to optimize performance and cost-efficiency for its ranking and recommendation algorithms.

Meta to Manufacture AI Chips in September — TechCrunch

Key Points

1

Meta's MTIA program has four new chips in the pipeline with testing completed within six weeks.

2

The company will manufacture chips through TSMC while working on design with Broadcom since March.

3

Meta is buying RAM from Samsung, storage from Sandisk, and fiber-optic equipment from Sumitomo Electric for its AI infrastructure.

4

Capital expenditures are expected to be between $125 billion and $145 billion this year as Meta ramps up compute capacity.

5

Meta plans to deploy 7 gigawatts of compute in 2023 and double that by next year, securing ARM chips for recommendation systems.

Why It Matters

If you're a data scientist or engineer relying on GPUs from Nvidia or AMD, Meta's DIY approach could shift the market dynamics. The modular chip design under MTIA aims to optimize performance specifically for AI workloads like ranking and recommendation algorithms, potentially offering better cost-efficiency than off-the-shelf solutions.

MetaAI ChipsMTIA ProgramTSMCBroadcom

Frequently Asked Questions

Why does this matter?

If you're a data scientist or engineer relying on GPUs from Nvidia or AMD, Meta's DIY approach could shift the market dynamics. The modular chip design under MTIA aims to optimize performance specifically for AI workloads like ranking and recommendation algorithms, potentially offering better cost-efficiency than off-the-shelf solutions.

What happened?

Meta plans to manufacture its own AI-specific chips starting this month, potentially disrupting the GPU market. The company is working with Broadcom and TSMC for design and manufacturing.

Comments

Subscribe to join the conversation...

Be the first to comment

Enjoyed this article?

Get it daily. 7am. Free. Reads in 5 minutes.

Join 2,069 builders reading daily.