Archive — US — techcrunch
Meta is reportedly testing in-house chips for AI training
Meta is testing in-house chips for AI training to minimize reliance on Nvidia. Manufactured with TSMC, the chip targets AI-specific workloads. Currently in a small deployment phase, if successful, Meta plans to scale production. Previous attempts at custom chips faced challenges, but saving on Nvidia GPU costs could significantly impact their $65 billion capital expenditure this year.
Key points
- Meta is developing an in-house chip for AI workloads to reduce reliance on Nvidia.
- The chip is being tested in small deployments and aims to handle AI training specifically.
- Manufactured in cooperation with TSMC, it follows several past efforts that failed to meet expectations.
- A successful test could lead to increased production and significant cost savings on Nvidia GPUs, which constitute a large part of Meta's projected $65 billion expenditure this year.
Why it matters
- This development reflects the ongoing trend in 2025 of companies seeking to develop proprietary hardware to optimize AI performance and reduce costs.
- Historically, firms like Google and Amazon have pursued similar strategies to decrease reliance on third-party manufacturers.
- As AI adoption increases, the demand for custom chips is expected to rise, pushing companies to innovate and invest in in-house solutions.
Meta's venture into in-house AI chips highlights a significant trend in the tech industry to innovate hardware solutions, potentially reshaping their cost structure and operational efficiency in a competitive market.