Meta's in-house chip development dream faces setbacks, shifting towards more external collaborations

As Meta Platforms reaches new chip supply agreements with AMD, Nvidia, and Google, its internally designed AI chips are facing obstacles.

According to sources familiar with the matter, last week, Meta abandoned its most advanced AI training chip in development due to design difficulties, shifting focus to a simpler version.

This decision highlights the challenges the tech giant faces in designing AI chips capable of competing with Nvidia, which currently maintains a dominant position in the AI chip market.

In-House Projects Encounter Obstacles

Meta’s self-developed AI chips are part of its “Meta Training and Inference Accelerator (MTIA)” project, which is part of the company’s broader effort to develop proprietary AI hardware and reduce reliance on external chip manufacturers like Nvidia. The goal is to lower costs and gain more control over data center infrastructure.

Under the MTIA project, Meta originally planned to complete the design of a chip codenamed Olympus by Q4 2026. However, concerns over technical complexity, manufacturing risks, and software stability have led Meta to reconsider this design.

Notably, Meta previously abandoned an early design codenamed Iris. Now, with internal teams struggling to build products that can truly rival Nvidia’s ecosystem, Olympus may face the same fate as Iris.

In a statement, a Meta spokesperson said, “We remain committed to investing in a diverse portfolio of silicon products to meet our needs, including advancing our MTIA product line. We will share more updates this year.”

Like Meta, many other tech companies, such as Microsoft, have encountered similar issues when designing their own AI chips. Last year, Nvidia CEO Jensen Huang publicly stated that most large tech companies would abandon their chip projects because those chips would not perform at Nvidia’s level.

The setbacks in developing in-house chips have led Meta to rely more on external partnerships. On Tuesday (February 24), Meta and AMD announced a collaboration to purchase $60 billion worth of AI chips; earlier this month, Meta also signed an agreement with Nvidia to buy its current and future AI chips.

Meanwhile, reports this week indicate that Meta has signed a multi-billion dollar deal with Google to lease AI chips from Google for developing new AI models. This procurement plan could be implemented as early as next year, but the progress of these discussions remains uncertain.

Meta previously projected its capital expenditure for 2026 to be between $115 billion and $135 billion, with most of the funds allocated to chips and servers.

(Source: Cailian Press)