Meta Platforms Inc. is focusing on making a transition to independent artificial intelligence (AI) operations by introducing its new in-house chip. This strategic move is designed to reduce its dependency on semiconductors supplied by Nvidia Corp and other external providers.
The latest version of the Meta Training and Inference Accelerator (MTIA) was revealed on Wednesday. This chip is crucial in ranking and recommending content across social media platforms like Facebook and Instagram. The very first MTIA product was introduced by Meta last year.
The tech giant’s shift towards AI services has dramatically increased the demand for computational power. Meta released its AI model last year as a competitive response to OpenAI’s ChatGPT. It also incorporated generative AI features into its social apps, such as the creation of custom stickers and chatbot characters that resemble celebrities.
In October, Meta declared its plan to invest up to $35 billion in AI-supporting infrastructure. This includes substantial investment in data centers and hardware. Mark Zuckerberg, Meta’s CEO, confirmed that AI will dominate the company’s investment areas in 2024. A large portion of this investment would still be directed towards Nvidia, known for its highly sought-after H100 graphics cards that power AI models.
Earlier this year, Zuckerberg had stated that Meta would buy 350,000 of Nvidia’s chips, each priced at tens of thousands of dollars. However, like other tech behemoths such as Amazon’s AWS, Microsoft, and Google’s Alphabet, Meta is striving to develop its in-house chips to eliminate expensive dependencies. Yet, these endeavors have so far not impacted the tech industry’s relentless demand for Nvidia’s AI accelerators.
Nvidia has greatly benefited from the surge in AI, securing a position as the world’s third most valuable tech company, trailing only Microsoft and Apple. Nvidia’s sales to data centers operators increased tremendously from $15 billion to $47.5 billion in fiscal 2024. Analysts predict that this figure will more than double in fiscal 2025.