The Information, a U.S. information technology (IT) outlet, reported on the 26th, citing sources, that Meta, the parent company of Facebook and Instagram, signed a deal to adopt artificial intelligence (AI) chips from Google, following Nvidia and AMD. The move is seen as Meta turning to a series of large-scale AI chip contracts after running into a wall with its in-house AI chip development.
According to the report, Meta signed a multiyear deal worth several billion dollars to rent Google's AI chip, the tensor processing unit (TPU). In addition to this computing rental agreement, Meta is also in talks on a purchase contract to supply Google TPUs directly to its data centers for installation.
Meta on the 17th signed a contract, estimated at tens of billions of dollars, to adopt Nvidia's AI chips, and on the 24th struck an AI chip supply deal with AMD worth $100 billion (about 143.72 trillion won).
Observers say Meta's push to sign AI chip adoption contracts stems from setbacks in the in-house chip development it had been pursuing. According to The Information, Meta halted its plan to develop cutting-edge in-house AI chips under the name "Meta Training and Inference Accelerator" (MTIA) and shifted its development direction to a simpler version.
Meta canceled development of a training chip codenamed "Olympus," and scrapped one version in the plan for another chip, "Iris." The decision was recently communicated to the AI infrastructure division.
Originally, Meta planned to develop the Olympus chip, install it in a large-scale server cluster, and use it for AI training.
However, Meta's management reportedly judged that carrying out this plan could leave it behind OpenAI or Google in the race to develop AI models. They cited concerns that the related equipment and software are not as stable as Nvidia's products and that the chip design is complex, likely causing difficulties in mass production.
A person working in Meta's in-house chip development unit told The Information, "Given the risks of development delays or redesigns, there is a lot of internal skepticism about whether we can build a chip that matches Nvidia's performance."
Regarding this, a Spokesperson for Meta said, "We remain committed to continued investment in a diverse silicon portfolio, including advancing the MTIA portfolio," adding, "We will share more this year." Meta is expected to invest $135 billion (about 194 trillion won) in AI infrastructure this year.