Meta, the parent company of Facebook and Instagram, has grown more dependent on AI for advertising targeting
Meta Platforms has hired a team based in Oslo, previously employed by Graphcore until late 2022, to bolster its infrastructure team in handling AI work at scale. This move is part of Meta’s push to compete with rivals in introducing generative AI products. The team brings expertise in designing and developing supercomputing systems to support AI and machine learning in Meta’s data centers. The move aims to improve how Meta’s data centers handle AI work, as the company experiences an increase in demand for AI-oriented infrastructure.
Meta, the owner of Facebook and Instagram, heavily relies on AI to enhance its advertising, curate content for its apps’ feeds, and filter out banned material from its platforms. To remain competitive with companies such as Microsoft and Google, Meta is also endeavoring to launch generative AI products that can generate human-like writing, art, and other content. Meta has hired ten engineers from Graphcore, a British chip unicorn, who had been building AI networking technology until December 2022 or January 2023. The Oslo-based team will contribute their expertise in supercomputing systems’ design and development to Meta’s infrastructure team to support AI and machine learning at scale in Meta’s data centers. The new hires’ specific projects at Meta have not been revealed.
According to a spokesperson for the startup, Graphcore closed its Oslo office as part of a wider restructuring in October 2022 due to challenges in competing with US-based companies like Nvidia and Advanced Micro Devices, who dominate the market for AI chips. Meanwhile, Meta has an in-house team working on various chip designs to improve the speed and efficiency of its AI work, including a network chip that performs an “air traffic control” function for servers. This type of networking is particularly important for large AI systems like ChatGPT and Dall-E, which cannot fit onto a single computing chip and need to be split over multiple chips strung together. Companies such as Nvidia, AMD, and Intel also produce similar network chips.
Meta is developing a sophisticated computing chip to train AI models and perform inference, but it will not be available until 2025. In addition to this, Meta is also creating a network chip that acts as air traffic control for servers. Efficient networking is critical for modern AI systems, which are too large to fit onto a single computing chip and must be split across several. Nvidia, AMD, and Intel all manufacture similar network chips.
Graphcore, a leading UK tech startup, was once regarded as a potential competitor to Nvidia’s AI chip system dominance, with investors such as Microsoft and Sequoia Capital showing interest in the company. However, Graphcore closed its Oslo office as part of a broader restructuring last year, as it struggled to compete with US-based firms like Nvidia and Advanced Micro Devices that dominate the AI chip market.
According to The Times, Microsoft reportedly canceled an early agreement to purchase Graphcore’s chips for its Azure cloud computing platform, dealing a blow to Graphcore in 2020. As an alternative, Microsoft used Nvidia’s GPUs to create the massive infrastructure that powers OpenAI, the developer of ChatGPT, in which Microsoft also has a stake. Sequoia, an investor in Graphcore, wrote off its investment, but still remains on the company’s board, according to an insider source cited by Insider in October. Although Graphcore confirmed the setback, a spokesperson said that the company was “perfectly positioned” to capitalize on the increasing commercial adoption of AI.
