An Oslo-based team that was developing artificial intelligence networking technologies at British chip startup Graphcore until late last year has been hired by Meta Platforms.
After Reuters discovered 10 persons whose LinkedIn accounts claimed they had worked at Graphcore until December 2022 or January 2023 and then joined Meta in February or March of this year, a Meta spokesman acknowledged the hirings in response to a request for comment.
“Our infrastructure team at Meta recently welcomed a lot of highly specialized engineers to Oslo. In order to support AI and machine learning at scale in Meta’s data centers, they contribute significant knowledge in the design and development of supercomputing systems, according to Meta spokesperson Jon Carvill.
The decision gives the social media behemoth more clout in its fight to better manage AI work in its data centers as it struggles to keep up with the demand for AI-oriented infrastructure from teams throughout the firm eager to develop new products.
Facebook and Instagram’s parent company, Meta, has come to rely more and more on AI technology to target advertising, choose content for its applications’ newsfeed, and remove prohibited information from its platforms.
Additionally, IBM is curren
tly trying to catch up with rivals like Microsoft Corp. and Alphabet Inc.’s Google in the release of generative AI products capable of producing literature, art, and other content that resembles what a human would produce, which investors regard as the next major growth sector for tech businesses.
According to the employment biographies of the 10 individuals on LinkedIn, the group had worked at Graphcore, which creates computer chips and systems designed for AI work, on networking technologies specifically for AI.
Carvill opted not to reveal the project they would be working on at Meta.
According to a Graphcore spokeswoman, the startup closed its headquarters in Oslo as part of a larger restructure that was revealed in October of last year. At the time, Graphcore was struggling to compete with American companies like Nvidia Corp. and Advanced Micro Devices, which control the majority of the AI chip market.
A team within Meta already develops several types of chips to speed up and maximize the effectiveness of its AI work, including a network chip that serves as a kind of server air traffic controller.
Modern AI systems like those powering the chatbot ChatGPT or the image-generation tool Dall-E, which are much too big to fit on a single processor chip and must instead be split out over multiple chips connected together, benefit greatly from efficient networking.
To help maintain smooth data flow within those computing clusters, a new class of network chip has evolved. Such network chips are produced by Intel Corp, AMD, and Nvidia.
Meta is developing a complicated computer chip in addition to its network chip that will be used to train AI models and carry out inference, which is the process by which trained models make decisions and produce solutions to questions. However, Meta does not anticipate that chip to be ready until approximately 2025.
Investors like Microsoft and the venture capital company Sequoia formerly viewed Graphcore, one of the most valuable IT businesses in the UK, as a possible rival to Nvidia’s dominant position in the market for AI chip systems.
According to a UK daily The Times, it encountered a roadblock in 2020 when Microsoft canceled an early agreement to purchase Graphcore’s chips for its Azure cloud computing platform. Instead, Microsoft built the huge infrastructure that powers ChatGPT developer OpenAI, which Microsoft also supports, using Nvidia’s GPUs.
According to a source familiar with the situation, Sequoia has since written off its investment in Graphcore to zero while continuing to serve on the board of directors. Insider initially mentioned the write-down in October.
A Graphcore representative acknowledged the failures but asserted that the business was “perfectly positioned” to benefit from the growing commercial adoption of AI.
The most recent funding round for Graphcore, which raised $222 million in 2020, resulted in a valuation of $2.8 billion.