Meta acquires AI networking chip team from Graphcore

Meta Platforms has recently hired a team based in Oslo that was previously working on artificial intelligence networking technology at the British chip company Graphcore. A spokesperson for Meta confirmed the hiring of the team after Reuters discovered that 10 individuals, who had been employed at Graphcore until December 2022 or January 2023, had updated their LinkedIn profiles to indicate that they had joined Meta in February or March of this year.

“We recently welcomed a number of highly specialised engineers in Oslo to our infrastructure team at Meta. They bring deep expertise in the design and development of supercomputing systems to support AI and machine learning at scale in Meta’s data centres,” said Jon Carvill, the Meta spokesman.

By hiring the team from Graphcore, Meta Platforms is bolstering its efforts to enhance the handling of AI work in its data centers. The company is facing increasing demand for AI-oriented infrastructure from teams across the organization seeking to develop new features.

Meta, the owner of Facebook and Instagram, relies heavily on AI technology for tasks such as targeting advertising, curating content for its apps’ feeds, and removing prohibited content from its platforms.

In addition to these existing uses of AI, Meta is now competing with companies like Microsoft and Google to release generative AI products capable of creating human-like writing, art, and other content. Investors see this as a major growth area for tech companies.

The LinkedIn profiles of the 10 employees hired by Meta indicate that they had previously worked on AI-specific networking technology at Graphcore, a company that specializes in developing computer chips and systems optimized for AI work.

Carvill declined to say what they would be working on at Meta.

Restructuring

Graphcore closed its Oslo office as part of a larger restructuring effort announced in October of last year. A spokesperson for the start-up stated that the company was facing challenges in competing against US-based firms such as Nvidia and AMD, which dominate the market for AI chips.

Meta already has an internal unit dedicated to designing various types of chips to improve the speed and efficiency of its AI work. This includes a network chip that acts as a sort of air traffic control for servers, according to two sources.

Efficient networking is particularly important for modern AI systems, such as the chatbot ChatGPT or the image-generation tool Dall-E. These systems are too large to fit on a single computing chip and must be distributed across multiple chips. A new category of network chip has emerged to facilitate smooth data transfer within these computing clusters. Companies such as Nvidia, AMD, and Intel all produce such network chips.

In addition to its network chip, Meta is also developing a complex computing chip for both training AI models and performing inference, a process in which trained models make judgments and generate responses to prompts. However, this chip is not expected to be ready until around 2025.

Graphcore, one of the UK’s most valuable tech start-ups, was once viewed by investors such as Microsoft and venture capital firm Sequoia as a potential challenger to Nvidia’s dominant position in the AI chip systems market.

However, the company suffered a setback in 2020 when Microsoft cancelled an early agreement to purchase Graphcore’s chips for its Azure cloud computing platform, according to a report by UK newspaper The Times. Instead, Microsoft used Nvidia’s GPUs to build the infrastructure for OpenAI, the developer of ChatGPT, which is also backed by Microsoft.

Sequoia has since written down its investment in Graphcore to zero, although it remains on the company’s board, according to a source familiar with the relationship.

A spokesperson for Graphcore confirmed these setbacks but stated that the company is “perfectly positioned” to capitalize on the growing commercial adoption of AI.

Graphcore was last valued at US$2.8 billion after raising $222 million in its most recent investment round in 2020. — Katie Paul, Krystal Hu and Stephen Nellis, © 2023 Reuters