Meta, the parent company of Facebook, has been slow to adopt expensive AI-friendly hardware and software systems for its main business, leading to a significant overhaul of its AI infrastructure over the past year. Despite significant investments, the company has fallen behind peers like Google in deploying its custom-built versions of GPUs, which are necessary for AI processing. The overhaul has led to increased capital expenditures of about $4 billion per quarter and has coincided with a period of severe financial squeeze, leading to layoffs. Meta is now prioritizing the development of generative AI products.
Meta CEO Mark Zuckerberg held a five-hour meeting with his top executives in September 2022 to discuss the company’s computing capacity and its ability to do cutting-edge artificial intelligence (AI) work. According to a company memo dated Sept. 20, the social media giant had been slow to adopt expensive AI-friendly hardware and software systems for its main business, hampering its ability to keep pace with innovation at scale. As a result, the company has been engaged in a massive project for over a year to overhaul its AI infrastructure, including leadership changes, a scrapped AI chip project, and a significant increase in capital expenditures of about $4 billion per quarter.
The overhaul has led to the pausing or canceling of previously planned data center builds in four locations, coinciding with a period of severe financial squeeze for Meta, which has been laying off employees since November. Meanwhile, Microsoft-backed OpenAI’s ChatGPT has surged to become the fastest-growing consumer application in history, triggering an arms race among tech giants to release products using so-called “generative AI,” which creates human-like written and visual content in response to prompts.
Meta’s belated embrace of the graphics processing unit (GPU) for AI work is a key source of the company’s trouble. Until last year, Meta largely ran AI workloads using the company’s fleet of commodity central processing units (CPUs), which perform AI work poorly. The company also started using the custom chip it had designed in-house for inference, but by 2021, that approach proved slower and less efficient than one built around GPUs. Meta was already several steps behind peers like Google, which had begun deploying its custom-built version of GPUs, called the TPU, in 2015.
Meta has also started retooling its data centers to accommodate the incoming GPUs, which draw more power and produce more heat than CPUs. As the work got underway, Meta made internal plans to start developing a new and more ambitious in-house chip that, like a GPU, would be capable of both training AI models and performing inference. The project is set to finish around 2025.
While scaling up its GPU capacity, Meta has had little to show as competitors like Microsoft and Google promote public launches of commercial generative AI products. According to four sources, Meta did not prioritize building generative AI products until after the launch of ChatGPT in November. Even though its research lab, Facebook AI Research, has been publishing prototypes of the technology since late 2021, the company was not focused on converting its well-regarded research into products. However, as investor interest soars, that is changing. Zuckerberg announced a new top-level generative AI team in February that he said would “turbocharge” the company’s work in the area. Chief Technology Officer Andrew Bosworth likewise said this month that generative AI was the area where he and Zuckerberg were spending the most time, forecasting that Meta would release a product this year.
