Tech giants have been racing to gain an edge in artificial intelligence for years, but one company may be playing catch-up as the pressure intensifies. In a memo from September, a senior executive at Facebook parent Meta reportedly noted that the company still lagged behind in A.I.
“We have a significant gap in our tooling, workflows, and processes when it comes to developing for A.I. We need to invest heavily here,” Santosh Janardhan, the head of infrastructure at Meta, wrote on an internal message board in September, Reuters reported Tuesday.
The memo also stated that Meta’s A.I. work would require it to “fundamentally shift our physical infrastructure design, our software systems, and our approach to providing a stable platform.”
Almost a decade after starting to invest in full-fledged A.I. research and pouring several billions of dollars into the effort, Meta seems to still have fallen behind in adopting the hardware and software to support its A.I. ambitions, the Reuters report said, citing interviews with people close to the matter, company statements, and the memo.
The tech behemoth, which also owns Instagram and WhatsApp, is now trying to grow its A.I. prowess amid the meteoric rise of OpenAI’s ChatGPT, a chatbot that can give detailed answers to prompts in a matter of seconds. But in addition to its shortcomings with A.I. infrastructure, Meta has been hamstrung in the race by its focus elsewhere, including on the metaverse, a virtual universe that has been one of CEO Mark Zuckerberg’s obsessions, according to Reuters.
In February, Meta announced a tool, LLaMA, that it would use in its chatbots and other tools as it tried to compete with rivals Google and Microsoft. But with two rounds of layoffs and leadership changes, Meta is scrambling.
Part of the issue identified in Reuters’ report is Meta’s late adoption of graphics processing units, or GPUs, which can help speed up tasks, cutting down the time needed to process vast amounts of data and making it a fit for certain A.I. projects. Meta was also reportedly using its own custom inference chips, a process in which A.I. algorithms answer prompts and make judgements. The approach of using GPUs and custom chips slowed the company’s A.I. progress, according to two sources with knowledge of the matter, Reuters reported. Meta eventually stopped using its in-house chips and opted instead for chipmaker Nvidia’s GPUs, but by this time it was already well behind its competitor, Google.
It was also around this time that Meta had a shake-up in its leadership on the A.I. infrastructure side—including Janardhan taking on his role as the head of infrastructure.
Meta has been increasing its capital expenditures, or money invested on the company’s fixed assets like equipment or land, through 2022, according to Reuters. That increase alone was as high as $4 billion per quarter through much of last year and is expected to continue to remain high through 2023 as well. This increase is, in large part, due to Meta’s bid to expand its A.I. capacity, Reuters reported.
But the Menlo Park, Calif.–based company, which began its full-scale A.I. research in 2013, believes it can still break new ground when it comes to this technology.
“Meta has a proven track record in creating and deploying state-of-the-art infrastructure at scale combined with deep expertise in A.I. research and engineering,” a Meta spokesperson told Fortune, echoing the remarks it made in the Reuters report.
“We’re confident in our ability to continue expanding our infrastructure’s capabilities to meet our near-term and long-term needs as we bring new A.I.-powered experiences to our family of apps and consumer products.”
The rush is real
The introduction of ChatGPT in late November came just weeks after Meta laid off 11,000 employees. ChatGPT, whose owner is backed by Microsoft, quickly gained millions of users and sparked companies even outside tech to adopt similar capabilities. Meta’s job culling came as part of its broader strategy in 2023 of pursuing a “year of efficiency,” which Zuckerberg said would involve flattening the company structure and adopting A.I. to improve productivity.
Meanwhile, companies like Google and Microsoft also unveiled A.I. chatbots in February. Weeks later, Meta said it was working on similar technology.
“We’re creating a new top-level product group at Meta focused on generative A.I. to turbocharge our work in this area,” Zuckerberg said in a Facebook post in late February. “Over the longer term, we’ll focus on developing A.I. personas that can help people in a variety of ways.”
He added that the A.I. could be applied in text-based apps like WhatsApp or visual filters on Instagram. Details about Meta’s plans for using A.I. across apps and the timeline for doing so remains unclear, although Meta CTO Andrew Bosworth told Nikkei Asia last month that the company hopes to commercialize its generative A.I. by December. Bosworth also said that he and Zuckerberg were spending much of their time on generative A.I.