The urgency shown by Zuckerberg comes after the company ceded the lead in open-source models to China’s DeepSeek, which unveiled its first model in January.

Meta Platforms, Inc. (META) CEO Mark Zuckerberg’s ruthless pursuit of superintelligence has prompted the social-media giant’s cofounder to go with a three-pronged strategy of focusing on talent, data and infrastructure.

Taking a radically different approach, Zuckberg is now looking to build multi-billion-dollar Graphic Processor Unit (GPU) clusters in “tents,” according to a SemiAnalysis report published Friday.

The report came to the spotlight after the Meta CEO mentioned the research in a Facebook post on Monday. Citing SemiAnalysis, Zuckerberg said the company is on track to be the first lab to bring a supercluster boasting a capacity of over 1 Gigawatt (GW) online.

A Business Insider story highlighted precedence. Building a datacenter in tents is likely “taking a page from Elon's strategy book,” the report said, adding that the world’s richest man built Model 3 cars in tents outside his Fremont, California factory in 2018 to expedite the move to the market.

In Musk’s own words, Tesla was on the brink of bankruptcy at that time, which prompted him to send out the infamous “going private” tweet, landing him on the SEC’s radar.

Amid Meta’s renewed AI push, Meta stock climbed to an all-time high of $747.90 on June 30. On Stocktwits, sentiment toward the stock improved to ‘neutral’ (45/100) by late Monday from ‘bearish’ a day ago. The message volume stayed at a ‘normal’ level.

META sentiment and message volume as of 2:05 a.m. ET, July 15 | source: Stocktwits

SemiAnalysis said, as part of reinventing its datacenter strategy, Meta began focusing on speed above all else when it comes to its datacenter buildout, aping a strategy that Musk’s AI startup xAI adopted in coming online in double quick time. 

Meta’s data center design has given priority to “getting compute online fast” over beauty or redundancy. 

The company is quietly building one of the world’s largest AI training clusters in Ohio, the report said. Zuckerberg has confirmed this by stating that the company is building several multi-GW clusters. The first, named “Prometheus,” would come online in 2026, and another, going by the name “Hyperion,” which can be scaled up to 5GW, will be built over several years.

SemiAnalysis said Meta will likely connect all these sites with ultra-high-bandwidth networks.

“Meta went full Elon mode,” the report said, pointing out that the company is building two 200 Megawatt (MW) on-site natural gas plants when the local power grid proved inadequate.

The urgency shown by Zuckerberg comes after the company ceded the lead in open-source models to China’s DeepSeek, which unveiled its first model in January. It was then that the company identified Talent and compute as its twin shortcomings.

Since then, the entrepreneur has set up a Superintelligence group aiming to achieve artificial general intelligence (AI), poached several AI researchers paying hefty compensation packages, restructured its AI division around the newly formed Superintelligence group, picked up stakes or bought out smaller AI startups and intensified datacenter infrastructure buildout.

Meta ended Monday’s session up 0.48% at $720.92, with the year-to-date gains at a robust 23%.

For updates and corrections, email newsroom[at]stocktwits[dot]com.<