Meta Plans to Construct AI Data Centres for Hundreds of Billions of Dollars

Early this year, when the company released its Llama 4 model, Meta fell behind in the race for generative AI. In comparison to competitors DeepSeek, OpenAI, Anthropic, and Google, the product was subpar.

Remarkably, Zuckerberg has since gone on a hiring rampage, personally luring top AI talent with compensation packages that range from $100 million to more.

That is only one of the components required for success in generative AI. Another is high-quality data. Hence, Meta's strange $14 billion deal to buy just under half of Scale AI and its leader, Alex Wang.

Zuckerberg's Vision: Exciting Plans for Meta's AI Infrastructure

The infrastructure—technical terms for the AI chips (GPUs, etc.), networking equipment, and data centres—is the third ingredient needed to create, hone, and operate massive AI models like the Llama series.

Moreover, SemiAnalysis revealed Zuck's ambitious plans for Meta's massive new AI infrastructure on Friday. Zuck posted on Facebook on Monday, confirming some information in SemiAnalysis's report.

According to the CEO, Meta plans to construct multiple new AI data centres with power consumption exceeding one gigawatt each. Data centre capacity is measured in this way, and anything in the 1-gigawatt range is incredibly large (or was until recently).

Tent-Based Data Centres

According to a spokesperson, Meta is currently building AI data centres in tents. These tents add capacity to the company's campuses, ensuring the entire “supercluster” facility is more than just a tent. This is advantageous because data centres contain sophisticated, costly equipment that requires cooling control to avoid overheating.

Likewise, Zuck is rapidly constructing data centres in tents, a move reminiscent of Elon Musk's Tesla Model 3 production in 2018, aiming to expedite AI compute capacity.

According to Semianalysis, Meta is focusing on speed in datacentre design, drawing inspiration from xAI's time-to-market and constructing more of them. Real estate and traditional datacentre investors are likely to be shocked by this move.

“From prefabricated power and cooling modules to ultra-light structures, speed is of the essence,” it stated.

Notably, tents can be hot, potentially causing challenges for AI data centres' prefab parts, as reported by Semianalysis, which could lead to workload shutdowns during hot summer days.

Meta plans to build full data centres without tents in the long term, but SemiAnalysis CEO Dylan Patel stated that the company needs these facilities to be operational as soon as possible.

Additionally, Meta is implementing 'tents' to reduce construction bottlenecks in its effort to build data centres quickly, addressing power, capacity, and construction crew constraints, according to Patel.

Zuck's Key Points from Monday

Meta plans to invest hundreds of billions of dollars in computing to build a highly skilled and elite team for its superintelligence initiative. SemiAnalysis reports that Meta is on track to become the first lab to launch a 1 GW+ supercluster online.

The company is developing multiple multi-GW clusters, including Prometheus, which will be launched in 2026; Hyperion, which can scale up to 5 GW over several years; and multiple Titan clusters, one of which covers a significant portion of Manhattan's footprint.

Meta Superintelligence Labs will boast industry-leading compute levels, with the greatest compute per researcher, and is eager to collaborate with top researchers to advance the frontier.

Read more news:

Elon Musk claims that a "massive cyberattack" caused waves of outages on his X

TikTok 'Craze' Caused Peak District's Severe Parking Situation

Microsoft will shut down Skype in May after nearly two decades