Meta Is Building Data Centers in Tents to Speed Up AI Development

Meta has built six large fabric tent structures to house computer equipment as part of an effort to cut data center construction time roughly in half, the company confirmed in June 2026. The approach borrows a strategy Tesla used when building its factories: putting equipment into temporary structures while permanent buildings are still under construction, rather than waiting for everything to be finished first.

Why Speed Matters

Normally, a large data center takes two to four years to build from start to finish. If you're Meta, racing to build the AI systems the company wants to create, waiting that long is costly. Getting powerful computers online six months faster means you can start training artificial intelligence models sooner, and begin earning money from those models earlier.

Meta tried this tent strategy at its data center campus in New Albany, Ohio. According to TechCrunch, the six tent structures cut build time by about 50 percent. That is a significant acceleration in an industry where every month counts.

The Ohio Campus and Its Power Needs

The New Albany campus will require over 1 gigawatt of electrical power when it is fully built out, according to Data Centre Magazine. To put that in perspective: a single large nuclear power plant produces about 1 gigawatt of electricity. This data center will rival that in terms of power consumption. The campus is scheduled to come online in 2026, as reported by Tom's Hardware, making it one of Meta's largest data center projects.

Two Big Projects With Different Goals

Meta is also building something even larger called Hyperion, a data center campus so big that Tom's Hardware compared its land area to Manhattan. Hyperion will take several years to build, a reminder that tents are useful for getting computers online quickly, but not for solving long-term capacity needs.

The two projects work together. The Ohio campus (Prometheus) uses tents to get capacity up and running fast. Hyperion builds the foundational infrastructure for the coming decade.

Why Meta Needs This Much Capacity

In July 2025, Meta said it has the computing power and resources to build what it calls Personal Superintelligence — AI systems tailored to individual people that are available all the time, not just tools for specific tasks.

This is connected to Meta's work on large language models — software that can read and write text and understand language the way humans do. In July 2024, Meta released Llama 3.1 405B, which the company described as the largest and most capable openly available language model at that time. Making models like this — and training bigger ones in the future — requires enormous amounts of computing power. The data centers Meta is building now are designed to support that work.

The Real Constraint: Power, Not Construction

Here is something important that does not get solved by building with tents. Connecting this much power to the electrical grid is slow. In the United States, getting 1 gigawatt of power delivered to a data center typically involves waiting in a queue with regional power companies for years. Tents speed up construction, but the power companies work on their own timeline. Whether the Ohio campus can actually use all its computers when the tents come down will depend on whether the power grid is ready. Right now, that is still an open question.

How Equipment Works Inside Tents

The tent approach works by putting large fabric enclosures around computer equipment while permanent buildings continue to be constructed. There are some engineering challenges worth noting. Keeping computers cool inside a fabric tent is different from keeping them cool in a purpose-built building. Fire safety, water leaks, and air flow all need careful attention. Meta has not publicly shared the technical details of how it solved these problems, so we do not yet know if operating these tent structures for years will cause unexpected issues. The 50 percent time saving is real, but the long-term reliability track record is still being written.

A Familiar Pattern From Technology History

Those who watched the early days of cloud computing will see something familiar here. In the mid-2000s, Google began deploying servers in shipping container-sized modules to speed up data center deployment and avoid waiting for conventional buildings to be completed. The tech industry eventually adopted this lesson. Meta's tent approach is similar — the building envelope is treated as a constraint to work around, not a requirement. The difference is urgency. In 2005, companies like Google were competing for search market share over months. In 2026, Meta is competing with companies like OpenAI, Google DeepMind, Anthropic, and xAI in the race to build the next generation of AI models, with new releases weeks apart.

What This Makes Possible

If tent-based data centers work well and can be replicated, Meta can put new computer chips online about twice as fast as traditional construction allows. In a world where powerful AI chips are hard to get and every month of delay is expensive, that speed gain matters. Getting a training cluster ready six months earlier compresses the entire timeline from developing an AI model to having it ready to use.

Whether other tech companies copy this approach depends on Meta being willing to share the technical details. Historically, Meta has been generous in releasing infrastructure research to the industry — most notably through the Open Compute Project — so there is reason to expect that the details will eventually become public.

For now, Meta has six working tent structures and a 1 gigawatt campus coming online in 2026. The race for AI computing power is being run on a tighter schedule than any previous data center buildout, and companies willing to treat building construction as flexible appear to be gaining ground.