Meta is housing AI chips in tents in Ohio to outrun its own construction timelines

Meta has built six tents outside New Albany, Ohio, and stuffed them with AI accelerators worth what is probably billions of dollars. The company prefers the term "rapid deployment structures," but the satellite imagery and local building permits surfaced by Michael Thomas, founder of the data center tracking outfit Cleanview, show exactly what they are: weatherproof fabric buildings erected fast and powered by generators parked next door.

The basic claim here is that you can skip most of the slow part of building a data center. A conventional hyperscale facility is a hardened concrete-and-steel project that takes a couple of years to permit, pour, and fit out. Thomas' review of city permits shows Meta started building five 125,000-square-foot tents between April and June of this year, and his satellite images show all of them already standing. Compressing that timeline is the entire point.

What's actually new here

Not much, and that is worth being honest about. Mark Zuckerberg described this plan to The Information last year, talking openly about using tents to house multi-gigawatt compute. Thomas' contribution is evidence that the plan is real and moving quickly, not a fresh strategy reveal.

The individual pieces are all borrowed. The tent itself is a direct lift from Tesla, which famously bolted a giant tent onto its Fremont factory parking lot in 2018 to hit Model 3 production targets when the permanent line couldn't keep up. The power setup, roughly 200 megawatts of modular gas turbines sitting next to the site, is the trick xAI used to bring its Memphis cluster online without waiting for grid interconnection, which is now the single longest pole in the tent for anyone trying to add large loads to the US electric system.

So the genuinely new element is the combination: pairing throwaway-fast structures with behind-the-meter gas generation to decouple your compute buildout from both the construction industry and the utility queue. That is a logistics decision, not a technical one. The chips inside, the networking, and the cooling are presumably the same hardware Meta would deploy in a permanent hall.

The trade-offs nobody is advertising

A fabric building does not get you out of the hard problems. Modern AI training racks dissipate enormous heat, and thermal management inside a tent in an Ohio summer is a real engineering constraint, not a footnote. Humidity, dust, and temperature swings are exactly the conditions data center design normally spends a fortune to eliminate. Meta is betting it can manage those with the mechanical systems alone rather than the building envelope, which is plausible for a deployment meant to be temporary or semi-temporary.

Then there is the power source. On-site gas turbines are fast, but they are generators burning fossil fuel next to a company that has spent years on clean-energy commitments. De Chant filed this under TechCrunch's Climate desk for a reason. "Rapid deployment" and "net zero" are pulling in opposite directions, and the turbines are the part of this story most likely to draw regulatory and local pushback.

Why Meta is in a hurry

The context that makes the tents legible is financial and competitive pressure. Meta has guided toward spending as much as $145 billion on data centers and other capital expenditures, a number Wall Street has greeted by marking the stock down about 5% on the year. When investors are nervous about capex, shaving construction cost and time off each gigawatt is one of the few levers you can pull without cutting the chip order itself.

There is also a product problem in the background. The Wall Street Journal recently reported that Meta's latest model, Muse Spark, is finished, but the developer APIs that would let anyone actually use it have been repeatedly delayed. That is the more telling signal for practitioners than any tent. Having the weights done and not shipping access usually points to safety review, serving infrastructure, or capacity bottlenecks rather than the model itself. Building compute capacity at speed and getting a finished model into developers' hands are different failures, and right now Meta appears to be hitting both at once.

What it actually changes

For the broader buildout, the tent is a tell. It shows that the binding constraint on AI capacity in 2026 is no longer chips or capital, it is the physical and electrical infrastructure to plug them in, and companies are willing to abandon decades of data center orthodoxy to route around that. Expect more behind-the-meter gas and more semi-permanent structures from the other hyperscalers if Meta's Ohio sites run without incident.

For everyone watching model releases, the lesson is more sobering. You can erect a building in eight weeks. You cannot will an API into existence on the same schedule, and the part of the stack that determines whether any of this compute reaches developers is the part Meta hasn't managed to accelerate. TechCrunch has reached out to Meta for comment. The tents, at least, are already up.