Meta builds AI data centers in tents to slash deployment time and cut billions in costs

Meta Deploys Tent-Based Data Centers to Accelerate AI Infrastructure

Meta has built six massive weatherproof tents outside New Albany, Ohio to house AI servers, marking an unconventional approach to the AI infrastructure race. The structures, officially termed rapid deployment structures, represent a dramatic shift from traditional data center construction methods1

1

. Each tent spans approximately 125,000 square feet and contains AI chips likely worth billions of dollars, according to Michael Thomas, founder of Cleanview Energy, a data center tracking firm2

2

.

City permits reviewed by Thomas show that Meta started constructing five tents between April and June 2026, and recent satellite images confirm all structures have been completed1

1

. This timeline represents a significant acceleration compared to the five traditional buildings at the same site, which took approximately two to three years to complete2

2

.

Tesla and xAI Inspire Unconventional Construction Strategy

The tent approach borrows directly from Tesla, which used similar canvas structures in the parking lot of its Fremont, California factory during the frantic Model 3 production ramp. Meta CEO Mark Zuckerberg first revealed plans to use weatherproof tents for multi-gigawatt data centers in a conversation with the Information last year1

1

.

The strategy also takes cues from Elon Musk's xAI, which built a 100,000-strong AI data center in just 19 days in 2024—a process that typically takes four years, according to Nvidia CEO Jensen Huang2

2

. Meta is now applying this technique to at least three sites across the United States, including locations in Tennessee2

2

.

Behind-the-Meter Power and the Prometheus Project

The Ohio site, part of Meta's Prometheus project, uses 200 megawatts of modular gas turbines installed on-site to provide power1

1

. This behind-the-meter approach allows the company to operate independently of the power grid, similar to xAI's Memphis Supercluster, which initially relied on portable power generators2

2

.

According to Cleanview's analysis, there's currently about 2GW of power capacity available from behind-the-meter data centers, with an additional 1GW expected to come online this year for a total of 3GW. If ongoing projects stay on schedule, total capacity could reach 13GW by the end of 2027—equivalent to the output of 13 nuclear power plants and enough to power New York City2

2

.

Pressure Mounts Amid Delayed AI Rollouts and Massive Capital Expenditures

The data centers in tents have emerged as Meta struggles to release AI models to developers. A recent Wall Street Journal report indicated that while the company's latest model, Muse Spark, is complete, the APIs that developers rely on to access large language models have been repeatedly delayed1

1

.

Meta has committed to spending up to $145 billion on AI data centers and other capital expenditures, a figure that has contributed to the company's stock trading down 5% this year1

1

. Deploying AI compute resources in tents represents one method to reduce costs while maintaining the pace of infrastructure buildout.

The urgency reflects broader industry challenges. Nearly half of the AI data centers scheduled to open this year have been cancelled or significantly delayed, creating a critical bottleneck for companies dependent on access to AI chips3

3

. As communities across the United States increasingly oppose years-long data center construction projects, the tent approach could signal a broader trend in how tech companies deploy AI infrastructure3

3

.