DeepSeek, the Chinese AI company that rattled global tech markets earlier this year, is now planning to build a 1 gigawatt AI data center in Inner Mongolia. Some capacity is expected to come online by early 2028.

To put 1GW in perspective, that’s roughly the output of a nuclear power plant. It’s also more power than many mid-sized cities consume.

What’s actually happening on the ground

DeepSeek has started hiring server maintenance engineers and delivery managers in Ulanqab, a city in Inner Mongolia that’s quietly becoming one of China’s preferred data center corridors. The job postings, noted around April 2026, suggest the company is well past the planning phase and actively preparing for physical deployment.

This isn’t DeepSeek’s first footprint in the region. The company reportedly completed deployments of its models in Hohhot’s New District, also in Inner Mongolia, back in February 2025. So the buildout in Ulanqab represents an expansion of an existing regional strategy, not a cold start.

The company, headquartered in Hangzhou, is simultaneously developing its V4 AI model. The new data center is expected to provide the compute backbone for training and running that model at scale.

Why crypto investors should care about an AI data center

DeepSeek’s project has no direct connection to cryptocurrencies, tokens, or digital assets. The research confirms this explicitly.

Second, there’s the GPU supply chain factor. DeepSeek’s massive infrastructure buildout means more demand for high-end chips. When DeepSeek’s R1 model launched in late January 2025, it triggered substantial market reactions and movements in tech stocks precisely because investors recognized the cost-effectiveness implications.

This is the Jevons paradox at work. When something becomes more efficient, people don’t use less of it. They use more. DeepSeek’s R1 demonstrated that you could train competitive AI models at a fraction of the usual cost. The market’s response wasn’t to assume less compute would be needed. It was to realize that cheaper compute means everyone wants more of it. Hence, a 1GW data center.

The competitive landscape is shifting fast

The R1 model’s January 2025 release didn’t just impress AI researchers. It spooked investors in Western tech companies who had assumed that cutting-edge AI required Western-scale budgets.

Investors watching the intersection of AI and crypto should track several things in the coming months: whether DeepSeek secures dedicated power purchase agreements for the Inner Mongolia facility, how the V4 model’s capabilities compare to competitors when released, and whether the company’s infrastructure timeline holds. Early 2028 is ambitious for a project of this scale, and construction delays in data center buildouts have become almost routine across the industry.