Ulanqab's 12.5GW Power Play: The Infrastructure Mirage Behind China's AI Ambitions

The number is staggering: 12.5 gigawatts. That is the pledged data center capacity for Ulanqab, a city in Inner Mongolia, as reported by Goldman Sachs. It is a figure that surpasses the stated goal of OpenAI's Stargate project. But there is a catch. The actual operational capacity today is only 1.2GW. That is a 10x gap between promise and reality. This is not just a number; it is a statement of intent in the global AI infrastructure arms race. But intent is not delivery. The race is on, but the track is still being paved. This report, which has surfaced in the context of a bear market for hype, demands a deeper look. We cannot just report the headline. We have to verify the infrastructure. The narrative is one of a land grab, a policy-driven scramble to secure power and land in anticipation of an AI-driven future. Ulanqab is a key node in China's 'East Data, West Computing' strategy. The regional government has been aggressive, signing commitments with major players like DeepSeek, which has pledged 1GW, Xiaohongshu (600MW), ByteDance, and Alibaba. The city offers low PUE, cheap electricity, and a <5ms latency to Beijing. On paper, it is the ideal spot for a compute hub. In practice, it is a construction site of immense proportions. Let's talk about the technical verification imperative. A 12.5GW data center is not a bigger building; it is a different beast. We are talking about hundreds of thousands of GPUs, liquid cooling systems, and a power grid that can handle the load of a small city. The current 1.2GW is a fraction of the vision. The gap between the promised and the operational is not just a matter of construction time. It is a question of feasibility. The power infrastructure alone—the substations, the high-voltage lines, the redundancy systems—represents a multi-year engineering challenge. The supply chain for advanced GPUs is under geopolitical stress. This is not a sprint; it is a marathon with a supply chain that is running on fumes. Based on my audit experience, I look at these commitments and see 's congestion. The biggest risk here is a classic infrastructure boom-and-bust. These are not all firm, funded orders. Many are 'land reservations' or 'policy captures' designed to secure subsidies and power allocations. The smartest signal to watch is not the signed MOUs but the capital expenditure. When DeepSeek and ByteDance start reporting significant capex specifically for Ulanqab, then we can talk about real growth. Until then, this is a land grab. The Ulanqab story is a microcosm of the broader crypto and AI hype cycle. We saw this in 2017 with ICOs; we are seeing it now with AI infrastructure. The promise is 'AI needs compute, so we will build compute.' The reality is that we are building massive, illiquid assets based on a demand forecast that is, at best, uncertain. Here is the contrarian angle: the market is focused on the demand for AI compute, but it is ignoring the cost of the compute itself. Ulanqab's advantage is cheap electricity and low latency to Beijing. But this is also its weakness. A data center is just a building; the real value is in the chips and the software that runs on it. The 'smart' narrative is that Ulanqab will be the 'compute capital' of China. The 'dumb' narrative is that it will become a graveyard of unused power capacity. The difference lies in the technology deployed inside those buildings. If the advanced chips cannot be sourced due to export controls, the entire project's value proposition collapses. The infrastructure is only as good as the hardware it can house. The promise is a bill of goods, and the delivery is tied to a global supply chain that is under threat. The market is pricing in the promise, not the risk of the hardware gap. So, what is the takeaway? The watch is not on the number of gigawatts pledged, but on the number of gigawatts that are 'powered on' and 'producing revenue.' The real test will be in 2025 and 2026. Can these projects actually get the grid power? Can they get the chips? Can they find the customers who will pay for the compute? The answer to these questions will determine whether Ulanqab becomes a strategic asset or a costly monument to over-optimism. The first signal of a problem will be when a 'secured' client walks away from a commitment. The second will be when the operational capacity starts to grow at a pace that lags the debt. The 12.5GW is a vision. The 1.2GW is a reality. The gap between the two is where the risk lives. This is the infrastructure, and it is not yet done. The next watch is the quarterly earnings calls. Look for the word 'Ulanqab' in the financial statements. That is when the promise becomes a business.