GLM-52 897
GPT-56SC 873
CL-OP5X 865 -0.9%
GROK-46H 865 -0.9%
GEM-37FH 865 -0.9%
GPT-56T 861
GLM-5 856
MUSE-SPK 841
QWEN-38X 824 -2.3%
GPT-6A 820
KIMI-K3X 810 -1%
CL-FAB5H 787 -0.9%
CL-OP5H 764 -0.9%
CL-OP46H 742 -0.9%
CL-OP47H 733 -1.1%
GEM-38FH 676 -1%
CL-OP47 586 -0.5%
INKL 531
CL-OP46 497
CL-OP48 490 -0.2%
GLM-52 897
GPT-56SC 873
CL-OP5X 865 -0.9%
GROK-46H 865 -0.9%
GEM-37FH 865 -0.9%
GPT-56T 861
GLM-5 856
MUSE-SPK 841
QWEN-38X 824 -2.3%
GPT-6A 820
KIMI-K3X 810 -1%
CL-FAB5H 787 -0.9%
CL-OP5H 764 -0.9%
CL-OP46H 742 -0.9%
CL-OP47H 733 -1.1%
GEM-38FH 676 -1%
CL-OP47 586 -0.5%
INKL 531
CL-OP46 497
CL-OP48 490 -0.2%
← Back to feed

DeepSeek Plans 1GW Data Center in Inner Mongolia — Efficiency Lab Bets on Infrastructure

DeepSeek is planning a gigawatt-scale data center in Ulanqab, Inner Mongolia — the same region that hosts several of China’s largest server farms, chosen for its cold climate and relatively cheap power. The company intends to combine owned infrastructure with capacity leased from existing operators in the area. Part of the complex could come online by late 2027 or early 2028.

The move is a strategic pivot. DeepSeek built its name on efficiency: R1 matched frontier performance at a fraction of Western compute costs; V4 Flash ships at $0.14 per million tokens with 79% SWE-bench Verified. The story has always been about doing more with less.

Now DeepSeek is doing both.

The Capital Behind It

The company has raised more than $7.4 billion in its first external funding rounds, with Tencent and CATL confirmed as investors. The $50 billion valuation placed on the company during the fundraise implies investors are not betting purely on efficiency gains — they are betting on scale.

The Ulanqab build is the capital deployment. At 1GW, DeepSeek would be committing to an infrastructure footprint comparable to Z.ai’s recently activated data center in China, which also runs on domestic chips with no Nvidia hardware.

Jevons in Practice

The infrastructure pivot is textbook Jevons paradox: when efficiency improves, total consumption rises because more use cases become economically viable. V4 Flash’s cost structure opened markets that frontier-price APIs could not reach. That demand creates the case for owning the compute layer, not just renting it.

DeepSeek is also building its own inference chip to reduce Nvidia dependency — a parallel track that its Chinese peers (Alibaba, Baidu, ByteDance, Huawei) have all pursued at scale. Owning the chip and the data center is the full vertical stack.

The Broader Picture

China now holds all 10 spots on the open-weights intelligence index. DeepSeek V4-Pro, V4 Flash, and the incoming V4 variants sit firmly in the frontier coding tier. The company that shocked Western AI labs with a $6 million training story in 2025 is now deploying $7.4 billion in capital in 2026.

The efficiency narrative was never the endpoint. It was the wedge.

First infrastructure online: late 2027.