Expired50% confidenceFactExact time
Training a model of the same scale as those in the US requires 50,000 top-tier NVIDIA cards or 200,000 Huawei 950s.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
7/25/2026
First Seen
This claim has expired. It is kept for the record and no longer describes the current state.
Valid until: 10/23/2026(expired)
Sources
DeepSeek五步阶梯通往AGI：两万张卡的克制路线图
Related Claims
UnverifiedA 70-billion-parameter model corresponds to approximately 1.4 trillion training tokens under the Chinchilla optimum63% similarUnverifiedA hyperscale data center with an investment of billions of dollars may require only 50–200 people for daily operations.62% similarUnverifiedxAI has built the Colossus 2000MW-class training cluster and is training Block5 versions with 6 trillion and 12 trillion parameters.61% similarUnverifiedCOCO 数据集包含约12万张训练图像、80个类别、超过150万个目标实例，由微软研究院发布60% similarUnverifiedThe MNIST dataset contains 60,000 training images and 10,000 test images, each a 28×28 pixel grayscale image of a handwritten digit.58% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/615430API
curl https://kongchang.com/api/v1/knowledge/claims/615430MCP
get_claim(id=615430)