Unverified50% confidenceFactExact time
The cost pressures on frontier large model companies primarily stem from compute chip procurement for model training and compensation for top research talent
1
Sources
50%
Confidence
Long-term
Relevance
9/29/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedTraining frontier large models requires enormous computational investment, and there are diminishing marginal returns as model scale increases.76% similarUnverified训练2.8万亿参数模型代价高昂，便宜和开源并不等于不需要算力，算力需求从闭源巨头向分散的企业定制场景转移76% similarUnverifiedPausing large model training is costly, involving enormous compute expenditures and potentially affecting the pace of product iteration and competitive market position.74% similarUnverifiedTraditional large model training relies heavily on manually annotated data and external corpus collection, which are typically the bottlenecks constraining iteration speed.70% similarUnverified国产模型在编程能力追赶上面临高质量代码训练数据获取成本高、专业程序员RLHF标注资源稀缺、算力基础设施受出口管制影响等多重结构性挑战69% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/959053API
curl https://kongchang.com/api/v1/knowledge/claims/959053MCP
get_claim(id=959053)