Expired50% confidenceFactExact time
When AI clusters scale to tens of thousands of GPUs, overall efficiency can plummet to below 40%, with more than half of computing power consumed in data transmission and waiting.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
7/2/2026
First Seen
This claim has expired. It is kept for the record and no longer describes the current state.
Valid until: 9/30/2026(expired)
Sources
Related Claims
UnverifiedWhen scaling to tens of thousands of GPUs, overall compute efficiency can plummet below 40%, with more than half of compute power consumed by data transfer and waiting — a phenomenon known as the 'communication wall.'80% similarUnverifiedThe marginal cost of AI models decreases as scale expands; the more users there are, the lower the average cost per inference after optimization through batching and GPU utilization.79% similarUnverifiedIn a complete AI application system, GPU inference typically accounts for only 30%–40% of total compute, with the remaining 60%–70% of the compute load borne by the CPU.76% similarUnverifiedWhen token usage exceeds 60%–70%, the AI's output quality degrades noticeably and its reasoning ability declines significantly.74% similarUnverifiedThe main components of AI inference costs are depreciation and electricity for high-end GPUs such as the A100/H100, which are capital-intensive rigid expenditures that are difficult to compress quickly as revenue declines.73% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/51578API
curl https://kongchang.com/api/v1/knowledge/claims/51578MCP
get_claim(id=51578)