DeepSeek states GPU in theoretical billing revenue per hour 38 USD

AT A GLANCE

DeepSeek Disclosed an inference-efficiency calculation, but per GPU hour 38 USD is only theoretical billing revenue and does not represent actual revenue or profit.

Article

DeepSeek Disclosure states GPU can generate per hour 38 USD in billing revenue,AI inference profitability draws attention

Dongcha Beating AI Flash,DeepSeek recently open-sourced the DeepInfer inference framework and disclosed that its cluster can process per day 155000000000000 Token. Based on DeepSeek API off-peak pricing, this translates to approximately GPU per 38 USD in theoretical billing revenue. Assuming uninterrupted operation throughout the year, annualized billing revenue exceeds 330000 USD。

DeepSeek improves inference efficiency through technologies such as expert parallelism, speculative decoding, asynchronous scheduling, and KV Cache optimization, but the above figures are theoretical billing revenue and do not equal actual revenue or profit. Actual returns also depend on GPU utilization, the proportion of paid requests, hardware and electricity costs, and service latency.AI The industry is facing simultaneous demand growth and profitability pressure. On the one hand, demand for small and medium-sized enterprise AI applications and Agent deployments continues to rise; on the other hand, investments in model training, inference services, and infrastructure continue to erode some companies' profits. According to reports, CoreWeave's second-quarter revenue increased by approximately 112%year on year to 2575000000 USD, but it still recorded a net loss of 626000000 USD.

Analysis indicates that improved inference efficiency is expected to expand the profit margin of AI services, but efficiently producing Token does not necessarily mean commercial profitability. As model call volume grows, companies still need to resolve the balance among compute costs, service pricing, and customers' willingness to pay.

Original link https://m.theblockbeats.info/flash/373499

Key points

DeepSeek Recently open-sourced the DeepInfer inference framework.
The materials state that DeepSeek disclosed that its cluster can process per day 155000000000000 Token。
Based on DeepSeek API off-peak pricing, an average of GPU per hour corresponds to approximately 38 USD in theoretical billing revenue.
Assuming uninterrupted operation throughout the year, annualized theoretical billing revenue exceeds 330000 USD。
DeepSeek Uses technologies such as expert parallelism, speculative decoding, asynchronous scheduling, and KV Cache optimization to improve inference efficiency.
The materials explicitly state that theoretical billing revenue does not equal actual revenue or profit.
Actual returns also depend on GPU utilization, the proportion of paid requests, hardware and electricity costs, and service latency.
According to reports, CoreWeave's second-quarter revenue increased by approximately 112%year on year to 2575000000 USD, but it still recorded a net loss of 626000000 USD。

AI-assisted interpretation

The following is analysis, separate from reported facts. Verify important claims independently.

These figures convert the cluster's capacity to process Token using API off-peak pricing into a theoretical billing amount. They show potential billing capacity, not money the company has already received, much less profit after deducting GPU, electricity, and operating costs.

Why it matters to readers

If inference efficiency improves and GPU is fully utilized with sufficient customers willing to pay,AI services' revenue potential per unit of compute may improve; however, the materials also note that efficiently producing Token does not necessarily result in commercial profitability.

Beginners should distinguish between “theoretical billing revenue,” “actual revenue,” and “profit”: the former is an estimate calculated from pricing and processing capacity, while the latter are also affected by actual orders, utilization, and costs.

Risks and unknowns

The materials do not provide DeepSeek original disclosures or technical materials, so DeepInfer, daily 155000000000000 Token, and the relevant calculation methodology cannot be independently verified.
Actual GPU utilization, the proportion of paid requests, hardware costs, electricity costs, and operating costs were not disclosed.
Approximately 38 USD and exceeding 330000 USD are both theoretical billing revenue and cannot be used to confirm actual revenue or profit.
The CoreWeave data are presented as “according to reports”; the materials do not provide its original financial filings.

DeepSeek states GPU in theoretical billing revenue per hour 38 USD

Article

AI-assisted interpretation

Why it matters to readers

Risks and unknowns

Related Developments

Related concepts

Keep following

Security Center