
Rocky|Mar 01, 2025 16:15
The AI circle exploded again on the weekend, brothers! DeepSeek's latest report: 'Make $560000 in one day with a profit margin of 545%!'
The brothers of AI Infra should cry on Monday when they go to work, and have a good weekend. The sky is about to collapse. Brothers of Silicon Valley, are you ready to work overtime? If the profit margin cannot reach the level of DeepSeek, it can only indicate that our own Infra team is incompetent! 😂
Do Deepseek still lack nodes? This profit margin is too good, I want to go mining! I'll calculate how much money can be earned on one node for you, why bother with BTC mining and digging a der? I'm planning to fly to Hangzhou next week, it's too explosive. No wonder my friends in the mining circle have been running to Hangzhou recently. What an opportunity!
Let me calculate for you: DeepSeek's current products, DeepSeek V3 and R1 inference services, currently occupy a peak total of 278 nodes, with an average of 226.75 nodes occupied (8 H800 GPUs per node).
Assuming a GPU rental cost of $2 per hour, the total cost is $87072 per day; If all tokens are priced according to DeepSeek R1, theoretically the total daily revenue would be $562027, with a cost profit margin of 545%.
So how did they make so much money? The strategy is like China's new energy vehicle charging station, a peak shaving and valley filling strategy. Due to the high service load during the day and low service load at night, DeepSeek has implemented a mechanism to deploy inference services on all nodes during high load periods. When the load is low at night, reduce the number of inference nodes for research and training, in order to ensure efficiency and maximize profits.
Let me calculate for you the cost and expense comparison of Openai's latest release, GPT 4.5, a few days ago. On February 28th, OpenAI officially released the research preview version of its latest model GPT-4.5, which is claimed to be the "highest emotional intelligence" universal big language model. However, its API call price of $75 per 1 million tokens input has skyrocketed 30 times compared to GPT-4o's $2.5, and compared to DeepSeek's normal price, GPT-4.5's input price has reached an astonishing 280 times. Many people commented: Openai is crazy about stealing money, only fools would use it at such a high price. With such high costs, can Jack Ma recoup his investment in the Year of the Monkey? Really, without comparison, there is no harm!
And last week during DeepSeek Open Source Week, losing one iteration update a day was terrifying. Looking forward to the release of R2, it will blow up the AI street in the United States again! On Monday, American AI stocks trembled! 🥲
Share To
Timeline
HotFlash
APP
X
Telegram
CopyLink