DeepSeek's "high profits" indicate that through extreme infrastructure optimization, very high computing power utilization and performance can also be achieved. However, there are still many disagreements about whether DeepSeek is a nuclear bomb for computing power. Well-known investor Duan Yongping agrees with NVIDIA CEO Jensen Huang's earlier viewpoint, stating that DeepSeek will stimulate the market's pursuit of more efficient AI models and believes that the demand for computing power will continue to grow. However, some foreign technology bloggers argue that DeepSeek has already 'knocked down' NVIDIA, and based on DeepSeek's current extremely high utilization of computing power, it can easily meet the global AI demand.
During the past week's Open Source Week, DeepSeek's "Five Days of Five Explosions" amazed the market. Just when everyone thought this feast was about to come to an end, DeepSeek produced an even more shocking "surprise"—a profit margin of 545%, with the theoretical daily profit of the V3/R1 inference system reaching 3.46 million yuan.
While the market marvels at this ultra-high "super profit," it is also more concerned with several questions: How to interpret this 545% profit margin? Is it a nuclear bomb for computing power? What does this mean for the Cloud Industry Chain? What does it mean for big model peers? What does it mean for the ecosystem? Several technology bloggers have the following main viewpoints:
The profit margin of 545% is currently still theoretical earnings, with DeepSeek's R1 model profit margin at about 85%. If priced according to V3, the profit margin would drop to about 70%. Even so, this number is still very impressive.
For the computing power industry chain, DeepSeek's case proves that even under relatively limited hardware conditions (using H800), extreme infrastructure optimization can achieve very high computing power utilization and performance.
However, there is still considerable disagreement about whether DeepSeek's innovation will reduce the demand for computing power. Prominent investor Duan Yongping agrees with NVIDIA's CEO Jensen Huang's view that the demand for computing power will continue to grow. However, some foreign technology bloggers have stated that DeepSeek has already "knocked down" NVIDIA, and that given DeepSeek's current ultra-high utilization of computing power, the Global demand for AI isn't that high.
Additionally, DeepSeek's case demonstrates that the similarities between AI Cloud Computing and traditional Cloud Computing have become more apparent. AI Cloud Computing will also face the challenges of "idle rates during low periods" and "stability during peak periods."
DeepSeek's open source and technology disclosure have set a new benchmark for the entire industry. Peers may face greater competitive pressure, and a new round of price wars is on the way.
For the industry ecosystem, DeepSeek will attract the industry to build both B2B and B2C businesses based on open-source technology and output, forming a complete upstream and downstream of the Industry Chain.
1. How to interpret this profit margin?
First, it is important to clarify that the 545% profit margin published by DeepSeek is based on a "theoretical" calculation under specific conditions, that is, assuming all tokens are priced according to the R1 model, and does not consider factors such as the lower pricing of V3, the proportion of free services, and night discounts. In reality, according to DeepSeek's official statement, their actual profit margin is far less exaggerated.
According to the interpretation of technology blogger 180K, the profit margin of DeepSeek's R1 model is about 85%, and if priced according to V3, the profit margin would drop to around 70%. Even so, this number is still very impressive.
180K indicates that this point can be understood more deeply by comparing it to the profit margin of Anthropic. According to TD Cowen's breakdown, Anthropic's profit margin for 2024 is expected to be 61%. If according to DeepSeek's standards, and considering the profit margin of AWS's Cloud Computing (assumed to be 25%-40%), Anthropic's profit margin could reach 74%. In extreme cases, if AWS's profit margin is assumed to be 50%, Anthropic's profit margin could even reach 85%, comparable to DeepSeek's R1 model.
This indicates that although OpenAI and Anthropic may not be as extreme in cost control as DeepSeek, they can still achieve similar high profit margins through higher pricing and more generous customers (at least for now). It should be noted that OpenAI is often reported as "losing money" because, during financing, investors typically focus on financial accounting profits and losses, rather than the theoretical costs from the perspective of large model leasing. Operational expenses such as model training costs, data licensing fees, personnel, and advertising are usually included.
2. Is it a nuclear bomb for computing power?
The case of DeepSeek proves that even with relatively limited hardware conditions (using H800), through extreme infrastructure optimization, it is possible to achieve high computing power utilization and performance, which has a huge impact on the entire computing Industry Chain.
Firstly, Technology blogger 180K believes that the importance of 'Effective Computing Power' will become prominent. The Industry will pay more attention to 'Effective Computing Power' (Computing Power x Computing Power Utilization Rate), rather than just simply stacking computing power.
Additionally, the upper limit of domestic chips is expected to be raised. If the H800 can achieve such results, then through infra optimization, the performance limit of domestic chips may further improve.
Moreover, Technology blogger Information Equality believes that the 'Jevons Paradox' continues to be effective. The improvement in computing power efficiency will not reduce the demand for computing power, but instead will stimulate the emergence of more application scenarios, driving the continuous growth of computing power demand. Just as Barclays predicted in June last year, by 2026, the industry's capital expenditures will be sufficient to support '12,000+ applications at the ChatGPT level.'
However, in the short term, the logic of computing power demand may be questioned. Some companies, especially CIOs or CFOs of overseas large enterprises, may face pressure from investors and bosses to explain why their ROI is much lower than DeepSeek's.
Famous investor Duan Yongping also stated on Xueqiu that the experience of DeepSeek indeed proves that a lower computing power in the model pre-training stage can achieve relatively good training results. Moreover, he agrees with Jensen Huang's statement that DeepSeek's innovation will not decrease the demand for computing power.
Previously, Jensen Huang stated in an interview in February that the market's understanding of DeepSeek was completely wrong. He said that the emergence of R1 does not mean that the market no longer needs computing power resources, but rather stimulates the market's pursuit of more efficient AI models, thereby promoting the development of the entire industry.
However, foreign Technology blogger Zephyr believes that DeepSeek has already 'knocked out' NVIDIA. Moreover, based on DeepSeek's currently high utilization of computing power, it is more than enough to meet the global AI demand.
DeepSeek has already 'knocked out' NVIDIA.
The reason I say this is that DeepSeek is currently processing 600 billion tokens daily on 300 H800 nodes (a total of 2400 H800) and outputting 150 billion tokens.
If the computing power expands 100 times (i.e., 0.24 million H800s), it can process 60 trillion tokens per day and output 15 trillion tokens.
However, the global demand for AI is not this high.
3. What does it mean for the Cloud Industry Chain?
DeepSeek's success cases make the similarities between AI Cloud Computing and traditional Cloud Computing more evident. AI Cloud Computing will also face challenges such as 'idle rate during low peaks' and 'stability during peak periods.'
Technology blogger with 180K followers believes that the scale effect of Cloud Computing will become more pronounced. DeepSeek's practice shows that large-scale clusters and high concurrency utilization can significantly reduce costs. The positive externality of the number of users is more apparent, that is, the more users there are, the stronger the ability to smooth fluctuations and the lower the demand for computing power redundancy.
The competitive advantage of Cloud vendors may change. Cloud vendors with their own businesses (like Alibaba, Tencent, Apple, etc.) may have more cost advantages than those without their own businesses because they can use the inference clusters as a base for all businesses, achieving greater scale effects.
Moreover, there is room for profit margin improvement in Cloud Computing. DeepSeek's case indicates that in the AI era, through extreme infra optimization, there is potential for further improvement in the profit margin of Cloud Computing.
In addition, the appeal of private cloud deployment may decrease. Extremely sparse MoE models may not be suitable for individual or 'half-hearted' business deployments, as the costs of small-scale GPU deployments may far exceed those of large companies. This may lead more enterprises to choose public cloud or hybrid cloud models.
Regular Cloud Computing/AI applications need to reserve more space for high-intensity user concurrency. Users have a higher tolerance for DeepSeek's 'server busy' message, but not for other applications. This may result in further declines in the profit margins of regular Cloud Computing/AI applications.
4. What does this mean for large model peers?
DeepSeek's open-source and technology disclosure have set a new benchmark for the entire Industry.
The information equity of Technology bloggers suggests that DeepSeek's case shows that the 'bottom line' of inference costs has been significantly lowered, far below previous expectations. Moreover, a new round of price wars may erupt, and peers will face greater pressure to cut prices to remain competitive.
Additionally, DeepSeek provides all inference teams with clear optimization paths and goals, increasing the pressure in the future.
In this situation, OpenAI's High Stock Price subscription model will also face challenges, as the expensive $200 monthly subscription fee is somewhat awkward.
5. What does this mean for the ecosystem?
DeepSeek's strategy focuses on foundational models and cutting-edge innovations, attracting the industry to build both B2B and B2C businesses based on open-source technology and output, forming a complete industry chain.
Technology blogger Geek Park states that the profit margins for ecological partners are increasing. Cloud platforms and upstream/downstream businesses can theoretically achieve high revenue and profit margins by deploying DeepSeek's services.
Looking ahead to the subsequent ecosystem, the differentiation in model architecture may become a key competitive factor. Because the architecture of DeepSeek V3/R1 differs significantly from mainstream models, this requires suppliers to adapt, resulting in higher development difficulty.
Furthermore, DeepSeek's open-source initiative lowers the difficulty for the community to replicate its inference system, which is beneficial for the prosperity of the ecosystem.
Technology blogger 180K indicates that the entire industry may start to focus on Infra. To some extent, the importance of Infra is increasing, and valuations can also rise.
In summary, DeepSeek's extremely high profit margins are not only a numerical miracle, but also a profound insight into the entire AI industry. It reveals the enormous potential for infra optimization, drives the transformation of computing power, cloud services, large models, and the ecosystem, and heralds the arrival of a more efficient, low-cost, and competitively intense AI era.
Editor/new