With its cute whale logo, DeepSeek’s recent release could have constituted another Chatgpt copy. What makes it so newsworthy – and what sends competitor stocks to the tail – creates very little. It effectively puts the monkey wrench into the US’s investment concepts required to train high-function large language models (LLMs).
DeepSeek allegedly spent just $6 million to train its AI models. Twined in the reports that Openai spent $8-100 million on chatting GPT-4, or they set aside $1 billion for the GPT-5. DeepSeek said the investment level was questioned and caused large players like Nvidia to drop the value of stocks by $600 billion in a day – TSMC and Microsoft were happy with AI’s long-term financial viability. If there is significantly less than the previous assumption that can be trained on AI models, what is the reservation for the overall AI spending?
Although DeepSeek’s interruption led to important discussions, some key points seemed to be lost in the shuffle. But the greater focus of the news is on the innovation costs and possible economic impacts of AI. Here are three important insights from this message:
1. DeepSeek’s $6 million price misleading
Companies need to master the total cost of ownership (TCO) of their infrastructure. Although DeepSeek’s $6 million price is sold out, that may just be the cost of its pre-training run rather than the entire investment. The total cost (not only running, but also building training in-depth searches) can be much higher. Semi-analysis by industry analysts firms shows that the company behind DeepSeek spent $1.6 billion on hardware to make its LLM a reality. Therefore, the possible costs are in the middle.
Whether it’s true costs, the emergence of DeepSeek will focus on potentially transformative cost-effective innovations. Innovation is often stimulated by limitations, and DeepSeek’s success highlights the way engineering teams are innovative when they optimize their resources when facing real-world constraints.
2. Inference is what makes AI valuable, not training
It is important to pay attention to how much AI model training costs, but training represents a small part of the total cost of building and running an AI model. reasoning – The way AI changes people’s work, interaction and life is where AI is truly valuable.
This raises the Jewen School paradox, an economic theory that shows that the overall consumption of the resource may actually increase as technology advances make the use of resources more efficient. In other words, as training costs fall, reasoning and agency consumption will increase, and overall spending will follow suit.
In fact, AI efficiency may lead to an increase in AI spending, which should boost all ships, not just Chinese ships. Assuming they ride the efficiency wave, companies like Openai and Nvidia will also benefit.
3. Still, unit economics is the most important
Improving AI efficiency is not only about reducing costs; it is also about optimizing unit economics. Motley fool predicts that this year will be a year for AI efficiency. If appropriate, companies should pay attention to reducing their AI training costs and AI consumption costs.
Organizations that build or use AI need to understand their unit economics, rather than picking out impressive figures, such as DeepSeek’s $6 million training cost. True efficiency requires allocating all costs, tracking AI-driven needs, and maintaining constant labeling in cost to value.
Cloud Unit Economics (CUE) is related to measuring and maximizing cloud-driven profits. CUE compares your cloud cost to revenue and demand metrics, revealing the efficiency of your cloud spending, how it changes over time, and the best way to be more efficient (if you have the right platform).
Given that it is inherently more expensive than traditional cloud services sold by the public, understanding the tips are more practical in artificial intelligence. The company that builds an agent application can calculate the cost per transaction (such as cost per bill, cost per delivery, cost per transaction, etc.) and use it to evaluate the return on investment for specific AI-driven services, products, and features. Rate. As AI spending increases, companies will be forced to do so. No company can always pay endless dollars in experimental innovation. Ultimately, it must be commercial.
Improve efficiency
What a meaningful $6 million figure, DeepSeek may offer a watershed to wake up the tech industry and inevitably need to be efficient. Hopefully this will be open to vulnerability training, inference and proxy applications to unlock the true potential of AI and ROI.
