The new wave evolves fiercely, while the old wave cuts through thorns.
Author: Key Frame
Image source: Generated by Boundless AI
The AI arena is quietly witnessing a profound shift in technological power.
The transformation triggered by DeepSeek has yet to calm down, as the competition among large models enters the "post-violent computing era," where the importance of efficiency is vividly highlighted, and the power dynamics in AI are also facing reconstruction, with OpenAI's dominance being increasingly challenged.
The new wave evolves fiercely, while the old wave cuts through thorns; the winner of the "changing banners on the city walls" has yet to be determined. The key to victory lies in how to gain ecological support through open-source while achieving commercial monetization through closed-source.
01. China's AI Projects Surge with Policy Support
The development of domestic AI has been quietly brewing. 2023 is seen by industry insiders as a watershed year for artificial intelligence development.
AI scientist Fei-Fei Li once said, "Historically, 2023 is expected to be remembered for its profound technological changes and public awakening."
Before this, the exploration and innovation in artificial intelligence technology had already been numerous.
In 1956, John McCarthy first proposed the concept of "Artificial Intelligence" at the Dartmouth Conference, marking the official birth of AI as a discipline.
However, by 1973, due to bottlenecks in AI research, funding for AI significantly shrank, leading to a "winter" in development.
In 1986, the revival of neural networks was sparked by "AI father" Geoffrey Hinton's introduction of the backpropagation algorithm, bringing new hope to AI development. By 2017, Google proposed a self-attention mechanism, replacing RNN/LSTM, which became the core architecture for subsequent large language models (LLMs)…
Looking back at the development of AI in China, 2023 is also the "year of the beginning of the domestic AI era."
According to Tianyancha, there were over 20 financing events directly related to large models in the first half of 2023, with more than 100 various large models released domestically. By July 2024, the number of generative AI large models that completed filing and went online is close to 200.
To this day, only a dozen companies have the opportunity to enter the finals. Consulting firm Frost & Sullivan pointed out that the number of competitors in the general foundational large model field has shrunk to over 20, mainly dominated by internet companies, cloud computing giants, and AI startups.
Everyone is a witness to this "war" without gunpowder. Looking back from the beginning of 2025, perhaps after experiencing the "hundred model battle" in 2024, DeepSeek will be able to throw a "thunderbolt" in the global tech industry at the start of 2025, pushing domestic AI to achieve a "critical leap" and establish a foothold.
Companies with continuous innovation capabilities are gradually dominating the market, with the application scope of artificial intelligence rapidly expanding from image and text to video and multilingual advertising generation.
Meanwhile, large models and agent technologies have entered an accelerated development phase. Whether it's optimizing user experience on the C-end or providing enterprise solutions on the B-end, agents and large models are redefining the connection between technology and society.
Currently, there are three forces in the finals: first, the internet giants and cloud service providers represented by Alibaba and ByteDance, which are involved in large models; second, the national AI team represented by iFlytek, which provides both solutions and hardware products through G/B/C linkage; third, AI startups like Zhipu and DeepSeek, with a few still insisting on foundational model innovation.
The situation of the upstream and downstream of the industry chain is polarized, and the development paths of model vendors are diverging. Even the "AI six little tigers" face road differentiation. For example, Baichuan Intelligent has shifted to large models in the medical industry; Lingyi Wanshi has entrusted the training of super large models to Alibaba; and Yuezhi Anmian and MiniMax focus on C-end applications and products.
Industry insiders generally believe that compared to the upstream and downstream of the industry chain, model vendors in the midstream are facing profitability challenges. By 2025, the number of players in the large model finals who can still innovate at the foundational model level will further decrease.
02. From "Burning Money Faith" to "Efficiency Revolution"
If "cost, AI Agent, multimodal" are the three keywords of the current AI industry, representing the evolution direction of large models in 2024, then they may also signify key nodes for large models to land in the industry.
First, cost is undoubtedly a key determinant of a company's survival. The enormous demand for computing resources to train and deploy large-scale AI models cannot be ignored, which forces companies to bear high computing and operational costs.
DeepSeek-R1 has precisely addressed the pain points of companies in efficiency and cost control, achieving performance comparable to or even surpassing leading models with relatively low computational investment.
The traditional AI development model often relies on the logic of "scale supremacy," pursuing ultra-large-scale models and ultra-large-scale computing clusters. The lightweight model and open-source strategy of DeepSeek R1 have lowered the barriers to AI application, promoting the popularization of mid-range computing facilities and distributed data centers.
Upstream in the industry chain, NVIDIA is beginning to face pressure for structural adjustment in demand due to the emergence of DeepSeek.
ASIC chip manufacturers are welcoming new development opportunities. Since ASIC chips can provide hardware acceleration for specific AI applications, they have significant advantages in energy efficiency and cost control, making them more suitable for the trend of distributed computing development.
For the computing service side, regional data centers, with their low latency and proximity to application scenarios, are starting to undertake latency-sensitive application demands such as intelligent quality inspection in manufacturing and financial risk control.
Cloud computing giants like AWS and Alibaba Cloud are adjusting their construction strategies for some large data centers, increasing investment in edge computing and distributed computing layouts.
On the application side, the decline in computing costs will drive the accelerated penetration of artificial intelligence in industries such as manufacturing, finance, and healthcare.
On the code hosting platform GitHub, a large number of integrated application cases based on the DeepSeek model have emerged (awesome deepseek integration), forming a positive cycle of "demand driving supply," achieving bidirectional empowerment of "computing power + industry."
Artificial intelligence technology will accelerate its penetration into various industries, becoming an important engine for driving industrial upgrades and economic development.
However, it is worth noting that the technological breakthrough of DeepSeek R1, while lowering the barriers to AI application, may also trigger the "Jevons Paradox."
The Jevons Paradox, proposed by 19th-century economist William Stanley Jevons, found that as the efficiency of coal use increased, the total consumption of coal actually rose. This paradox reveals a profound economic law: improvements in efficiency do not necessarily lead to a reduction in resource consumption; instead, they may stimulate demand growth due to lower costs and expanded application scope, ultimately leading to an increase in total resource consumption.
Microsoft CEO Satya Nadella cited the Jevons Paradox to explain the potential impact of DeepSeek R1, which is quite insightful.
Nadella believes that more affordable and accessible AI technology will lead to a surge in demand through faster adoption and broader application. As the barriers to AI technology lower, fields that previously could not apply AI due to cost constraints, such as small and medium-sized enterprises and edge computing scenarios, will see a surge in new application demands, resulting in an exponential increase in computing power utilization density.
The explosion of emerging application scenarios will also accelerate the fragmentation of computing power demand. Cutting-edge fields like intelligent driving and embodied robotics have enormous real-time computing demands, far exceeding the speed of optimization provided by DeepSeek technology. Even if single-task efficiency improves several times, the concurrent demand from millions of intelligent terminals will still create a massive computing power consumption black hole.
03. The Synergy of "Open Source" and "Closed Source"
With the explosive popularity of the open-source large model DeepSeek, keywords like "open source" and "free" are frequently appearing.
If, before DeepSeek, domestic large model companies had many disagreements regarding the paths of "open source" and "closed source," now the calls for "open source," "open ecology," and expanding social circles seem to have become mainstream.
Under the impact of DeepSeek as a catfish, domestic large model companies are showing a more "open" posture, hoping to accelerate the establishment of their own developer and application ecosystems.
The key differences between open-source and closed-source models can be observed from three dimensions: foundational conditions, technical aspects, and commercialization.
From the foundational conditions, open-source models rely on public datasets and community-contributed data as their data sources, supported by distributed, developer-owned GPU clusters for computing power, providing equal access opportunities for developers, researchers, and enterprises, thus promoting technological innovation and sharing.
Closed-source models, on the other hand, are developed by companies or teams, using proprietary data such as user behavior logs, private databases, and cleaned public data as their data sources, with users only able to use these models through interfaces or platforms provided by the company.
From the perspective of profit scenarios, open-source models do not directly generate revenue, but they typically achieve profitability through additional services (such as cloud computing, technical support, training, customized development, etc.). Companies can provide value-added services commercially, forming a sustainable income source based on open-source models.
The commercialization path of closed-source models is relatively straightforward, with enterprises achieving profitability through licensing, subscription services, and platform fees. Closed-source models can bring high profits to companies because customers need to pay for usage rights and services.
Open source and closed source are not "incompatible"; in the future, it is likely that a form of interaction between open source and closed source will emerge, where open source accelerates the popularization and innovation of AI technology, while closed source ensures that technology can achieve substantial commercial development and maintain stability.
The future winners will be versatile players who can master both open-source and closed-source capabilities, gaining ecological momentum through open source while capturing value through closed source.
As Nadella said, "There will not be a winner-takes-all situation in super-large-scale AI; the open-source model will balance the closed source."
Epilogue
DeepSeek will play an important role in the current AI era, much like Android did in the mobile internet revolution.
Reconstructing the industrial ecology, triggering a chain reaction, and accelerating the development of upper-layer applications and the unification of lower-layer systems. This will mobilize the ecological forces that span software and hardware as well as upstream and downstream, prompting all parties to increase investment in the collaborative optimization and vertical integration of "models - chips - systems," further weakening the advantages of the CUDA ecosystem and creating opportunities for the development of the domestic AI industry.
Through technological innovation, DeepSeek has reduced reliance on high-end imported chips in the AI model training process, showcasing a feasible technological path for domestic enterprises and greatly enhancing their confidence in independently developing computing power chips.
The competition is not just a technical choice between open source and closed source; it also involves the discourse power, market dominance, and allocation of computing power in AI development. And this battle for AI power has already begun.
免责声明：本文章仅代表作者个人观点，不代表本平台的立场和观点。本文章仅供信息分享，不构成对任何人的任何投资建议。用户与作者之间的任何争议，与本平台无关。如网页中刊载的文章或图片涉及侵权，请提供相关的权利证明和身份证明发送邮件到support@aicoin.com，本平台相关工作人员将会进行核查。
