DeepSeek has swiftly ascended to become the fastest application to surpass 30 million daily active users.

Like a pebble dropped into a tranquil pond, DeepSeek has rippled through the technology industry, causing widespread commotion.

From foundational chips and cloud services to upper-level terminal devices and internet applications, DeepSeek has permeated the entire technology industry chain at an astonishing rate.

DeepSeek's emergence has not only disrupted the competitive landscape in the domestic AI sector but also brought about five significant impacts on the industry through its "full-stack breakthrough":

01

Pushing Large Model Vendors to Adopt Inference Models or Even Alter Strategic Directions

The foundational model arena is characterized by "three highs": high barriers, high investments, and high risks, making it a realm reserved for a select few with top-tier technical prowess and substantial capital reserves. Competition here is fiercely intense.

This is not merely a contest of technical strength but also a comprehensive battle involving capital endurance, talent density, and ecosystem integration capabilities. Any slight lapse may result in falling behind or being eliminated.

One of DeepSeek R1's innovative aspects lies in its long-chain thinking mode, which, while not an industry first, is compelling large model vendors to reassess their strategic directions and even accelerate the research, development, and deployment of inference models.

ByteDance: Reflecting on its sluggish pace and lack of sensitivity to new opportunities

Recently, ByteDance CEO Liang Rubo acknowledged at an all-hands meeting that despite recognizing significant technological shifts after OpenAI released the long-chain thinking model o1 in September 2024, ByteDance failed to respond swiftly enough. He stated that had ByteDance paid timely attention, it could have achieved this milestone earlier. He further noted that compared to startups, ByteDance's sensitivity to new opportunities in large models is notably inadequate.

The Dark Side of the Moon: Simultaneous Launch of Multimodal Inference Model k1.5

On January 20, coinciding with the release of DeepSeek R1, The Dark Side of the Moon unveiled the Kimi k1.5 multimodal thinking model, equipped with state-of-the-art (SOTA) multimodal inference capabilities.

Recently, The Dark Side of the Moon researcher Flood Sung shared the complete thinking process behind the k1.5 model in an article titled "The Long Long Long Long Long Thinking Behind Kimi k1.5," sparking intense discussions within the AI community. He revealed that The Dark Side of the Moon had validated long chains of thought a year prior but initially focused on long texts due to high costs.

02

Accelerating Open-Source and Free Availability of Foundational Models to Drive Down Inference Costs

The evolution of AI consistently follows a similar trend: inference costs continuously decrease, and computational efficiency steadily improves, significantly reducing the computational requirements for AI capabilities of the same level.

As of September 2024, the open-source community Hugging Face hosted over one million AI models, affirming the popularity of open-source models and their revolutionary significance in democratizing technology.

Accelerating the open-source and free availability of foundational models is not only a crucial step towards technology democratization but also a vital engine driving the ongoing reduction of inference costs.

The advent of DeepSeek signifies a reconfiguration of the binary opposition between open-source and closed-source models, completely overturning the past industry perception that "open-source equals low-end."

As open-source models begin to compete on par with closed-source giants in core capabilities, the strategic choices of technology enterprises undergo a fundamental transformation.

03

Comprehensive Penetration into Top Mobile Internet Ecosystem Applications

Recently, top internet applications such as WeChat and Baidu Search have integrated DeepSeek.

Notably, the full integration of Tencent's product matrix, encompassing AI assistant Tencent Yuanbao, QQ Music, Tencent Docs, Tencent Maps, and more, marks the first time China's internet giants have collectively opened up core scenarios to third-party AI models.

This shift indicates that large model vendors are transitioning from a "self-developed model moat" strategy to an "ecosystem thinking of plug-and-play models."

04

Domestic Computing Power + Domestic Models Accelerate AI Implementation in Enterprise-Level Scenarios

The synergistic breakthrough of domestic computing power and domestic models is clearing the obstacles that have historically hindered AI implementation in enterprise-level scenarios.

Currently, domestic GPU cards like Ascend, Haiguang, Moore Threads, BiRen Technology, and Tianzhi Core have been adapted for DeepSeek.

In terms of application scenarios, numerous regions across the country have announced the application of DeepSeek in government systems.

For instance, on February 16, Shenzhen officially provided DeepSeek model application services to all districts and departments citywide, leveraging the government cloud environment to achieve an integrated AI-enabled upgrade of government applications based on DeepSeek.

Jiangxi's Ganzhou City has completed the deployment of the DeepSeek series of large models in the government environment, becoming the first prefecture-level city in the province to successfully deploy and launch this large model.

05

Accelerating Edge AI Applications to Promote Integration of Large Models with Smart Terminals

DeepSeek's emergence has not only significantly lowered the barriers to AI model training and deployment but also fostered the deep integration of large models with smart terminals, ushering in an era of universal access to edge AI with its "low cost and high efficiency" advantages.

The consumer electronics industry, as the "pioneer" of edge AI, has taken the lead in embracing this trend. Currently, mainstream domestic smartphone manufacturers such as Honor and OPPO have announced the integration of DeepSeek, embedding its capabilities into smartphones and other terminal products.

However, the potential of edge AI extends far beyond this.

In the future, with breakthroughs in embodied intelligence technology, DeepSeek will further empower emerging terminals such as robots and smart cars.

For example, domestic leading robot manufacturers like Unitree and CloudMinds have made significant strides in the flexibility and autonomous decision-making capabilities of robots. In the future, if they can integrate DeepSeek-level models, it will infuse them with an "AI brain," enabling a leapfrog upgrade from "mechanical execution" to "intelligent interaction."

Conclusion

The AI tsunami sparked by DeepSeek continues to rage.

From innovations in underlying technologies to expansions in application scenarios, from academic research breakthroughs to commercial application implementations, the AI wave since ChatGPT in 2023 has never so profoundly reshaped the landscape of various industries.

Remarkable foundational models have not only altered people's perceptions and expectations of AI but are also propelling human society into a new era brimming with infinite possibilities.

END This article is an original work of "Intelligent Evolution Theory." Welcome to follow us.