The waves generated by DeepSeek continue to ripple through the global technology landscape, creating unprecedented impacts not only in China but also reaching the United States, Germany, and beyondDuring the Spring Festival, DeepSeek achieved the astonishing feat of topping the iOS app charts in over 100 countries and regions worldwide, surpassing even the popular ChatGPT in download numbersJust 20 days after its launch, the daily active user count skyrocketed past 20 million, and within 72 hours, more than 16 domestic chip companies had completed adaptations to leverage this new technologyMajor cloud computing platforms were clamoring to incorporate support, with all three leading domestic telecom operators connecting to its services, reinforcing the simultaneous launch of the National Supercomputing InternetNumerous smartphone and automobile brands eagerly announced their integration with DeepSeek, marking a seismic shift in the tech ecosystem.

DeepSeek is not just another AI application; it represents the fastest-growing AI solution in the world, igniting a frenzy across industries such as chip production, cloud computing, telecommunications, application development, endpoints, and even the capital marketsBehind this excitement lies not only a triumph of technology but a collective breakthrough for the domestic AI commercial ecosystem.

In the past decade, Chinese AI chips have often played second fiddle to NVIDIA's dominant presence in the marketThe head of Huawei’s Ascend team once remarked that "a chip without an ecosystem is like a gun without bullets." The advent of DeepSeek has pulled the trigger on that gun, initiating a new era where domestic chip manufacturers can finally find their footing.

DeepSeek combines cutting-edge multi-head latent attention (MLA) and mixture of experts (MOE) technologies with reinforcement learning (RL) inference algorithms to optimize its processing power (PTX). This infusion of innovative technologies has not only bolstered AI model performance but also significantly reduced costs associated with training and inference

Advertisements

The economical yet highly capable nature of DeepSeek has offered domestic AI chip manufacturers a unique opportunity for technical validation and commercialization.

Within a mere three days of releasing the open-source protocol, at least 16 domestic AI chip companies announced that they had successfully adapted their capabilities to integrate with the DeepSeek modelNotable names include Huawei Ascend, Muxi, TianShu Smart Chip, Moore Threads, Haiguang Information, Birun Technology, TaiChi YuanQi, and moreNotably, Huawei Ascend and Silicon-Based Flow jointly introduced a DeepSeek inference service that set a new benchmark for performance comparison between domestic chips and international top-tier GPUsSuir Technology’s smart computing centers have deployed thousands of self-developed accelerated cards across cities such as Qingyang and Wuxi, achieving a 40% reduction in model inference costsWithin hours, Birun Technology managed to adapt to the entire series of DeepSeek models, marking a new record for responsiveness among domestic chip enterprises.

This race for compatibility illustrates a culmination of technological advancements by domestic chip manufacturers—efforts that have been over a decade in the makingFor example, Moore Threads’ advanced computing cluster and TianShu Smart Chip’s FP8 mixed precision architecture have found novel commercialization pathways through their collaboration with DeepSeekSuch partnerships have accelerated the optimization of deep learning frameworks and the adaptation of distributed training, facilitating the creation of a closed-loop ecosystem combining “domestic computing power + domestic large models.” Estimates from CITIC Securities suggest that by 2025, the market share of domestic AI chips may soar from less than 15% to an impressive 35%.

Investors are responding with vigor; as per CITIC Securities, domestic AI chip shipments are projected to skyrocket by 180% month-on-month by January 2025, while the second-hand price of NVIDIA’s A100 has plummeted by 40%. Li Zhenya, the president of the Huawei Ascend product line, candidly noted, “Clients used to ask if our products could be compatible with CUDA; now the question has shifted to when DeepSeek will be compatible.”

The rise of DeepSeek has also ignited a wave of enthusiasm among cloud service providers and telecommunications operators

Advertisements

This new technology has democratized access to AI, creating a previously unseen level of engagement and usage among diverse sectors and ordinary consumers alikeDeepSeek's economics are striking; for instance, the API pricing for DeepSeek-R1 is merely around 3% of what one would pay for OpenAI's flagship modelThis stark difference is poised to generate a monumental surge in the demand for intelligent upgrades across industriesFurther amplifying its reach, DeepSeek is entirely open-source, allowing developers to freely integrate and adapt the platform for targeted AI applications tailored to specific industries.

The advantages of DeepSeek—namely its low cost, high performance, and open-source nature—are luring numerous developers and enterprises into the fray, while simultaneously providing cloud computing companies and operators with new commercialization opportunitiesThe adage “whoever holds the model holds the future” has never rung more true.

In a surprising turn, international giants like Microsoft Azure and Amazon AWS have quietly integrated DeepSeek, with NVIDIA’s NIM platform optimizing CUDA compatibilityAs discussions of "technology decoupling" loom in the West, commercial interests have prompted industry leaders to vote with their feet and embrace this new paradigm.

Meanwhile, domestic cloud providers are not standing idly byAs of now, at least 10 leading Chinese cloud service giants—including Huawei Cloud, Tianyi Cloud, Tencent Cloud, Alibaba Cloud, and Baidu Intelligent Cloud—have officially announced their support for DeepSeekIn addition, 12 intelligent computing companies have joined the movement, such as Unisoc and Beijing Supercomputing.

With all three major telecom operators entering the fray, the competition in computing services has escalated dramaticallyChina Telecom's “Xirang” platform has achieved full-stack domestic inference, while Mobile Cloud has unveiled a "plug-and-play" customized solution for DeepSeek

Advertisements

Unicom Cloud has even allowed clients the flexibility to toggle between private and public deploymentsAccording to consulting firm Omdia, AI computing service prices in China have plummeted by 27% within three months, while utilization rates have surged by 1.8 times.

As the capital markets respond enthusiastically to these developments, companies like Qingyun Technology and Parallel Technology have seen their stock prices soar substantially, even as the three major telecom operators recorded temporary spikes in their valuations.

This transformative revolution is rewriting the rules of engagementSimilar to how cloud computing disrupted traditional IT infrastructure, the current democratisation of computing power, primarily driven by cloud service providers, has lowered the barriers for small and medium enterprises to adopt AI from a daunting mountain climb to a simple subway commuteAnalysts believe that cloud companies act as amplifiers of DeepSeek’s capabilities, creating expanded opportunities for low-barrier deployment of “killer” applications.

The eruption of AI applications is witnessing a significant moment with DeepSeek at the helmRecent research from CITIC JianTou Securities highlights that just as the growth of Android and iOS applications has shaped the tech landscape, it’s crucial to focus not only on the number of users actively engaging with the large models but also on the community of developers, particularly those representing small and mid-sized enterprises.

Indicators of community engagement, such as GitHub Stars and Forks, serve as essential metrics reflecting a project’s popularity and user affinityDeepSeek’s V3 and R1 models, having been available for under two months, have already amassed a substantial number of Stars and Forks, rivalling earlier models like Llama and significantly outpacing the newly released Llama3. This clearly indicates a high level of endorsement from developers concerning DeepSeek’s open-source models, with R1 alone accumulating around 57,000 “likes” from fellow developers on GitHub

Community-driven initiatives have emerged, demonstrating that smaller models, even with around 8 billion parameters, can run on personal notebooks, thereby significantly lowering the barrier for localized deployment and heralding a wide array of developments in AI applications.

Moreover, DeepSeek is driving a powerful force for intelligent transformation across multiple sectorsAs discussions about DeepSeek gain momentum, it’s ushering in tremendous opportunities for AI adoption across various fields including manufacturing, healthcare, education, transport, and agricultureFor instance, organizations like Weiyun Group have integrated the DeepSeek-R1 model to bolster smart writing assistance, while NetEase Youdao has embraced DeepSeek-R1 to elevate the quality of online educationSimilarly, DingTalk’s AI assistant witnessed a remarkable surge in user-created intelligent assistants— surpassing one million within just three days of integrating DeepSeek.

These examples are merely the tip of the icebergWith projections indicating that 2025 will be a breakthrough year for AI applications in China, we can expect auxiliary diagnostic systems in healthcare to react faster than physicians flipping through medical histories, financial risk control models capable of making decisions in milliseconds, and even distinctions within live-streamed e-commerce environments where virtual hosts can engage continuously for hours without stuttering.

What’s equally fascinating is the swift adoption of DeepSeek among smartphone and automobile manufacturersWithin days of its introduction, several mobile brands—including Huawei, Honor, OPPO, and Nubia—have officially announced their integration with DeepSeek-R1. Reports suggest that Vivo will likely join this trendThe automotive industry isn’t lagging behind either; both Geely and GAC have announced their deep integration with DeepSeek, effectively driving advancements in smart vehicle technologies.

In reality, many smartphone and automotive brands have already launched their own intelligent assistants—Huawei's Xiaoyi and Honor's YOYO being prime examples

Nevertheless, the eagerness to integrate an external technology like DeepSeek suggests a desire to leverage its popularity as a selling point to attract consumers.

Moreover, DeepSeek's cost-effective models are anticipated to catalyze the commercialization of AI-driven gadgets, such as AI glasses and AI toys, with analysts predicting that 2025 will likely herald a year of explosive growth within the AI hardware marketMany experts believe that glasses represent an excellent platform for AI innovation, and we might soon witness a "battle of the glasses" unfold in the coming yearIn contrast to current AI glasses utilizing larger models with higher resource consumption, DeepSeek’s emergence is likely to spark a renewed interest in AI glasses.

In essence, DeepSeek acts like a universal key, unlocking the door to intelligent upgrades across diverse industries, rejuvenating established sectors under the transformative energy of AI.

However, the dramatic rise of DeepSeek has left competitors grappling with uncertaintyThe ascendance of this application has significantly impacted the global competition for large models, prompting both domestic and international players to deploy strategies focused on tech upgrades, ecosystem collaborations, and price adjustments.

OpenAI appears to have taken the brunt of this disruptionOn February 1, they hastily launched a new inference model named o3-mini, touting enhanced reasoning and conversational capabilities, along with functionalities for integrated web searchingAlthough the new o3-mini is offered for free, its pricing remains higher than DeepSeek's competitive offeringsBy February 6, OpenAI expanded ChatGPT's search capabilities to a broader user base without the need for registration, thereby lowering barriers for engagement.

Sam Altman, the CEO of OpenAI, acknowledged how DeepSeek’s rise has eroded their technological lead, leading to a future reorientation of their open-source strategies

Kevin Well, the chief product officer, noted that the company is contemplating making older, non-cutting-edge models like GPT-2 and GPT-3 open-source.

Meanwhile, in the shadow of DeepSeek's success, both major domestic internet companies and the so-called "Six Little Dragons of AI" seem somewhat overshadowed.

Alibaba Tongyi has reacted remarkably quicklyOn January 29, Alibaba Cloud upgraded its flagship Tongyi Qianwen model to Qwen2.5-Max, with over 200 trillion tokens in pretrained data, outperforming numerous current mainstream models in various open assessments, including DeepSeek-V3. Additionally, recent reports from sources such as The Information indicate that Apple’s iOS may integrate with Alibaba's Tongyi series, further boosting the model's visibilityAs iOS 18.4 is set to launch in April, it may feature local language versions of Apple Intelligence, thereby enhancing user experiences with more effective voice assistance, intelligent recommendations, and personalized services.

On the other hand, the company facing the most awkward position among all domestic large model providers seems to be BaiduOnce a trailblazer in embracing the "All in AI" approach over a decade ago, Baidu finds itself losing ground in terms of technology, product experience, and user perceptions in the current market landscape.

Baidu had aggressively positioned itself, once voicing opinions aligned with scale laws, relying heavily on massive investments and large cluster models for development but now appears increasingly rigid in its technological approach—lacking diversity and flexibilityAs for the direction stemming from its technological choices, it has strayed from exploring groundbreaking algorithms.

Baidu's emphasis on a closed-source model further exacerbates its situation, diverging from DeepSeek's open-source routeCEO Robin Li previously claimed that open-source models would require a greater parameter scale to match closed-source models, implying higher inference costs and slower processing speeds

Advertisements

Advertisements