In earlier section, I discussed that embodied AI, i.e. robots and humanoid, will be among the next big AI moves out of China. I believe another major trend will be application of AI technologies in vertical industries that have yet to be affected by the horizontal foundational LLMs, let alone experiencing broad adoption of AI tech. AI applications will be the most critical deciding factor in future AI advancements.
Here China’s industrial superiority and huge market will give it a distinct advantage in the AI competition.
AI in applications
Broadly speaking, there are 3 levels in the AI ecosystem. At the foundational level are AI hardware such as Nvidia’s Blackwell chips, Huawei’s Ascend, and data centers. In the middle, you’ll find the large language models such as ChatGPT, Llama, DeepSeek and Qwen. At the top are the various AI applications, e.g. in healthcare, banking, manufacturing, war fighting, etc.
This architecture is similar to the earlier generation of digital technology, with hardware like semiconductor chip, PC, mobile devices at the bottom, operating systems like Windows, Android, HarmonyOS in the middle, and applications such as Amazon, WeChat, TikTok on top.
In a mature ecosystem, the customer-facing activities will happen at the application level and the bulk of revenue and profit generated in the ecosystem should be here.
Today, in the AI ecosystem, just like the early days of the digitalization era, most of the revenue and profits is going to the foundational chip level, with Nvidia as the biggest winner. Over time, however, the revenue and profit of the industry will flow upward to the LLM and application levels. Eventually, the bulk of value will reside in the applications.
In such a mature AI ecosystem, still years away, multiple vertical AI applications will be built upon the horizontal LLM models and that is where AI will start to touch our lives in the most direct and impactful way. For example, Manus, a new Chinese AI assistant built on top of the Alibaba Qwen LLM, is able to perform tasks such as designing websites and planning travel itineraries upon user request. Such a development is an illustration of how we are progressing towards AGI. Putting aside the question true AGI or singularity is ever going to arrive, which I have my doubts, there is little dispute AI will be used more broadly in the economy and our daily lives.
As we are already seeing the slowdown of the scaling law, the hyper-scalers such as OpenAI and Meta are facing a diminishing return on raw compute. They also face a lack of new data as online data used for training is nearly exhausted. However, vertical industry data, much of which offline today, has barely been tapped yet.
China’s strategy is to drive AI from the chip and LLM levels to the application level as fast as possible. Given China has the world’s most industrialized economy, there are numerous ways to incorporate AI to enhance productivity and deliver higher ROI for AI users. China has a wealth of manufacturing, industrial, banking and healthcare data that can be used to train vertical models to lead AI applications.
This is why in the last two months, the majority of large Chinese car companies (BYD, Cherry, Geely, Nio), telecom companies (China Telecom, China Mobile), energy companies (PetroChina, Sinopec, CNOOC, China Nuclear), industrial manufacturers (Sinochem, Baosteel, Jiangnan Shipping), and financial institutions (ICBC, BOC, CCB, Ping’an Insurance) have announced they are incorporating DeepSeek into their business operations.
Such integration is also going beyond these large businesses. Numerous hospitals, local government agencies, and schools are integrating DeepSeek into their workflows.
A good example of AI application can be found at a major Shanghai hospital. It was reported in February that Shanghai Zhongshan Hospital officially released the beta version of China’s first AI model in the cardiovascular field focused on cardiovascular diagnosis and treatment via AI agent.
CardioMind, co-developed by Zhongshan Hospital and Shanghai Academy of AI for Science, aims to become an AI doctor specialized in cardiovascular disease diagnosis and treatment with top-level experience, according to Ge Junbo, member of the Chinese Academy of Sciences and director of Zhongshan Hospital’s cardiology department. “We’re feeding CardioMind data and teaching it to think like a top expert,” Ge announced in the press release.
In addition to general cardiovascular disease diagnosis and treatment knowledge, CardioMind has also absorbed hundreds of thousands of electronic medical records from the cardiology department of Zhongshan Hospital, learned the ways doctors think in diagnosis and treatment, and picked up various difficult cases.
“CardioMind’s knowledge is precisely focused on all types of cardiovascular diseases,” Ge said, adding that the AI model can comprehensively process various examination data from electrocardiograms, ultrasound images, and laboratories and make diagnosis and treatment conclusions by processing the information.
The cardiology department of Zhongshan Hospital admitted 820,000 patients last year, according to Ge. “With the help of CardioMind, our doctors can serve more patients, reduce the overall workload, and improve the quality of diagnosis and treatment.”
CardioMind transforms the expertise of top doctors from leading medical institutions into “digital diagnosis and treatment capabilities” that can be replicated, Ge said. The promotion and application of CardioMind can accelerate the utilization of high-quality medical resources in hospitals.
Another example of such AI deployment is in military combat scenarios. The Chinese military is already starting to deploy AI and embodied AI into its operations.
AI is already used in aerial dog fights. However, there is a critical flaw in current AI air combat systems – their reliance on trajectory-based predictions, which struggle to account for sudden, non-linear manoeuvres executed by human pilots.
The Northwest Institute of Mechanical and Electrical Engineering, a key research arm of Norinco, a top Chinese arms supplier, is solving this problem with upgrades to current AI used in air combat. Research team at the institute has developed a technology that combines advanced infrared imaging with AI-driven predictive modelling to anticipate opponent fighter pilot’s moves by detecting subtle wing-tail movements.
Using a modified YOLOv8 neural network, the system analyses infrared imagery to detect millimetre-level deformations in an opponent’s control surfaces – such as a F-15’s 1.5-metre rudder or two-metre elevator – during flight.
These real-time observations feed into a long short-term memory (LSTM) network improved with attention-weighing mechanisms, enabling the AI to predict manoeuvres before they fully unfold. Human pilots rely on instinct and unpredictability, but every physical maneuver has mechanical precursors. By decoding these signs – a rudder tilt, an elevator shift – the new model resolves the “black box” of human decision-making and can make split second strike decisions.
As AI gets integrated into real life situations, the scope and scale of a country’s industrial base and consumer markets will play a decisive role in the advancement of future AI innovations.
My third prediction is about AI’s commercial development. I believe AI will follow the same trajectory as all the past technological developments – it will get cheaper as AI penetrates into mass market. Affordability, efficiency, and innovation will be the focus of competition in the long term.
As shown in industry after industry, Chinese companies excel in delivering low cost, high quality products and solutions as the Chinese market is the most competitive globally. As a result, China is likely to have the world’s most competitive AI application markets, just like with EV, solar panel, batteries, smart phones, etc.
Already we are seeing Chinese companies take a different approach to AI development from OpenAI, Meta, and Tesla. Rather than building a moat protected by high Capex investment, proprietary software and high profit margin, DeepSeek, Alibaba, Unitree and BYD are developing an ecosystem through open sourcing, engineering optimization, and free/low-cost software. The goal is to drive quick adoption, scale and fast iteration, thus achieving long term market share gains.
Mass market adoption of AI
As the technical frontier for AI matures and applications move to the center stage, affordability becomes a critical issue for broad adoption. Cost remains one of the most formidable barriers to AI implementation, particularly for large language models.
DeepSeek has dismantled this obstacle by offering an open-source strategy and cost-effective training solutions, making AI accessible to businesses and entities that previously found it prohibitively expensive.
This is probably the most important contribution of DeepSeek and China in AI development, democratizing AI for businesses and consumers who otherwise couldn’t afford advanced AI solutions.
This isn’t merely a tech breakthrough; it’s a shift in AI economics that is forcing competitors, both in China and abroad, to rethink their own business models.
As a result of DeepSeek’s breakthrough, a new framework, called Chain-of-Experts (CoE), has emerged as the next frontier beyond the high compute, high cost classic LLMs, also referred to as dense models, which activate every parameter simultaneously during inference leading to extensive computational demands.
The aim of CoEs is to make LLMs more resource-efficient while increasing their accuracy on reasoning tasks, by addressing the limitations of the classic LLMs. This new algorithmic framework vastly accelerates the speed and relevance of model output by activating “experts” — separated elements of a model, each specializing in certain tasks — sequentially instead of in parallel.
This structure allows experts to communicate intermediate results and gradually build on each other’s work rather than crunching vast amount of data that may be largely irrelevant yet take up much time and compute resources.
Washington’s effort to strangle China’s AI progress through chip restrictions has inadvertently pushed China down this path. Since Chinese companies do not have access to the best AI chips, they are forced to explore innovative ways to achieve high AI performance with limited compute. DeepSeek’s breakthrough is to achieve at-parity, even superior, performance without using the best hardware and incurring the heavy associated costs.
As often said, necessity is the mother of inventions. Washington’s bellicose actions have completely boomeranged.
The low cost mass market adoption approach is not restricted only in horizontal foundational LLMs. The same strategy is being pursued by Chinese manufacturers when they incorporate AI into their products at the vertical application level.
BYD, Geely and Cherry, all top EV makers, have started to offer free autonomous driving AI software in their cars. BYD has extended free autonomous driving to all its models including the sub-$10,000 Seal model. In contrast, Telsa charges $8000 for its autonomous driving software package or a $200 monthly fee.
Chinese consumer electronics companies are integrating AI into low cost consumer products that are widely used already. Companies like Huawei and Xiaomi have embedded AI into their smartphones and home devices, making AI widely accessible in markets with lower purchasing power.
Unlike AI models in the US, which often require high-end computing infrastructure, China’s AI is optimized for lower-power devices, which is more practical for regions with weaker digital infrastructure. This gives China a strategic edge in markets where western AI solutions are too expensive or incompatible.
Additionally, China has built over 1.9 million 5G base stations across Belt and Road Initiative nations. The Beidou satellite navigation system, China’s alternative to GPS, is in use in over 120 countries, providing location-based AI services for industries such as agriculture, urban planning and security.
With affordable AI-powered products, investments in digital infrastructure and a growing influence in AI governance, China is positioning itself as the leading AI player in developing economies.
China’s AI expansion is guided by a clear strategy. The government 2017 New Generation Development Plan laid out a road map to make China the global AI leader by 2030, focusing on breakthroughs in AI infrastructure, applications and industrial integration.
The plan aligns with China’s “digital silk road” initiative, which promotes digital infrastructure projects, including fibre optic networks, 5G, cloud computing and AI-powered services. The digital silk road initiative is an extension of the Belt and Road Initiative (BRI) launched by President Xi in 2013.
Another consequence of driving down AI costs is the recalibration of China’s tech ecosystem. While tech giants like Alibaba and Tencent have dominated AI investments in the past, startups with fewer resources, exemplified by DeepSeek and Manus, can compete and innovate at the same footing without massive upfront investments.
This is driving a shift towards a more diverse AI landscape, one where smaller, more agile firms can challenge the incumbents. It also raises the stakes for China’s AI talent pool, as engineers and researchers gravitate toward the country’s most promising frontier firms.
One of the most noteworthy characteristics of DeepSeek is its open source stand on AI tech standard. The same open source approach is being adopted by companies such as Ubtech Robotics and Alibaba in their humanoid tech and chip tech programs. Ubtech has open sourced its humanoid design while Alibaba has done the same with its latest RISC-V chip design. Design plans, software, and detailed schematics are made freely available.
The open source approach empowers developers everywhere to innovate without starting from scratch, effectively lowering barriers to entry in these high tech areas. As noted by organizations like the IEEE Robotics and Automation Society, sharing technological blueprints can accelerate innovation across the industry.
Open sourcing technology isn’t just about saving time and cutting costs. It also ignites a global movement to standardize technology and enhance collaboration among research institutions and tech companies.
Open source collaboration also helps to prevent market fragmentation caused by excessive reliance on proprietary systems, instead fostering rapid iteration and large-scale adoption to ensure a dynamic synergy between technological innovation and market applications.
This democratization of technology is how open-source software like Linux transformed the computing world, offering a powerful, community-driven alternative to proprietary systems.
China’s drive to offer low-cost, open source AI will spur a wave of innovations and broad adoption among both Chinese and non-Chinese businesses in different industries, eventually benefiting everyone in the ecosystem.
