A doctor anywhere likely loves using AI as a second opinion on her lawyers’ legal advice, to prepare for a meeting or to fact-check his claims. However, she perhaps still has only scorn, when she is wearing the other shoe. She has a list of reasons why AI should not be trusted for her domain.
One can reverse the positions of the doctor and the lawyer above, and the arguments will be the same. We can create similar examples using any pairs not just from doctors and lawyers, but also from analysts, videographers, writers, coders or any of dozens of other professionals. The conclusion is always the same.
So, when a surveyor comes calling, whether you trust AI, what is the likely answer from the Doctor busy using a chatbox for her legal case? In societies where most people are service-sector professionals or experts, the displayed or expressed distrust towards AI, driven by views towards one’s own field of expertise, tends to be higher, even if the usage is not.
But not all societies are built that way.
The trust chasm
It is hard to overstate how psychologically different the AI moment feels in China versus the West. The Western conversation still carries a tone of “Are we sure this is real?” dominated by analysts, journalists, podcasters and other voices, who are not much different from the doctor above.
The “slop” discussions are amplified by a section of technology companies facing disruption pressures that increasingly engulf not just database companies but an increasing number of the B2C segments.
The Chinese conversation, exemplified by a panel discussion that went viral this week, increasingly resembles a consumer-tech conversation: “Which AI app should I use, and how fast can I integrate it into what I already do?”
This difference is not merely cultural trivia. It has investment consequences. It changes the speed at which products become habits, habits become workflows, and workflows become revenue.
And if there is one under-appreciated rule in technology investing, it is this: the biggest fortunes are usually not made by the cleverest debate. They are made by the lowest-friction adoption.
The following numbers are from a small number of select surveys. With more surveys conducted regularly, and realities turning rapidly, the precise numbers analysts like to work with could be different with a different selection, but the gap is wide enough to not affect the conclusions.
In China, possibly because of the less dominant role of service-sector professionals discussed above, as well as many cultural factors, and because of less circular effects of media doomerism, 83% of people see AI products and services as more beneficial than harmful. This is as per the Stanford HAI’s AI Index 2025. In the United States, that number is 39%. In Canada, 40%. In the Netherlands, 36%.
This is not unique to China. Indonesia sits at 80%. Thailand at 77%. India at 71%. The pattern is clear: in emerging economies with younger demographics, manufacturing-heavy employment, and less entrenched professional services sectors, AI optimism runs high.
We should not pretend surveys are destiny, particularly for us at GenInnov, who have repeatedly ignored the most discussed ones like the MIT survey in preference for the explosive actual token growth.
But sentiment is still the first and easiest proxy for adoption friction. A country where the default stance of opinion-setters is “let’s try it” has a different usage trajectory than a country where the default stance is “let’s regulate it, litigate it, and write an op-ed about it.”
China’s generative AI user base reportedly reached 515 million by June 2025, a 36.5% adoption rate within six months, more than doubling from late 2024. If these numbers are even directionally correct, and we believe they are, then China is not merely experimenting with AI.
At least as per some of the more quoted surveys, only one in four American companies have deployed AI in any meaningful way. Some claim 80% of Western organizations report seeing no tangible impact on their bottom line from AI investments. Others say 42% of companies abandoned most AI initiatives in 2025, up from 17% the year before.
To be fair, these numbers are increasingly being widely contested, with a series of new surveys painting a different picture, partly reflective of the fast-changing nature of the industry. We are not believers of any Western paralysis type arguments. The note is about the less discussed other side.
China never loved the cloud; it often loved the new
China has spent the better part of a decade resisting cloud computing. Some have tagged China’s “Cloud Lag” as a sign of weakness. Historically, Chinese firms spent only 2% of their revenue on IT, half the global average.
While a dollar spent on hardware in the US generated over three dollars in software spending, that same dollar in China generated only about 50 cents. The reasons were partly demographical in the lower cost of labor, and partly cultural.
The most observable impact was on subscription revenues. China, and many other countries in the middle-income categories, never adopted subscriptions the way it happened in the West.
The other impact was on the use of cloud platforms. Enterprises preferred owning servers. Whether they were for cost reasons, or for the balance sheet funding, or privacy distrusts, linkages to the era of software piracy, or some other reason, one can argue, but the end result has been stark.
China’s entire cloud market generated roughly US$46 billion in revenue. The United States exceeded $380 billion, more than eight times larger, despite similar GDPs. The SaaS market, which drove much of America’s cloud adoption, tells the story most clearly: China’s SaaS market in 2020 was $5.2 billion, barely 4% of the $120 billion US market.
Of course, history also shows China’s adoption of other new technologies following different trajectories. This is the same country that skipped credit cards entirely. In 2017, Chinese mobile payments totaled $12.8 trillion while the United States managed roughly $50 billion.
What mobile payments proved is that China does not resist technology but can follow paths that need to be assessed independently and without drawing parallels with what could be happening elsewhere.
Is AI that kind of technology? The data suggests yes. More importantly, as we discussed in the Tech is Dead as We Know It, the AI compute cannot happen on the type of distributed hardware Chinese systems are used to. China skipped cloud. It cannot skip the cloud for AI.
Aided by the low base, and spurred by the enthusiastic adoption, the growth rates of cloud services in China could achieve levels unthinkable in the West. Some expect a doubling in five years, but the logic suggests the possibilities of an even quicker surge.
We must also keep in mind that Beijing is not leaving the AI adoption to market forces or consumer preferences. The State Council’s “AI Plus Action Plan” sets explicit targets: AI penetration in intelligent terminals must exceed 70% by 2027 and 90% by 2030.
In many ways, they are compliance targets that local governments and state enterprises must meet. Once again, these have massive implications for the cloud industry’s growth in China.
Alibaba: fingers in all the pies
Is Alibaba like Amazon? Or, Google? Or Microsoft? It has traces of the most attractive aspects of each, but it is also completely different. The best way to summarize is to say it has exposure to multiple AI-facets in ways far deeper than any of the Western giants. Let’s start with what it is doing in models.
In December 2025, Alibaba’s Qwen models recorded single-month downloads on Hugging Face that exceeded the combined total of the next eight most popular models, Meta, DeepSeek, OpenAI, Mistral, Nvidia, Zhipu AI, Moonshot and MiniMax. Cumulative downloads crossed 700 million. Nearly 400 models have been open-sourced under the Qwen family. Over 180,000 derivative models have been built on Qwen backbones.
Whether download counts translate to revenues is a reasonable question. Alibaba effectively subsidizes much of China’s AI rollout through open-source generosity. Quarterly adjusted EBITA in Alibaba’s Cloud Intelligence business rose by just $132 million during the twelve months through September 2025, modest returns on billions in capital expenditure.
But the strategic logic is clear. Every derivative model built on Qwen, every developer who learns the Qwen API, every enterprise that integrates Qwen into production, they all become part of an ecosystem that increasingly runs on Alibaba infrastructure.
Alibaba holds 35.8% of China’s AI cloud market. That is more than its next three competitors combined, which are ByteDance’s Volcano Engine (14.8%), Huawei Cloud (13.1%), and Tencent Cloud (7%). In the September 2025 quarter, Alibaba Cloud revenue grew 34% year-over-year. AI-related products have now logged nine consecutive quarters of triple-digit growth.
But the market share tells only part of the story. Alibaba has systematically invested in virtually every promising AI venture in China.
Alibaba’s Strategic AI Investments
In some ways, this is the much-despised “round-trip” model. Investment agreements that stipulate portions of the capital that must be used to purchase Alibaba’s computing power. More realistically, these could be perceived as business-promotion investments in the worst case, which lock users into Alibaba’s infrastructure in an explosively growing industry.
Sanctions and Aegaeon advantage
The most significant constraint facing any Chinese tech firm is the US export control regime, which restricts access to cutting-edge Nvidia chips. Alibaba’s response has been to stop thinking like a hardware buyer and start thinking like a software architect.
The result is Aegaeon, a GPU pooling system that enables token-level virtualisation. In traditional cloud setups, a GPU assigned to an idle task sits wasted. Aegaeon decouples memory from compute, breaking requests into individual tokens and scheduling them dynamically across GPU pools. The claimed results: 82% reduction in Nvidia GPU usage for certain LLM workloads. A task that once required 1,192 GPUs was served using just 213.
Whether these numbers hold across all workloads is unclear. But if even directionally correct, Aegaeon effectively multiplies Alibaba’s chip stockpile and provides competitive moats. It allows price competition without margin destruction while providing a buffer against tightening sanctions.
Google-like integration with the old
We have long been awaiting Google to press home the advantage of decades of user data it has through its extensive ecosystem. We have repeatedly discussed the potential moat it has in articles last year.
The announcements last year had begun to show how it could take the AI utility to consumers that its competitors may not be able to match. With Personal Intelligence announced this week, we are likely to witness the actualization.
Coincidentally, within hours, Alibaba announced the integration of its Qwen AI app with Taobao, Alipay, travel service Fliggy, and mapping service Amap. Google’s Personal Intelligence entered public testing in the US immediately. Alibaba’s integration entered public testing in China immediately.
The ambition is to help Qwen’s 100 million users shop, book travel, order food, and make payments via a single AI-powered interface. Early demonstrations showed Qwen completing tasks like recommending products, booking flights, and processing payments without switching between apps.
In semantics, this is another agentic AI or AI that performs tasks. Alibaba’s competitors could announce something similar soon. But few have the advantage of past information that Google or Alibaba has. Alibaba has two other advantages: one, ChatGPT-like model popularity, and second, Amazon-like market share in cloud infrastructure, particularly in a country where advanced compute power is more scarce than anywhere.
The strategic implications are significant. If AI shopping gains traction, Alibaba’s ecosystem of Taobao (commerce), Alipay (payments), Fliggy (travel), Ele.me (food delivery), Freshippo (groceries), Amap (navigation), and others becomes not just a collection of services but an integrated platform where AI handles the orchestration.
This is just one example. One can create other examples in enterprise and finance spaces on how Alibaba is in a unique position to expand traditional businesses using its diverse strengths.
Qwen: open and complete; Android of Chinese AI
Unlike ChatGPT’s closed, API-dependent model, Qwen operates as a full model family under permissive Apache 2.0 licensing.
The suite spans language models (from 0.6B to over 1 trillion parameters), vision-language models, audio understanding, code generation, mathematical reasoning, translation across 92 languages and video generation. Alibaba has open-sourced over 300 models since August 2023, and the “full-stack, open-source” approach is intentionally promoted outside China as well for potential medium-term opportunities against global giants.
Over 290,000 customers across robotics, healthcare, education, finance, and automotive have deployed Qwen via Alibaba’s Model Studio as of January 2025, with API calls increasing 15x year-over-year.
Lenovo integrated Qwen3 into its AI agent, Baiying, which now serves over 1 million business customers for operations and IT management across 119 languages. FAW Group, one of China’s largest automakers, built its internal AI agent OpenMind on Qwen for policy analysis and intelligent reporting.
Signify (Philips Lighting) launched the industry’s first GenAI agent for smart street lighting, enabling natural language commands for dimming strategies and maintenance. SK Telecom’s A.X 4.0, one of the five consortia selected for Korea’s sovereign AI, is built on top of Alibaba’s Qwen 2.5.
Another Korean Sovereign AI aspirant, Naver, is also building it non-sovereignly on Qwen, apparently, and coming under pressure because of that. Singapore’s national AI program is also building its flagship model on Qwen.
Over on consumer hardware side, Alibaba launched its Quark AI Glasses in November last year, which is a smart eyewear with dual micro-OLED displays and deep integration across Alibaba’s ecosystem.
Users activate via “Hello Qwen” for real-time translation, AI Q&A, meeting notes, price comparison on Taobao, payments through Alipay, and navigation via Amap. More significantly, Xiaomi has integrated Qwen into its AI assistant Xiao Ai across smartphones and the SU7 electric vehicle.
Effectively, vehicle owners can generate images on the car’s infotainment system via voice commands and receive visual guidance for vehicle operation. MediaTek deployed Qwen3 on its Dimensity 9400 smartphone platforms with 20% faster inference through optimized speculative decoding. Apple Silicon, Nvidia, AMD and Arm have all released optimized Qwen deployments for their respective hardware ecosystems.
Qwen models are relatively young in the Vision-Action-Language space, but are making rapid gains. Specific models have been integrated with Nvidia’s Drive AGX Orin platform for autonomous vehicles.
Alibaba unveiled its automotive intelligent cockpit solution powered by Qwen and running on Qualcomm’s Snapdragon 8397 platform. Early investments and developments have been noted in Robotics as well.
Qwen’s platform power is most visible when one sees it being used by the other model developers. We discussed Singapore and Korea’s sovereign development efforts using Qwen foundations. The aggregate derivative model count tells how pervasive the issue is.
Stanford’s HAI analysis found that 63% of all new fine-tuned models on Hugging Face in September 2025 were built on Chinese base models, primarily Qwen and DeepSeek. DeepSeek’s R1 reasoning models were indeed developed on Qwen foundations, specifically utilizing Qwen’s vision encoders and base language models. In the US, Fei-Fei Li’s team used Qwen as the base for its S1 inference model.
Baba: where the biggest model-maker meets the biggest Cloud provider
There is something almost disconcerting about the combination. The largest cloud provider in the world’s fastest-adopting AI market, developing the most downloaded open-source models globally, integrated with the dominant commerce and payments infrastructure, now building custom silicon to boot.
Try to construct this portfolio elsewhere: you would need OpenAI’s model prowess, Amazon’s cloud dominance, Google’s consumer reach, Meta’s open-source philosophy, and a chip design team, and then somehow graft them onto a population that wants to use AI. The combination does not exist outside China. Inside China, it is called Alibaba.
Before one turns to the “Brutalism” of the Chinese market, the 97% price slashes and the scorched-earth rivalry with Huawei, Tencent, and ByteDance, it is important to note that Alibaba’s “moats” are increasingly compounding rather than just coexisting.
The ModelScope community now hosts over 70,000 open-source models for 16 million users, while its Model Studio marketplace supports a staggering 800,000 agents across every conceivable vertical from healthcare to smart-home robotics.
When the Qwen App reached 30 million monthly active users within just 23 days of its November 2025 launch, it signaled something more profound than a traffic spike; it confirmed that the “Cloud Lag” of the last decade is dissolving into a desperate, centralized rush for compute.
In a market where the government has set an unyielding 90% AI penetration target by 2030, the company that provides the “Android of AI” effectively owns the operating system of the world’s second-largest economy.
The risks, however, are as massive as the potential. The primary skeptical case rests on whether all the spend can ever translate into sustainable, high-margin profits. Enterprise monetization remains the great “unproven” of the East.
While American firms have a decades-long habit of paying for software-as-a-service (SaaS), Chinese enterprises are still learning to view AI as a core OpEx rather than a hardware one-off. Furthermore, the “Silicon Curtain” of US export controls remains a persistent threat.
Alibaba’s Aegaeon pooling system, which stretches GPU capacity by 82%, is a brilliant software buffer, but it is not an absolute immunity. If the hardware spigot is tightened further, even the most efficient software will eventually hit a ceiling.
Alibaba’s current 380 billion yen ($53 billion) capex commitment is more than it spent on cloud in the entire previous decade. If the market continues to commoditize base models through price wars, Alibaba’s survival will depend on its ability to integrate AI so deeply into the daily lives of its 53 million 88VIP members and its millions of corporate merchants that it becomes impossible to switch.
Whether they can charge meaningful tolls at the gate or end up subsidising everyone else’s passage will separate a good thesis from a great investment.
Nilesh Jasani is the founder and CEO of GenInnov Pte Ltd Singapore. This article first appeared on www.geninnov.ai and is republished with permission. Read the original here. Read more at www.geninnov.ai/blog