Alibaba

Baidu

ByteDance

Meituan

Pinduoduo

Tencent

Most Followed

KrASIA Connection

China's Auto Sector

Earnings Reports

Inside China's Startups

Venture Roundup

Our Events

View All

Collections

News

Features

Insights

Ecosystem

DeepSeek

Kuaishou

Li Auto

Xiaohongshu

Xiaomi

Xpeng

On the Radar

Events

Latest

About

36Kr (EN)

Consumer

Robotics

Fintech

In the span of a month, <a href="/company/deepseek">DeepSeek</a> has gone from an obscure venture to a global sensation, upending Silicon Valley’s hierarchy in the artificial intelligence race and sparking a regulatory firestorm. The company’s open-source V3 and R1 models, launched in January, have fueled an adoption surge that is already forcing industry heavyweights—and governments—to take notice.
Yet, as with any disruptive force, DeepSeek’s ascent has come with baggage. Governments are launching investigations, security researchers are raising red flags, and lawmakers in the US are debating whether DeepSeek should even be available to the public.
DeepSeek’s namesake AI assistant debuted publicly on January 11 as a free app, and the numbers since have been staggering.
According to data from <a href="https://mp.weixin.qq.com/s/lWbzaiIm8hRMu9xgy5NFKg">Aicpb.com</a>, a platform tracking global AI adoption, DeepSeek averaged 22.2 million daily active users (DAUs) in January—equivalent to roughly 41.6% of ChatGPT’s DAUs (53.2 million). DeepSeek has also overtaken ByteDance’s Doubao, another Chinese AI assistant that managed about 17 million DAUs during the same period.
DeepSeek’s first 20 days post-launch have left an impression, averaging 21.6 million DAUs—surpassing ChatGPT’s 14.6 million in the same period.
DeepSeek’s monthly active users (MAUs) reached 33.7 million, making it the fourth most used AI app globally, behind ChatGPT (349.4 million), Doubao (78.6 million), and Nova (56.6 million), a proofreading AI tool.
Aicpb.com’s tally accounts only for (mobile) app usage and omits web users.
As of end-January, the DeepSeek app ranked first in 157 countries and regions on the iOS App Store. The top five countries where the app has gained the most traction are China (30.7%), India (13.6%), Indonesia (6.9%), the US (4.3%), and France (3.2%).
The sheer velocity of DeepSeek’s expansion has prompted swift responses from governments worldwide, with some already initiating probes, imposing restrictions, or outright banning the app.
In Europe, Italy was the first to launch an investigation, quickly followed by France, Germany, Ireland, the Netherlands, Portugal, and the UK. Outside of Europe, Australia, Taiwan, South Korea, and—most significantly—the US have all moved to curb or scrutinize DeepSeek’s operations.
While US consumers still have access to DeepSeek, federal agencies such as the National Aeronautics and Space Administration (NASA) and the US Navy have warned employees against using the app on government devices. Legislators, meanwhile, are weighing an outright ban.
The current proposal comes from US senator Josh Hawley, who introduced a bill that would criminalize the import and export of AI products between the US and China. If passed, the law could punish individuals who knowingly download Chinese AI technology, including DeepSeek, with up to 20 years in prison or a USD 1 million fine—or both.
The crackdown mirrors the US government’s handling of TikTok, which faced a Supreme Court-backed ban that was only halted following direct intervention from Donald Trump, who recently returned to the White House for his second presidential stint. DeepSeek, however, may face an even steeper uphill battle, given the national security implications of AI technology.
whereRelatedArticle
The fear surrounding DeepSeek isn’t entirely unfounded. Since its launch, security researchers have uncovered troubling vulnerabilities in its infrastructure.
One of the most damning findings came from <a href="https://www.wiz.io/blog/wiz-research-uncovers-exposed-deepseek-database-leak">Wiz Research</a>, which reported that it had accessed a publicly exposed database belonging to DeepSeek. The database contained over a million lines of log streams, including chat histories, secret keys, backend details, and other sensitive information. The database reportedly allowed full control over operations, meaning anyone who discovered the exposure could manipulate DeepSeek’s internal data. DeepSeek patched the vulnerability after being notified.
Despite the regulatory storm, industry leaders have responded to DeepSeek’s emergence with measured optimism. OpenAI CEO Sam Altman, speaking in Tokyo on February 3, acknowledged DeepSeek’s capabilities but said he remains confident in his company’s lead:
<blockquote>“DeepSeek is certainly an impressive model, but we believe we will continue to push the frontier and deliver great products, so we’re happy to have another competitor, … We’ve had many before, and I think it is in everyone’s interest for us to push ahead and continue to lead.”</blockquote>
DeepSeek’s breakneck rise presents a paradox: it is both the biggest success story in open-source AI to date, and the fastest-growing regulatory headache for governments worldwide.
Yet, from a user perspective, DeepSeek’s popularity speaks to a hunger for powerful AI models, regardless of where and how they are made. With 70% of its users outside China, the app has demonstrated broad global appeal even as governments scramble to decide their next move.

DeepSeek hit 33.7 million users in January—but is a crackdown imminent?

East Asia

China

Artificial Intelligence

Deep Tech

… algorithms to improve business performance may face disruption as independent AI agents reduce their influence over consumer choices.
Another possible outcome is that every app or platform could eventually be powered by its own AI agent. While still speculative, AI agents could evolve to network and coordinate with one another, enabling dynamic and complex interactions across multiple systems.
This would mark a significant update from current models. OpenAI’s Operator, for instance, is (currently) designed to interact with existing platforms through a browser interface. Future AI agents, however, could be capable of planning and synchronizing actions across robots, agents, users, and systems.
In the context of e-commerce, consider how this progression might unfold:
<ol>
<li aria-level="1">AI assistants (current): These AI-powered tools assist users in product discovery, price comparisons, and recommendations within the platforms they are embedded in.</li>
<li aria-level="1">Platform-specific AI agents: These agents help users track and make purchase decisions within a specific platform, learning from explicit user input and behavior as well as backend data collected and analyzed by the platform.</li>
<li aria-level="1">Independent AI agents: These agents operate across multiple platforms, tracking products and making purchase decisions based on predefined user preferences. However, their capabilities may be constrained by the framework established by their provider. For example, OpenAI’s Operator interacts with browser screens but may lack access to backend data that platform-specific agents can leverage.</li>
<li aria-level="1">Future AI agents: Every digital touchpoint—retail sites, logistics providers, payment gateways—could be fronted by AI agents capable of communicating with each other. These AI agents wouldn’t just retrieve data but could dynamically negotiate, plan, and execute transactions based on user-defined goals.</li>
</ol>
<h3>The future of commerce: Fully automated, or human in the loop?</h3>
Agentic commerce will likely streamline routine shopping, but will humans be entirely removed from the process? Not necessarily.
Everyday essentials like groceries and household supplies are well-suited for full automation. AI agents can monitor inventory, track price fluctuations, and make purchases at optimal moments—removing the need for manual oversight.
On the other hand, high-value or experience-driven purchases—such as cars, luxury goods, and fashion items—are less likely to be entirely delegated to AI. While agentic AI can streamline research, facilitate comparison shopping, and even assist in negotiations, the final decision will likely remain with the human buyer, where personal preference and tactile experience play a crucial role.
After all, would you trust an AI agent to pick your next car or luxury handbag based purely on past data and algorithmic predictions?
<h3>Risks and challenges of agentic commerce</h3>
Despite its promise, agentic AI presents serious challenges that must be addressed before it reaches mass adoption:
<ul>
<li aria-level="1">Trust and reliability: AI agents will need near-perfect accuracy in decision-making. The risks of purchasing the wrong item, misunderstanding a user's intent, or making unauthorized purchases could erode consumer trust.</li>
<li aria-level="1">Ethical and security concerns: Delegating financial authority to an AI raises concerns about fraud, unintended purchases, and misuse of personal data. How do users retain control over autonomous systems acting on their behalf?</li>
<li aria-level="1">Business model disruption: If AI agents optimize purely for price and efficiency, what happens to brand loyalty, advertising, and retailer-driven engagement? Businesses may have to rethink how they engage customers.</li>
</ul>
Agentic commerce is on the horizon, and OpenAI’s Operator is a significant step toward making it real. As AI agents become more capable, shopping is expected to shift from a manual, decision-heavy process to an optimized and more automatic experience.
Still, the transition won’t happen overnight. Retailers, AI developers, and regulators will need to collaborate on guardrails that ensure transparency, security, and user control.
For now, we stand at a tipping point: the shopping cart is still in your hands—but maybe not for much longer.

Agentic commerce is near: AI could soon turn into your personal shopper

OpenAI

Just weeks after Honda and Nissan revealed <a href="https://kr-asia.com/honda-and-nissan-in-merger-talks-to-secure-ev-future">plans to explore a merger</a> that could have created the world’s third largest automaker, those discussions appear to be unraveling. According to a <a href="https://asia.nikkei.com/Business/Business-deals/Nissan-to-suspend-merger-talks-with-Honda">Nikkei Asia</a> report, Nissan has informed Honda of its decision to suspend talks, citing fundamental disagreements over valuation and governance. Meanwhile, conflicting statements from both companies suggest that discussions remain ongoing—though uncertainty looms large over the fate of the deal.
Sources cited by Nikkei Asia said that Nissan balked at Honda’s proposal to make it a subsidiary, a deviation from the original plan to form a joint holding company with near-equal influence. The automakers had agreed to determine their respective valuations based on average share prices before the memorandum of understanding (MoU) was signed in December. However, rough estimates suggest the arrangement would heavily favor Honda, whose market capitalization is more than five times Nissan’s size—an imbalance that has become a key point of contention.
Another sticking point was Nissan’s turnaround plan. Honda had set a January deadline for Nissan to present a clear restructuring roadmap before moving forward with integration. Nissan, which has struggled with declining sales and profitability in recent years, unveiled a plan to cut 9,000 jobs globally, with layoffs in North America and Thailand. But Honda reportedly found Nissan’s proposals inadequate and urged more aggressive restructuring measures.
According to <a href="https://www.reuters.com/markets/deals/nissan-ceo-uchida-tells-honda-ceo-mibe-he-wants-terminate-merger-talks-asahi-2025-02-06/">Reuters</a>, Nissan CEO Makoto Uchida met with Honda’s CEO Toshihiro Mibe on February 6 to inform him that Nissan wants to formally withdraw from the MoU. Nissan’s board is expected to finalize this decision before it announces its latest earnings.
Despite the reports, both automakers have refrained from making a definitive statement. In a carefully worded response, Honda and Nissan acknowledged media reports but maintained that they are “in the stage of advancing various discussions… We plan to establish a direction and make an announcement around mid-February.”
A report from <a href="https://www.automotivedive.com/news/nissan-honda-merger-talks-ongoing/739318/">Automotive Dive</a> also claims that talks remain ongoing, citing a Honda spokesperson who said that more details will be shared at an appropriate time.
If talks collapse, Nissan will be left to pursue its turnaround independently, while Honda may seek alternative partnerships to solidify its position in the automotive market.
For now, the merger’s fate hangs in the balance, and all eyes are on Nissan’s next move.

Nissan hits the brakes on Honda merger talks, throwing the deal into limbo

Japan

Honda

Nissan

Transportation

Electric Vehicles

Mobility

… https://console.kr-asia.com/wp-content/uploads/2025/02/HairOriginals-Piyush-Wadhwani-Jitendra-Sharma.jpg 1920w" sizes="(max-width: 1024px) 100vw, 1024px" /><figcaption id="caption-attachment-151925" class="wp-caption-text">Photo of Piyush Wadhwani (left) and Jitendra Sharma, co-founders of HairOriginals. Photo courtesy of the company.</figcaption></figure>
whereAdBanner.Uptake
<h3>Recent deals completed in China:</h3>
<ul>
<li aria-level="1"><a href="http://fwjtech.com/">Azimuth Technology</a>, a Shanghai-based indoor positioning technology company, has secured nearly RMB 100 million (USD 14 million) in a Series A funding round. The round was led by a BeiDou research institute under China North Industries Group Corporation (Norinco), with Shenzhen Capital Group (SCGC) and Shanghai Ruihe Asset Management participating. Azimuth focuses on integrating indoor navigation with outdoor satellite positioning systems to provide unified high-precision positioning solutions. —<a href="https://36kr.com/p/3133663375677955">36Kr</a></li>
<li aria-level="1"><a href="http://www.lexangre.com/">Lexangre</a>, an AI-powered consumer hardware startup, has raised nearly RMB 200 million (USD 28 million) in an angel round of financing. The investment was led by IDG Capital, with Monolith, Matrix Partners, ZhenFund, HongShan, and Vitalbridge participating. —<a href="https://36kr.com/p/3142202686265865">36Kr</a></li>
<li aria-level="1"><a href="https://www.tangjimed.com/en/">Tongee Medical Technology</a>, a Hangzhou-registered medtech company, has raised nearly RMB 100 million in a Series C funding round from Everest and Quzhou Lushi Fund. The funding will be utilized for market expansion, regulatory clinical trials, and capacity building. Tongee specializes in developing medical devices for metabolic diseases such as obesity, diabetes, and fatty liver disease. —<a href="https://36kr.com/p/3153353327467016">36Kr</a></li>
</ul>
whereRelatedArticle
<h3>Ampotech, Nibertex, Yuwell, and more made recent headlines:</h3>
<ul>
<li aria-level="1"><a href="https://www.ampotech.com/">Ampotech</a>, a Singapore-based energy management solutions provider, has secured an undisclosed investment from Clime Capital’s Southeast Asia Clean Energy Fund II (SEACEF II).</li>
<li aria-level="1"><a href="https://nibertex.com/">Nibertex</a>, a deep tech startup specializing in waterproof breathable membranes, has closed its pre-Series A funding round with backing from Foxmont Capital Partners, ADB Ventures, and regional family offices.</li>
<li aria-level="1">Yuwell Medical, a China-based manufacturer of home-use medical devices, has announced a USD 27.2 million investment in Inogen through its Hong Kong subsidiary. Yuwell will acquire a 9.9% stake in Inogen’s common stock as part of the deal.</li>
</ul>
If there are any news or updates you'd like us to feature, get in touch with us at: <a href="mailto:hello@kr-asia.com">hello@kr-asia.com</a>.
<style>.addcaption{font-size:14px!important;line-height:20px!important;color:inherit;padding-top:8px;border-top:1px solid #e8e8e8}.entry-content blockquote p{font-size:22px;line-height:33px}@media only screen and (max-width:767px){.entry-content blockquote p{font-size:20px;line-height:32px}.entry-content h3{font-size:24px;line-height:32px}}</style>

Deals in brief: Syfe bids for ASX-listed Selfwealth, Malaysia’s KMP backs Swift Bridge, SoftBank in talks to acquire chipmaker Ampere, and more

SoftBank

Venture Capital

Private Equity

Investment

Merger &amp; Acquisition

Singapore

Malaysia

Australia

Southeast Asia

India

Europe

Middle East

Oceania

Electric vehicle registrations are rising rapidly in Myanmar due partly to "parallel imports" brought in by unauthorized dealers, but a lack of repair and maintenance support could put the reputation of major brands at risk.
The country's growing EV market is readily apparent in the commercial capital of Yangon. In addition to Chinese-made EVs, one can see <a href="/company/toyota">Toyota Motor's</a> bZ4X—even though the Japanese automaker does not import the model, which is built through a joint venture in China.
Myanmar had 6,166 passenger EVs registered as of December 2024, more than triple the number a year earlier, according to the country's Road Transport Administration Department. The military regime has promoted EV imports by waiving tariffs in a bid to reduce fuel consumption.
The regime has largely banned imports of gasoline-fueled vehicles, and domestic production of such cars has stalled since the 2021 military takeover. Prices of used cars have soared, making EVs appear less expensive by comparison. A compact EV from Chinese automaker <a href="/company/byd">BYD</a> costs around MMK 150 million (USD 72,200), on par with a preowned upmarket gas vehicle.
Around 80 local companies have imported more than 50 models of EVs over the past two years, according to a state-run newspaper, which noted that brands like Toyota, BYD, and <a href="/company/saic-motor">SAIC Motor</a>-owned MG are especially popular. This has raised concerns that inadequate repair and maintenance services could damage these brands.
Several authorized Toyota dealerships all provided the same answer when contacted about purchasing the bZ4X—they do not import or repair the EV. When asked about after-sales services at a local showroom selling the bZ4X, a representative referred customers to the vehicle's importer.
In some cases, dealerships of Dongfeng Motor Group are selling the bZ4X or another Toyota EV, the bZ3, even though the Chinese automaker has no relation to the models.
"The time has come for companies to seriously consider providing after-sales services for the EVs flowing into the market," said an executive at a Japanese automaker.
Chinese companies have already begun taking action. BYD has partnered with three local companies deemed to have the proper import licenses and sufficient repair capabilities, and encourages prospective buyers to go through its official channels. <a href="/company/gac">GAC</a> has opened official dealerships for its Aion brand of EVs in Yangon and elsewhere.
Japanese, US, and European automakers are hesitant to engage in Myanmar's EV market, worried about investor perception. The family of Min Aung Hlaing, the head of the military regime, is said to run an import business, and this keeps Japanese and other companies from jumping on the country's EV bandwagon.
This article first appeared on <a tabindex="0" href="https://asia.nikkei.com/Spotlight/Myanmar-Crisis/Myanmar-EV-registrations-triple-but-brands-face-parallel-import-risks" rel="noopener noreferrer" data-token-index="1">Nikkei Asia</a>. It has been republished here as part of 36Kr’s ongoing <a tabindex="0" href="https://www.kr-asia.com/nikkei-teams-up-with-36kr-in-asia-tech-news-coverage/" rel="noopener noreferrer" data-token-index="3">partnership with Nikkei</a>.

Myanmar EV registrations triple, but brands face "parallel import" risks

Toyota

SAIC Motor

Myanmar

… stating that the brand remains firmly focused on casual apparel.
But within these constraints, Uniqlo’s innovations are starting to fall short.
In 2023, Uniqlo once again partnered with Toray Industries to develop "Pufftech," hoping to replicate the success of Ultra Light Down. Pufftech, made from ultra fine hollow fibers, was initially priced at RMB 499 (USD 69.9)—matching the cost of Uniqlo’s lightweight down series.
However, sales of Pufftech jackets have been underwhelming. According to data from third-party analytics firm Zhiyi Tech, during the 2024 Singles’ Day shopping festival, Uniqlo’s bestselling winter items did not include Pufftech. Additionally, Uniqlo’s Chinese website made no prominent mention of the new line in its warm clothing category.
Uniqlo had posted a detailed nine-step washing guide for Pufftech jackets on its official Xiaohongshu page, covering everything from reading the care label to ensuring the padding “returns to its original shape” after drying. The extensive instructions drew criticism in the comments: “It’s 2024—why is fast fashion still making clothes that can’t be machine-washed?”
Between September and November 2024, Uniqlo’s revenue in China totaled JPY 178.6 billion (USD 1.2 billion), remaining flat year-on-year, while operating profit declined by approximately 10%. In multiple financial reports, Uniqlo’s parent company attributed weak performance in Greater China to unseasonal weather and insufficient regional product differentiation. Third-party reports indicated that, for the first time, Uniqlo has closed more stores than it has opened in Greater China.
Bound by its fast fashion DNA and a focus on essential casual wear, Uniqlo has long relied on technology as a key selling point. But from the start, perhaps Yanai had set up an impossible challenge: convincing the market that a fast fashion brand could win with innovation alone.
KrASIA Connection features translated and adapted content that was originally published by 36Kr. This <a href="https://36kr.com/p/3133443722681088">article</a> was written by He Zhexin for 36Kr.

Is Uniqlo still an innovator? The fast fashion giant faces a new tech reset

Retail

Consumer Services

… traditional strategy: price cuts.
Since launch, the Camry has undergone multiple rounds of price reductions. Initially priced at RMB 171,000 (USD 23,940), the model can now be purchased for less than RMB 140,000 (USD 19,600). Even the Avalon, positioned slightly above the Camry, has seen its price drop below RMB 150,000 (USD 21,000).
However, for Toyota, price cuts come at the expense of dealer profitability. A dealership insider told 36Kr that the wholesale price dealers pay for Toyota vehicles typically ranges from 90–95% of the official retail price, depending on the model and timing. "Right now, most Toyota cars are selling at 70% to 80% of their original price—meaning dealers are selling at a loss." The only way for dealers to stay afloat is by relying on year-end rebates from Toyota.
With low margins and declining demand, some dealers are exiting the business altogether. "Last year, especially in central cities like Wuhan, Toyota dealers exited in large numbers," an industry source said.
Toyota’s decision to consolidate its twin models is a direct response to this profit erosion.
In fact, resource integration is becoming an industry-wide trend. Geely, for example, recently merged its Zeekr and Lynk &amp; Co brands, shifting from aggressive expansion to a more strategic resource optimization approach. Meanwhile, Honda and Nissan have also announced their own <a href="https://kr-asia.com/honda-and-nissan-in-merger-talks-to-secure-ev-future">consolidation plans</a> to tackle declining sales and operational challenges.
whereRelatedArticle
Unlike some global automakers that are taking a cautious approach in China, Toyota is betting big. It aims to boost its annual production in China to 2.5–3 million vehicles by 2030, an 80% increase from its 2024 sales of 1.776 million units.
To achieve this, Toyota is ramping up investments in localized R&amp;D. Its Toyota IEM R&amp;D Center is spearheading developments in intelligent and electric vehicle technology.
Li, the newly appointed general manager of Toyota China, also recently shared that starting in 2025, Toyota will build a dedicated R&amp;D system in China, mirroring Volkswagen’s strategic shift.
From technology partnerships with Huawei to investments in range-extender systems and small-engine plug-in hybrids, Toyota is charting an aggressive course for China’s evolving market.
KrASIA Connection features translated and adapted content that was originally published by 36Kr. This <a href="https://36kr.com/p/3131886737267465">article</a> was written by Xu Caiyu for 36Kr.

Toyota shakes up its China strategy as sales slip

The tech industry and stock markets have been trying to grasp how a small, relatively unknown Chinese company was able to develop a sophisticated artificial intelligence chatbot on par with <a href="/company/openai">OpenAI's</a> ChatGPT at a fraction of the cost.
One possible answer being floated in tech circles is distillation, an AI training method that uses bigger "teacher" models to train smaller but faster-operating "student" models.
<a href="/company/deepseek">DeepSeek</a> claims to have achieved a similar level of performance as OpenAI's o1 model at a fraction of the cost through "optimized codesign of algorithms, frameworks, and hardware."
This sparked a sharp selloff in tech shares as investors considered whether the Chinese company's low-budget approach signaled the end of the AI investment race and the dominance of US tech giants.
But questions soon arose, with some in the industry speculating that the company had piggybacked on OpenAI's developments.
Such speculation was fueled when Bloomberg reported that Microsoft and OpenAI had launched a probe into whether DeepSeek improperly obtained data from OpenAI to train its own model. OpenAI told the Financial Times on January 28 that it has seen evidence of distillation, though it did not make that evidence public.
<a href="/company/microsoft">Microsoft</a> and DeepSeek did not immediately respond to Nikkei Asia's request for comment.
Distillation itself is not a new technique and is not necessarily controversial. <a href="/company/nvidia">Nvidia's</a> Minitron and Falcon 3, which was developed by the Technology Innovation Institute in the UAE, both used the technique, likely using their own families of large language models (LLMs) as the teacher. It has become increasingly popular since 2024 amid demand from businesses wanting to utilize LLMs in their services.
Big LLM models, however, are "difficult to handle, and you would need a vast number of graphics processing units (GPUs) for its deployment," said an engineer at an AI startup in Japan.
GPUs are the main reason AI systems are so expensive. Nvidia's signature H100 chips, for example, can cost USD 30,000–35,000 each. Distillation drastically cuts development time and costs, and results in models that can operate faster than their bigger counterparts.
The issue for DeepSeek is whether its low-cost model is based more on distillation than innovation.
"There is a question on whether they're able to use existing large language models to distill their results," Kirk Boodry, an analyst at Astris Advisory Japan, told Nikkei Asia. "It seems to be coming up quite a bit in discussion. People are like, 'I don't know how much of this is really cutting edge.'"
Kazuhiro Sugiyama, consulting director at Omdia, is skeptical that DeepSeek could drastically disrupt the current AI ecosystem. Its impact is "temporary and limited," he said, pointing out that although the Chinese company's chatbot shows signs of impressive innovation, the industry still needs to verify how much it holds up.
Analysts have also questioned whether the Chinese chatbot was indeed developed with a fraction of the budget of Western counterparts.
"When people talk about [DeepSeek's] headline numbers, like a couple months of development, [or spending] USD 6 million, what they're talking about is this very specific [model]," said Boodry from Astris. "The numbers that people are throwing around are probably way too low."
The Chinese company released a paper in December 2024 which set the figure for its V3 model at USD 5.6 million. This does not include costs associated with prior research and experiments. The training cost of OpenAI's GPT-4, is estimated to exceed USD 100 million.
Sugiyama said more companies will likely enter the race to develop LLMs, but the market position of the big players, including OpenAI, will probably not change. AI models will gradually "polarize," he predicted, with big companies like Microsoft and <a href="/company/google">Google</a> continuing to invest in bigger and more powerful models to be used on their services, and smaller players developing smaller, cheaper and more efficient models that are tailored to specific markets.
Hype aside, engineers do not doubt that DeepSeek has accomplished something that deserves to be acknowledged.
Even if the company used distillation, that alone would not be enough to develop a functioning model, according to one engineer. "It will need know-how to utilize GPUs efficiently and also come up with a way to do complex training," such as combining different models to come up with a better answer.
Another AI engineer said he was "not surprised" that a company like DeepSeek suddenly appeared. "There is a big trend of reducing the size of the AI model. … Over time, there will be many ways to achieve this."
This article first appeared on <a tabindex="0" href="https://asia.nikkei.com/Business/Technology/What-is-AI-distillation-and-what-does-it-mean-for-OpenAI" rel="noopener noreferrer" data-token-index="1">Nikkei Asia</a>. It has been republished here as part of 36Kr’s ongoing <a tabindex="0" href="https://www.kr-asia.com/nikkei-teams-up-with-36kr-in-asia-tech-news-coverage/" rel="noopener noreferrer" data-token-index="3">partnership with Nikkei</a>.

What is AI distillation and what does it mean for OpenAI?

<a href="/company/alibaba-cloud">Alibaba Cloud</a> has introduced <a href="https://qwenlm.github.io/blog/qwen2.5-max/">Qwen 2.5-Max</a>, its latest artificial intelligence model, claiming it outperforms OpenAI’s GPT-4o, Meta’s Llama-3.1-405B, and DeepSeek-V3 across multiple benchmarks. The announcement comes as AI development in China gains momentum, with new players entering the space and established firms adjusting their strategies.
In recent weeks, <a href="/company/deepseek">DeepSeek</a> has gained traction with a different approach to AI. Its V3 model, introduced late last year, was reportedly trained on a budget of just USD 5.6 million, a fraction of what larger firms typically spend. Then, just before the Lunar New Year, DeepSeek followed up with R1, a model said to be on par with OpenAI’s GPT-o1.
Unlike models that depend on large-scale computing infrastructure, DeepSeek has prioritized efficiency and lower costs. This approach has <a href="https://kr-asia.com/a-new-ai-disruptor-businesses-rush-to-integrate-deepseek-r1">attracted interest</a> from businesses seeking a balance between AI performance and cost-effectiveness. Amazon Web Services and Microsoft—an OpenAI backer—have also added R1 to their model catalogs, further raising its visibility.
whereRelatedArticle
Meanwhile, Alibaba is taking a different route. With Qwen 2.5-Max, the company is focusing on both AI performance and cloud infrastructure. The model is part of a broader rollout that includes a series of upgraded cloud computing services aimed at enhancing efficiency for AI applications.
To drive adoption, Alibaba Cloud is also launching a generative AI "empowerment program," offering free cloud credits, training, and co-marketing opportunities for developers and businesses using Qwen models. In its statement, Alibaba said the goal is to support AI adoption across industries while equipping enterprises with the tools to scale their applications.
For Beijing, these developments are likely encouraging. The releases of Qwen 2.5-Max and DeepSeek’s latest models signal China’s growing role in the global AI sector. Once viewed as trailing industry leaders like OpenAI, Meta, and Google, China’s AI firms are now demonstrating their ability to compete on equal footing—and, in some cases, exceed expectations set by global counterparts.

Alibaba Cloud fires back in the AI race with Qwen 2.5-Max

Alibaba Cloud