Alibaba

ByteDance

CATL

Huawei

Meituan

Tencent

Xiaomi

Most Followed

Market News

Listings (IPO)

Earnings Reports

Venture Roundup

Big Tech

Automotives

Retail and Commerce

Collections

News

Features

Insights

Press Releases

Anta

DeepSeek

Geely

Li Auto

Mixue Bingcheng

Pop Mart

Going Global

Latest

Collaborate

Earnings

Auto

Retail & Commerce

… post-training on GLM-5.1 and GLM-5.2.
It was also the first external team in the world to complete reinforcement-learning post-training on GLM-5.1.
Notably, Mind Lab equipped both Macaron-V1-Preview and Macaron-V1 with multiple LoRA expert modules. These modules can operate independently or collaborate with one another.
The company adopted this design because it found that a single LoRA adapter could not comprehensively improve all of a model’s capabilities. Using multiple LoRA adapters on the same model, however, could substantially improve its overall performance.
Mind Lab once used 200 different datasets to train 200 separate LoRA adapters, then attached them to a single model so they could collaborate on tasks.
The company made two findings:
<ol>
<li aria-level="1">First, as the number of collaborating LoRA adapters increased, the model’s task performance grew in a log-linear relationship. The more LoRA adapters that worked together, the better the results became.</li>
<li aria-level="1">Second, a model in which all 200 LoRA adapters collaborated performed better than a model using a single LoRA adapter trained on all 200 datasets.</li>
</ol>
The collaborative model delivered an improvement of about 25%.
This showed Mind Lab that model collaboration could push beyond existing limits on intelligence.
Chen said this remains an area that requires continued exploration. Among the open questions: how many LoRA modules need to collaborate for the arrangement to be meaningful, how work should be divided among them, when collaboration is necessary, and when a single LoRA module is enough.
Competition in post-training is also accelerating in China.
Companies in the sector have different strengths, but there are still relatively few teams capable of simultaneously making LoRA-RL work at the trillion-parameter scale, building infrastructure that can manage millions of LoRA adapters, and applying continual learning in commercial settings.
<h3>From post-training to continual learning</h3>
Mind Lab’s business is currently divided into three parts:
<ol>
<li aria-level="1">The first is the continued development of infrastructure systems and frontier research into post-training.</li>
<li aria-level="1">The second is updating and iterating on Macaron, its consumer product.</li>
<li aria-level="1">The third is serving enterprise customers.</li>
</ol>
This includes deploying less expensive training infrastructure for companies such as Microsoft Azure and Huawei Cloud, as well as providing MinT and LoRA models capable of continual learning.
The Macaron app is positioned as a personal agent for everyday life.
Users can generate customized mini applications through natural language instructions.
Drawing on long-term memory and everyday conversations, Macaron assists with daily tasks and offers features such as emotional companionship.
As a consumer-facing application, Macaron also generates user interaction data that is valuable for model training. Behavior observed in real-world use can help the team better understand how users interact with models.
Mind Lab said its enterprise customers are primarily AI-enabled hardware manufacturers.
Such hardware is a natural setting for continual learning models because the data generated by each user’s interactions is almost always different.
Personalized and differentiated data is valuable for continual learning within a specific domain.
In addition to its basic model-calling services, Mind Lab plans to offer customers a continual learning option. When a customer selects the option, data associated with strong model performance will be accumulated in a dedicated LoRA adapter. The adapter will be continually trained to produce a model that is better suited to the customer’s use case and delivers greater accuracy.
In a continual learning system, LoRA can effectively produce a newly updated version every day, with each version becoming better adapted to the user’s needs and use cases.
For Mind Lab, however, commercialization is only a way to validate the practical application of its technology at this stage. It is not something the company is rushing to pursue.
Chen said Mind Lab has already reached USD 10 million in annual recurring revenue and still has considerable room to grow.
“But we do not want to put all our computing resources toward fulfilling orders,” he said. “The priority still has to be doing the research well.”
Research remains a constant part of Mind Lab’s identity. Chen said the team currently has more than 30 people working on research, including foundational infrastructure and frontier research.
“Advancing LoRA is the role we want to play,” he said. “We want to solve the final few miles after a model has completed pretraining.”
LoRA is a technical direction with many possible variations.
These include how to implement linear attention, how to extend a model’s context window, and how to improve collaboration among LoRA modules.
All of these questions still require further exploration.
KrASIA features translated and adapted content that was originally published by 36Kr. This <a href="https://www.36kr.com/p/3916202023660929">article</a> was written by Wang Xinyi for 36Kr.

Mind Lab puts continual learning to the test with Macaron-V1

Artificial Intelligence

Deep Tech

Foundation and Large Models

China

TikTok's decision to build a massive data center in Brazil demonstrates how the Latin American country has become a strategic and geopolitical focal point for digital infrastructure.
Construction of the data center is underway at the Pecem Industrial and Port Complex in the northeastern Brazilian state of Ceara. This project of TikTok's Chinese owner, <a href="/company/bytedance">ByteDance</a>, is worth more than BRL 200 billion (USD 39.2 billion).
Operations are slated to begin in 2027, and the data center will be one of ByteDance's largest outside China. Local operator TikTok Brasil calls the data center "a decisive step toward expanding our contribution to the Brazilian digital ecosystem."
Factoring into the outsize investment is the global buildout of technological infrastructure. The proliferation of <a href="/industry/ai">artificial intelligence</a> is demanding massive amounts of computing power, and major technology companies are scrambling for data centers and other resources.
The amount of computing capabilities that companies can secure has become a key indicator of competitive advantage in the AI era. In this context, Brazil is emerging as an advantageous playing field.
Renewable energy sources supply nearly 90% of Brazil's electricity. Ceara, in particular, is known as one of the leading hubs for wind power.
With the vast power demands of data centers, constraints in electricity supplies have been a limiting factor in many areas. And with growing demands to reduce carbon emissions, major tech companies have been working to tap supplies of renewable energy.
A leading Brazilian energy company will supply wind power to TikTok's data center project. The site is expected to have initial power demand of 300 megawatts, eventually reaching 1 gigawatt. The plan is to have renewable energy cover all its power needs.
Data centers are most effective when close to active users. Brazil, with its population of more than 200 million, is one of the world's leading markets for digital services and a promising staging ground for expanding into the rest of Latin America.
Brazil had more than 130 million TikTok users as of October 2025, according to Statista, ranking third by country.
Ceara also has the advantage of being close to Fortaleza, a port city that serves as a hub for undersea cables. The cables link to Europe, North America, and Africa.
The data center project also has significant implications for compliance with international authorities. In the US and Europe, worries are mounting over data sovereignty and national security. TikTok, in particular, has faced intense scrutiny over concerns regarding data transfers to China.
The US enacted a law to bar the continued operation of TikTok under Chinese controlling ownership. For TikTok's service in the US, data is now transferred within infrastructure located within the borders. A dedicated data center network is also being established in Europe.
The investment in Brazil is seen as part of a strategy to place the infrastructure for managing user data in the country where the service is provided.
The Brazilian government is supporting ByteDance's investments. In addition to developing infrastructure at the construction site, officials are offering tax incentives for data centers.
"Brazil's strengths lie in the country's high regulatory predictability and the ability to maintain favorable commercial relationships with multiple spheres," said Julio Cesar Chaves of the Getulio Vargas Foundation, a Brazilian think tank.
Even as US-China tensions persist, Brazil has adopted equidistant diplomacy that does not favor either side. This approach is said to be a factor in drawing long-term investment. In 2025, Brazil became the top country for Chinese investments.
On the other hand, the investments could become a new source of tension between the US and Latin America. While the Trump administration is working to reduce Chinese influence in Latin America, Brazil and other countries in the region are deepening economic ties with the Asian power.
Concerns remain strong in the US over Chinese companies operating critical infrastructure. China's national intelligence law has been cited as a means of pressuring companies to cooperate in Beijing's intelligence activities.
This article first appeared on <a href="https://asia.nikkei.com/business/technology/why-tiktok-is-building-a-massive-data-center-in-brazil">Nikkei Asia</a>. It has been republished here as part of 36Kr’s ongoing partnership with <a href="https://www.kr-asia.com/nikkei-teams-up-with-36kr-in-asia-tech-news-coverage/">Nikkei</a>.

Why TikTok is building a massive data center in Brazil

TikTok

East Asia

Latin America

When Chinese automotive giant <a href="/company/byd">BYD</a> launched its electric luxury car, the Denza Z9GT, in Malaysia last month, the initial reaction was one of surprise, not only at its long list of features but also at its price tag.
For the Asia Pacific debut of the model, BYD set a price of MYR 358,800 (USD 87,700.4), a level analysts expect will provide "serious competition" for luxury competitors in the market.
"I would describe it as very attractive and competitively priced, given the design and the specs. Cars competing at this price range are looking at a very serious competition," said a Kuala Lumpur-based auto analyst of an investment bank who declined to be named.
For comparison, the electric Lotus Emeya sells for around MYR 459,000 (USD 112,192), while unofficial imports of the Tesla Model S Plaid start at MYR 450,000 (USD 109,992.2).
BYD Malaysia managing director Jacob Ma, however, insists the Chinese company "never competes with anyone."
"For us, we never compete with anyone or [a] specific brand or model. We are always competing with ourselves. Every day we are trying to elevate the quality, the technology, the service to our customers," Ma told Nikkei Asia. "The customers are all we need."
"It's (Z9GT) an original concept and all the equipment embedded in this vehicle, I believe … is something brand new," Ma said, "a second generation of transportation."
The Denza Z9GT can go 701 kilometers on a single charge and is equipped with BYD's second-generation blade battery, which needs nine minutes to go from a 10% charge to 97%.
Jim Kem, content editor at Malaysian automotive news portal Carlist.my, described the price as "quite unheard of," calling the equipment list "extremely long, bordering on an overkill."
BYD is looking to build on its presence in Malaysia, where it has sold more than 35,000 EVs since entering the market in 2022. Last year it sold 14,407 EVs, making it the country's biggest seller in the segment.
The price of the new Denza is eyebrow-raising not only because of the car's features but also the promised after-sales service. The model comes with a six-year or 150,000-km vehicle warranty and an eight-year or 160,000-km high-voltage battery warranty.
"We set this price … because we believe this is what Malaysians deserve," Ma said, "a better technology at a price they can reach."
"Margin is not always the priority," he said when asked if the price would squeeze BYD's profitability. "It's always about the service quality, because we know Denza is a new brand here," he added.
BYD's warranties, especially for the battery, stood out for Daryl Foong, group CEO of Aquawalk Group, who attended the launch, saying it is "very expensive to service an EV battery when it eventually fails."
"The current generation of BYD cars are really impressive. They are really putting themselves on an even pedestal with the rest of your traditional continental cars, the BMWs," Foong said.
In terms of EV tech innovation, he said, BYD is "miles ahead of the Europeans."
Another attendee, Akmal Nizam, said the price of the Z9GT would only fetch the midrange Mercedes-Benz E-class and BMW 5 series, which do not have the BYD car's fast-charging capacity or its bells and whistles. "My EV BMW iX doesn't have a fridge … unlike the Z9GT," Akmal said. The BMX iX retails at over MYR 500,000 (USD 122,213.5).
whereRelatedArticle
A major question for BYD is whether it can parlay innovation into brand recognition.
"The luxury [segment] is widely considered to be one of the hardest challenges in the auto business," the auto analyst said. "Obstacles include insufficient brand prestige, the cheap product stereotype associated with China, plus political risks, delocalization pressures, tariffs and regulations."
Kem of Carlist.my believes BYD will have a hard time matching the mystique, prestige, and brand perception of more established luxury nameplates. "These kinds of buyers are more attracted to the more intangible aspects of a luxury car—the romanticism and aspirational qualities of a marque's story … over flashy gadgetry."
Still, BYD is expected to do well in Malaysia and elsewhere in Asia, "particularly with the Chinese diaspora, which has shared language, shared cultural heritage and decades of close bilateral trade ties [with China]," the analyst said.
BYD is also expected to get a boost from China being viewed more favorably than the US in many countries, even if the automaker is not competing directly with American brands.
In a Pew Research survey released on July 15, respondents in 25 of 36 countries viewed China more positively than the US, the first time China gained the majority. Pew polled more than 42,000 people in 36 countries.
In Malaysia, 75% of respondents viewing China positively against 19% viewing the US favorably. Only Pakistan, at 90%, had a more positive view of China.
"This will definitely help BYD," the analyst said. "A country's image acts as an 'umbrella brand' that deeply influences consumer perception and brand power."
This article first appeared on <a href="https://asia.nikkei.com/business/automobiles/byd-takes-aim-at-malaysia-s-luxury-ev-segment">Nikkei Asia</a>. It has been republished here as part of 36Kr’s ongoing partnership with <a href="https://www.kr-asia.com/nikkei-teams-up-with-36kr-in-asia-tech-news-coverage/">Nikkei</a>.
Note: MYR figures are converted to USD at rates of MYR 4.09 = USD 1 based on estimates as of July 27, 2026, unless otherwise stated. USD conversions are presented for ease of reference and may not fully match prevailing exchange rates.

BYD takes aim at Malaysia's luxury EV segment

Transportation

Electric Vehicles

Mobility

… process moves forward, Shein is expected to accelerate its investments in technological innovation, decarbonization, supplier training, and other corporate social responsibility initiatives.
This also aligns with Guangdong’s 15th five-year plan outline, which calls for stronger efforts to promote the digital and intelligent transformation of industries, as well as their transition toward greener operations.
Shein has set a target of achieving net zero emissions by 2050, which has been approved by the Science Based Targets initiative. Using 2023 as its base year, Shein aims to reduce absolute greenhouse gas emissions across its entire value chain by 90% by 2050.
To support industrywide coordination, Shein has also worked with specialist third-party organizations to provide suppliers with comprehensive assistance, from greenhouse gas inventories and target-setting to transition planning.
Green decarbonization projects involving rooftop solar installations and energy efficiency improvements are also continuing across Shein’s supplier network.
Taken together, these factors suggest that Shein can no longer be assessed solely through the valuation framework applied to traditional apparel retailers.
According to data provided by China Insights Consultancy, the global fashion industry was valued at USD 1.7 trillion in 2025 and is expected to grow to USD 2 trillion by 2030. The online fashion market is projected to reach USD 792 billion by 2030 as online penetration continues to increase.
KrASIA features translated and adapted content that was originally published by 36Kr. This <a href="https://36kr.com/p/3917513813650824">article</a> was written by 36Kr Caijing.
Note: RMB figures are converted to USD at rates of RMB 6.77 = USD 1 based on estimates as of July 30, 2026, unless otherwise stated. USD conversions are presented for ease of reference and may not fully match prevailing exchange rates.

Shein’s IPO case rests on the technology behind its fashion business

Shein

Fashion

Creative

Retail

Consumer Economy

Hong Kong

… attempting to respond to the desire to feel accompanied.
The move may appear to be a departure from Cylingo’s existing business, but it is aimed at a more fundamental question: as society becomes increasingly atomized and emotional gaps widen within families, can a physical device provide a form of presence that is tangible, meaningful, and sustainable?
KrASIA features translated and adapted content that was originally published by 36Kr. This <a href="https://36kr.com/p/3904049563731849">article</a> was written by Huang Nan for 36Kr.

Why Cylingo is moving into home robotics after building a 60 million-user app

Robotics

Embodied Intelligence

… models designed for factory use.
As robotics technology enters the commercial validation stage, 2026 is shaping up as a key deployment phase.
Data from China’s Ministry of Industry and Information Technology showed that the country produced about 20,000 humanoid robots in 2025. Production exceeded 40,000 units in the first half of 2026 and is expected to surpass 100,000 units for the full year.
Chinese government policy is also continuing to support the sector. Special plans for 2026 issued by the Ministry of Industry and Information Technology and the State-owned Assets Supervision and Administration Commission of the State Council call for the establishment of capacity to deploy humanoid robots at the 10,000-unit scale by the end of the year.
Following the news about humanoid robots, BYD’s A-shares rose against the broader market on July 28. The stock gained more than 1% to close at RMB 93.35 (USD 13.8) per share, even as the ChiNext Index declined and more than 2,700 stocks fell across the Shanghai and Shenzhen markets. BYD shares rose more than 2% at one point during the trading session.
This <a href="https://xueqiu.com/2854806290/402661151">article</a> was adapted based on a feature originally written by SY and published on IPO Zaozhidao. KrASIA is authorized to translate, adapt, and publish its contents.
Note: RMB figures are converted to USD at rates of RMB 6.77 = USD 1 based on estimates as of July 30, 2026, unless otherwise stated. USD conversions are presented for ease of reference and may not fully match prevailing exchange rates.

BYD’s humanoid robot to begin work at D Space facility in August

Chinese <a href="/industry/ai">artificial intelligence</a> developer Moonshot AI has raised USD 3.5 billion in its latest funding round, lifting its valuation to USD 35 billion, as first reported by <a href="https://www.bloomberg.com/news/articles/2026-07-29/china-s-moonshot-ai-passes-funding-goal-to-hit-35-billion-value">Bloomberg</a>.
Moonshot entered the round at a pre-money valuation of USD 31.5 billion and initially sought to raise USD 1–2 billion, according to reports. China’s National AI Industry Investment Fund was said to be among the lead investors.
The company has also begun planning a pre-IPO round and approaching potential investors. It is seeking to complete the round at a pre-money valuation of USD 50 billion.
Moonshot could list in Hong Kong as early as this year. It began preparing for a Hong Kong IPO early this year, with China International Capital Corporation and Goldman Sachs serving as joint underwriters.
The company reportedly notified shareholders in May that it had begun dismantling its variable interest entity (VIE) and red-chip structures, removing potential obstacles to a Hong Kong listing. It may have targeted July for completion of the process.
Moonshot has a diverse shareholder base. Its investors include Chinese internet companies <a href="/company/alibaba">Alibaba Group</a>, <a href="/company/tencent">Tencent</a>, and <a href="/company/meituan">Meituan</a>; venture capital firms HSG, IDG Capital, Capital Today, 5Y Capital, Gaorong Ventures, and YF Capital; and strategic investors such as China Mobile, Andon Health, and 37 Interactive Entertainment.
At the USD 50 billion pre-money valuation it is seeking for the pre-IPO round, Moonshot would be valued at more than ten times the USD 4.3 billion recorded during its Series C round in 2025.
The higher valuation has coincided with <a href="https://kr-asia.com/moonshot-ai-sees-overseas-revenue-surge-as-kimi-k2-5-gains-traction-abroad">rapid revenue growth</a>. Moonshot’s annual recurring revenue had exceeded USD 300 million by July, while API sales accounted for more than 70% of total revenue and had become its main growth driver.
Moonshot released its latest large model, Kimi K3, on July 17. The launch attracted attention across the global technology sector and may have contributed to the company’s higher valuation expectations.
Daily sales increased sixfold after the model’s release, according to reports. As demand exceeded expectations and placed greater pressure on Kimi’s computing capacity, Moonshot announced on July 19 that it would immediately suspend subscriptions for new consumer users. It released Kimi K3 as an open-source model on July 27.
According to Moonshot, Kimi K3 has 2.8 trillion parameters, native visual understanding, and a context window of up to one million tokens. The company described it as the world’s first open-source model in the three-trillion-parameter class and said it was designed for advanced artificial intelligence applications, including long-horizon coding, knowledge work, and reasoning.
Moonshot said Kimi K3’s overall performance still trailed the strongest proprietary models, Claude Fable 5 and GPT-5.6 Sol. However, it said the model demonstrated frontier-level capabilities across its evaluation suite and outperformed every other model tested.
This <a href="https://xueqiu.com/2854806290/402739157">article</a> was adapted based on a feature originally written by SY and published on IPO Zaozhidao. KrASIA is authorized to translate, adapt, and publish its contents.

Moonshot AI targets USD 50 billion valuation ahead of Hong Kong IPO

Moonshot AI

… quickly.
In late June, Volcano Engine presented Seedance 2.5 at its Force Conference. The company said the model could generate 4K video and offered improved control over staging and visual detail.
The model was scheduled to launch later in July.
whereRelatedArticle
<h3>After the novelty</h3>
Whether the early success of AI filmmakers can continue is uncertain.
Wang Lei, dean of the Communication University of China’s School of Animation and Digital Arts, said the storytelling, staging, and editing in the viral Zombie Scavenger were “below average” by the standards of graduate film projects.
He believes much of the attention surrounding AI filmmakers comes from the public’s “shock.” Once that feeling passes, he said, the weaknesses will become clearer, and few creators or works will endure.
Hell Grind, a 95-minute feature released in theaters and described by its producers as the first film of its kind, was made by a San Francisco startup with sponsorship from Seedance.
Viewers praised its technical advances but criticized its thin story and limited artistic ambition.
Acclaimed director Jia Zhangke’s first AI work, Wheat Harvest, made with Kling AI, also received a wave of negative reviews.
The names of models and directors may attract an audience once. They cannot make the audience care.
When the surprise fades and the technology becomes widely available, the familiar question will return: what kind of work will people love, respect, and remember?
KrASIA features translated and adapted content that was originally published by 36Kr. This <a href="https://36kr.com/p/3898151587890824">article</a> was written by Lan Jie for 36Kr.
Note: KRW, RMB figures are converted to USD at rates of KRW 1448.68 = USD 1 and RMB 6.77 = USD 1 based on estimates as of July 30, 2026, unless otherwise stated. USD conversions are presented for ease of reference and may not fully match prevailing exchange rates.

Inspired by viral hits, aspiring filmmakers are using AI to break into the industry

Entertainment

Short Video

Video

Content

Creator Economy

… Ratings. "In this capacity, one of their goals is to support local economic development by attracting businesses and IPO candidates to establish operations in their cities and help raise tax revenue and create employment opportunities."
Other local governments looking to copy the Hefei model face an uphill battle. One major challenge is identifying quality candidates, as hundreds of municipalities often target similar high-tech industries.
"The supply is greater than demand because there are only a limited number of companies that are worth investing in," said a venture capitalist in China who spoke on the condition of anonymity. "It's difficult for a company to move to your city. So you can only attract second-tier projects."
Building industries like semiconductors, which involves a complex supply chain, can also extend far beyond the term of a local party leader. Hefei often supports portfolio companies for years after making an investment, such as by facilitating collaboration with experts from the Chinese Academy of Sciences for a nuclear fusion project, according to Fitch's Chen.
"The key is whether companies and the government have the right people to evaluate the risk, and to make measured investment decisions," said Davis Sun, senior director at Fitch. "Not every government has this asset."
Uncertainty surrounding electric appliances maker <a href="/company/dreame">Dreame</a> is shaping up to be a cautionary tale for local officials chasing startups.
Affiliates of the Suzhou-headquartered company incubated dozens of new ventures, ranging from electric motorbikes to robots to pool cleaners. These ventures, in turn, raised money from investment funds managed by Dreame but backed by local governments across the country, from Xiamen in the south to Jiaxing in the east and Dazhou in the west, according to corporate database Tianyancha. Some of these projects were heavily marketed but had little track record of actual sales.
Then, in early June, Dreame's CEO Yu Hao was barred from posting on social media platform Weibo, raising speculation that the company was under regulatory pressure.
Beijing has not publicly singled out Dreame. But on the same day as Yu's Weibo ban, the State Council published "guiding opinions" on the private fund market. The document outlined strict registration procedures for local governments when they establish new investment funds.
"In principle, no new funds may be established at the county or district level," according to the document. A spokesperson for China's securities regulator said "some state-owned funds have deviated from their intended functional positioning."
Following reports that some local governments have begun to scale back their support for Dreame-affiliated ventures, Yu told financial news outlet Caixin that the company will focus on four core areas. Dreame did not respond to a request for comment.
Sun said policymakers appear to be "targeting overcompetition between governments and reducing repetitive investments in the same sector." The guidance also vows to crack down on misuse of government funds, such as "fake equity, real debt" structures that involve clauses requiring a startup to buy back an investor's shares at a later date.
To mitigate risk, Tsang at S&amp;P expects LGFVs to take minority stakes or co-invest with more senior-level city funds. Despite the turbulence, industry observers expect that state involvement in funding early-stage Chinese startups will continue to deepen.
The tech rivalry between China and the US means both Chinese startups and US venture capital funds face intense scrutiny from their respective local regulators. Dollar-based funds accounted for just 5% of overall fundraising activity in China in the first half of the year, according to the industry association.
In April, the National Development and Reform Commission blocked US tech giant Meta's <a title="" href="https://asia.nikkei.com/business/technology/china-blocks-meta-s-2bn-acquisition-of-manus-after-security-review" target="_blank" rel="noopener">acquisition of Manus</a>, an AI company founded in China but later relocated to Singapore, indicating that the practice of moving a company and its assets overseas to raise funds is now subject to national security reviews.
"Startups in strategic industries still need to continue to rely on funding from local governments," said the Chinese venture capitalist. "There are a lot of startups in industries like space and nuclear energy, but foreign funds won't be able to invest due to restrictions."
This article first appeared on <a href="https://asia.nikkei.com/spotlight/policy-asia/hefei-s-cxmt-jackpot-fuels-china-local-governments-tech-bets">Nikkei Asia</a>. It has been republished here as part of 36Kr’s ongoing partnership with <a href="https://www.kr-asia.com/nikkei-teams-up-with-36kr-in-asia-tech-news-coverage/">Nikkei</a>.
Note: RMB figures are converted to USD at rates of RMB 6.78 = USD 1 based on estimates as of July 27, 2026, unless otherwise stated. USD conversions are presented for ease of reference and may not fully match prevailing exchange rates.

Hefei's CXMT jackpot fuels China local governments' tech bets

Investment