Know Asia's
Business changes,before it hits your market.

China's Alibaba has released an AI model that its research arm Damo Academy claims is more sensitive than radiologists at spotting early-stage colorectal cancer from CT scans, the latest sign that Chinese hyperscalers are building deeper into healthcare AI.

The model is called Coca. In a peer-reviewed paper published last week in the Annals of Oncology, Damo Academy reported that Coca correctly flagged five previously missed colorectal cancer cases in a cohort of more than 27,000 patients. Sensitivity hit 86.6%, specificity 99.8%.

What the test actually showed

Coca outperformed a group of 10 radiologists with varying levels of experience by 20.4% on sensitivity, the metric that measures how reliably a system identifies people who actually have the disease.

Sensitivity matters because the alternative diagnostic methods, colonoscopy and CT colonography, are invasive and uncomfortable, which is why screening compliance in the broader population is poor. A non-invasive CT-based pre-screen that does not miss disease changes that economics.

Coca was developed by Damo Academy in collaboration with Chinese institutions including Guangdong General Hospital, the paper specifies.

Why Alibaba keeps doubling down on healthcare AI

Damo Academy has been pushing into medical-imaging AI for years, with a particular focus on cancer detection. The new release fits a pattern: target diagnostic problems where the bottleneck is not data but interpretation, and where the productivity case is direct (one model can do the work of multiple radiologists at scale).

The same logic explains why Tencent and Baidu are building parallel medical-AI tracks, and why Beijing's industrial policy treats medical AI as one of the priority application sectors for domestic compute.

The wider picture

China's hospital system handles enormous volumes (Guangdong General alone runs millions of patient visits per year) and is structurally short on senior radiologists. AI models trained on that scale of imaging data have a structural advantage over Western peers working with smaller, more siloed datasets.

Coca is not yet a regulated diagnostic device. The Annals of Oncology publication is a research result, not market clearance. But it shifts the conversation from "can Chinese hyperscalers reach frontier AI" to "in which verticals will Chinese hyperscalers ship product first." Healthcare imaging is now firmly on that list.

Sources: SCMP, Annals of Oncology

Sources: SCMP, Annals of Oncology

China is testing a prototype truck-mounted nuclear reactor capable of 10 megawatts, enough to power a medium-sized AI data center, according to Wu Yican, chief scientific adviser at the Institute of Nuclear Energy Safety Technology and a member of the Chinese Academy of Sciences.

Wu, speaking to Science and Technology Daily, described the unit as the world's first 10-megawatt vehicle-mounted nuclear power system. The team is now actively looking for first deployments.

The Details

Output: 10 MW, sufficient for a medium-sized AI data center

Form factor: vehicle-mounted, transportable on a truck chassis

Operational lifespan: "decades without recharging," per Wu

Status: prototype, in testing, multi-year development behind it

Targeted use cases: remote regions and islands, emergency backup power in special environments, ship propulsion, space-system power, and AI computing or data centers

Why this is more than a science press release

The framing matters. Wu calls the system a "nuclear power bank" and pitches it explicitly as a solution to "battery anxiety" in different applications. That language is built for industrial customers, not for scientific peers.

It also lands at exactly the moment when the AI infrastructure conversation is dominated by the energy bottleneck. US hyperscalers are talking about gigawatt-scale data center campuses and are paying premium prices for power purchase agreements, with grid interconnection becoming the binding constraint. Microsoft and Amazon have written multi-billion-dollar checks for nuclear PPAs (Three Mile Island restart, X-Energy SMRs).

China's bet here is structurally different. Instead of waiting for utility-scale SMR builds, the pitch is a self-contained 10 MW box that can be parked next to a data center and run for decades. If the prototype scales, the unit cost and time-to-power look very different from a multi-year SMR construction project.

What is unproven

Important caveats: the announcement is a single source (SCMP, citing Science and Technology Daily) and reflects a research-stage prototype, not a commercial product. There is no third-party validation of safety claims yet, and no public regulatory framework for siting truck-mounted reactors at commercial data centers in any major jurisdiction.

What it does signal: China is willing to position nuclear-on-wheels as a serious answer to the AI energy problem, while the US debate is still anchored on grid expansion and stationary SMRs. If even one of the targeted use cases (islands, ships, remote AI compute) lands a paying customer, the sector reframes.

Sources: SCMP, Science and Technology Daily

Sources: SCMP, Science and Technology Daily

Japan's SoftBank Group is preparing to create and list a standalone AI and robotics company in the US, according to Financial Times reporting confirmed by CNBC and DealStreet Asia. The new entity is named Roze, and SoftBank executives are targeting a valuation of about $100 billion and an IPO as early as 2026.

Founder and CEO Masayoshi Son is driving the project. The plan is ambitious enough that the FT note flagged "uncertainties stemming from the conflict in the Middle East" as the main execution risk.

What goes into Roze

Roze is meant to bundle three buckets of SoftBank assets and pitch them as one operating story.

ABB Robotics, the Swiss-Swedish industrial robotics business that SoftBank agreed to buy from ABB last year. ABB Robotics is one of the world's largest suppliers of industrial robotics and machine automation.

Existing land, energy, and infrastructure assets from SoftBank's portfolio, the kind of inputs needed to build large-scale data centers fast.

Integration with SoftBank's broader AI hardware and software stack. The pitch is that Roze can build AI infrastructure faster and at lower cost by using robotics on the construction side.

Why Son needs the listing

SoftBank has made very large AI commitments. The firm has pledged more than $30 billion to OpenAI alone. SoftBank is also the lead financier of Stargate, a planned $500 billion US AI data center build-out announced jointly with OpenAI, Oracle, and others. The conglomerate is already constructing its own large-scale data center in Ohio.

A $100 billion IPO of Roze, even at half the targeted valuation, would generate the float Son needs to keep funding those commitments without leaning entirely on Vision Fund returns.

Vision Fund posted a $2.4 billion gain in the December quarter, helped by markup on the OpenAI position. SoftBank shares are up over 18% in 2026 and gave back 0.9% on Thursday after the FT report.

The market context

OpenAI itself is preparing for a potential IPO that bankers have penciled in at a $1 trillion valuation. If Roze prices anywhere near $100 billion, the spinout would arrive as the largest pure-play AI infrastructure listing in the US, ahead of CoreWeave and any of the smaller hyperscaler enablers.

The combination Son is selling is not just "AI infrastructure." It is "AI infrastructure built by robots." If the Roze story holds up to S-1 scrutiny later this year, that frame will reset how the next layer of AI infra companies position themselves.

Sources: CNBC, DealStreet Asia, Financial Times

Sources: CNBC, DealStreet Asia, Financial Times

South Korea's LG Energy Solution has secured a battery supply contract with BMW worth roughly 10 trillion won (~$7 billion), according to Chosun. It is the first major BMW supply deal for LGES.

BMW had previously sourced its EV cells primarily from China's CATL and Korea's Samsung SDI, with Northvolt as the planned European in-house alternative until that project's collapse. The LGES contract reshapes the supplier mix.

The Details

Order size: ~10 trillion won (~$7 billion)

Customer: BMW Group

Status: First major BMW supply contract for LGES

Source: Chosun report (other Korean and global outlets had not picked up the story by the time of writing, so finer details on tenor, cell chemistry, and delivery start are still pending confirmation)

Why this matters for the Korean cell makers

LGES has spent the last 24 months losing margin to CATL on the cost curve and losing share to Samsung SDI on premium European platforms. A direct BMW win, at this size, gives LGES a foothold in the German premium pipeline that pairs with its existing GM and Stellantis exposure in North America.

For BMW, the deal is strategic insurance. Northvolt's collapse left a structural hole in the European cell pipeline, and Beijing's grip on CATL's licensed-tech model in the US means BMW has limited room to scale Chinese cell sourcing inside Western trade frameworks. A Korean major partly fills both gaps.

What to watch next

If the Chosun number holds up, the deal slots LGES alongside CATL as one of the two structural winners of BMW's next-generation Neue Klasse platform, which the carmaker is using to anchor its 2026-2030 EV plan.

Korean and German trade officials have spent the spring talking about "strategic battery cooperation." This is the first contract of that size to actually land. The next data point will be confirmation from BMW's side, which had not been issued at time of writing.

Sources: Chosun

Sources: Chosun

On Thursday morning, the yen slid past 160 per dollar, and the yield on Japanese government bonds hit the highest level in nearly three decades. Hours later, the Bank of Japan and the Ministry of Finance moved in, buying yen and selling dollars. The currency snapped back from 160 to 155.5 in hours.

According to a Bloomberg analysis of central bank accounts, Tokyo likely spent around 5.4 trillion yen ($34.5 billion) on the day. Nikkei's read of BOJ money-market data put the figure closer to 5 trillion yen ($32 billion). It is Japan's first foreign exchange intervention since July 2024.

Why the yen broke

Three pressures lined up at once.

The Fed-BOJ rate gap is wide and not closing. The Federal Reserve held rates and warned about persistent inflation, citing high oil prices on the back of the Iran war. The market priced out near-term Fed cuts.

The BOJ talked tough on Tuesday, then did nothing. At its monetary policy meeting, the central bank signaled that hikes were nearing, but kept the policy rate unchanged. The market read it as bluff.

Speculators piled in. CFTC data showed short-yen positioning at the largest level since July 2024, around $7.5 billion notional, ahead of Japan's Golden Week holidays.

Katayama's phone warning

Finance Minister Satsuki Katayama issued a verbal warning about the yen's weakness on Thursday, telling reporters that "decisive action" was nearing. After the intervention took effect, she added an unusual instruction. "Please keep your smartphones with you at all times, whether you're going out or resting," she said, ahead of the long holiday week.

Top FX diplomat Atsushi Mimura backed the message on Friday. "I won't comment on what we'll do ahead. But I will tell you that Japan's Golden Week holidays have just started," he said. The yen jumped again on his comments, briefly touching 155.60.

The structural problem

FX intervention treats the symptom, not the cause. The cause is the rate gap. Until the BOJ actually moves, the carry trade rebuilds.

"The yen will remain under downward pressure on inflation concerns from high oil prices, slow BOJ rate hikes and the hawkish tone of other central banks," said Rinto Maruyama, FX and rates strategist at SMBC Nikko Securities, after the intervention.

Japan last burned through real reserves to defend the yen in July 2024, when the currency hit a 38-year low at 161.96. The Golden Week stretch through next Wednesday is the next pressure test.

Sources: Bloomberg, Nikkei, Reuters, Business Times SG

Sources: Bloomberg, Nikkei, Reuters, Business Times SG

China's Cambricon Technologies, the Beijing-based AI chip maker often called "China's little Nvidia," reported on Wednesday that Q1 net profit climbed 185% year on year to 1 billion yuan ($146 million), with revenue up 160% to 2.9 billion yuan.

The market reacted hard. Shares jumped the maximum 20% daily limit on Thursday to close at CNY 1,699.96, an all-time high, beating the previous peak of CNY 1,595.88 set on August 27, 2025. Market capitalization now stands at 716.8 billion yuan ($104.9 billion).

The Details

Net profit Q1: 1 billion yuan ($146.2 million), +185% YoY

Revenue Q1: 2.9 billion yuan, +160% YoY

Operating cash flow Q1: +834 million yuan ($121.9 million), reversing a 1.4 billion yuan outflow in the same quarter last year

Cambricon attributed the swing to faster customer collections, rising compute demand from China's AI industry, stronger product positioning, and faster rollout of AI use cases.

The company posted its first annual profit in 2025 after years of state-backed losses. Q1 2026 confirms that the operating model now stands on customer revenue, not just subsidies.

Moore Threads also turned the corner

Cambricon was not alone. Moore Threads, which listed on Shanghai's STAR Market at the end of 2025, reported a Q1 profit of 29.4 million yuan ($4.3 million), reversing a 112 million yuan loss a year earlier. Revenue was up 155% to 738 million yuan. Two of China's most-watched domestic AI chip names are profitable in the same quarter, for the first time.

Why Beijing keeps winning the funding side

The earnings come as US export curbs continue to bite. According to Bloomberg, the price of Nvidia's flagship B300 server in the Chinese gray market has been pushed to about 7 million yuan (~$1 million) per unit on tightening restrictions.

Every dollar the gray market spends on Nvidia is a dollar Cambricon, Moore Threads, Hygon, and Huawei Ascend can compete for. The Q1 numbers suggest that the substitution is no longer a thesis. It is an income statement.

Sources: Yicai Global, Business Times SG

Sources: Yicai Global, Business Times SG

South Korea's Samsung Electronics reported a Q1 group operating profit of 57.2 trillion won (~$38.5 billion) on Thursday, up 756% year on year. Group revenue hit 133.9 trillion won (~$90 billion), up 69%. Both are records for a single quarter, and both came in well above analyst consensus.

The number that mattered most: the Device Solutions semiconductor arm posted 53.7 trillion won (~$36 billion) in operating profit, almost 48 times the same quarter last year, when the same division earned about 1 trillion won.

What is driving the print

The AI data center boom has constrained the supply of memory chips and pushed prices higher across HBM, server DRAM, and DDR5. Samsung's chip business is the biggest single beneficiary inside the Korean tech complex.

Chip-division revenue alone reached 81.7 trillion won in Q1, up 225% year on year. By contrast, Samsung's Device eXperience arm, which sells smartphones, TVs, and home appliances, posted 52.7 trillion won in revenue and just 3 trillion won in profit. Display added another 6.7 trillion won in revenue and Harman 3.8 trillion won.

The chip division now produces well over 90% of group operating profit, a structural shift compared with even three years ago, when the company's mobile and consumer-electronics businesses still carried meaningful weight.

The HBM gap to SK Hynix is closing

Samsung said in February it had become the first chipmaker in the world to mass-produce HBM4 memory chips and ship them to unnamed customers. SK Hynix held a 57% revenue share in the global HBM market in Q4 2025 and has been the more profitable of the two on AI memory.

Samsung's earnings release made clear it sees no slowdown. Server memory demand is expected to remain strong into the second half as hyperscalers continue to build for AI adoption and as agentic AI use cases scale, the company said.

The risk on the other side

Higher memory prices are a tax on Samsung's own smartphone and home-appliance businesses, which buy the chips. The Middle East conflict adds pressure on Samsung's raw-material and energy supply chain, the company flagged.

Samsung shares are up about 90% in 2026 and gave back early gains on the day of the print, ending 0.8% lower. The market has spent the year buying the AI-memory thesis. The Q1 number just hardened it.

Sources: CNBC, Korea Herald, Bloomberg, Business Times SG

Sources: CNBC, Korea Herald, Bloomberg, Business Times SG

China's State Grid Corp, the country's largest electricity utility, has earmarked 6.8 billion yuan ($994.7 million) in 2026 to procure 8,500 robots for its grid operations, according to internal documents reported by Humanoid Daily.

It is one of the largest single industrial deployments of robotics announced anywhere this cycle and the clearest sign yet that China's humanoid sector is moving from demos to paying customers.

The Details

Procurement breaks into a mixed-fleet model:

Quadruped "robot dogs": 5,000 units, 1.5 billion yuan budget, deployed for autonomous inspections of substations and mountainous terrain.

Humanoid robots: 500 units, 2.5 billion yuan budget, focused on high-risk live-line work.

Other categories make up the remaining 3,000 units.

Named core suppliers: Unitree, AgiBot, and UBTech, three of the most prominent Chinese humanoid platforms. The 5 million yuan ($690,000) average price per humanoid is far above any of the consumer-facing units these companies have shown publicly, reflecting the heavy specialization required for live-line and substation work.

Why this is the first 10-billion-yuan vertical

State Grid's 2026 Embodied Intelligence Development Plan explicitly targets an 80% penetration rate for autonomous intelligent agents in high-risk grid scenarios by 2027.

The deployment will trigger follow-on orders from Southern Power Grid and other regional utilities, lifting the broader Chinese energy sector to over 10 billion yuan in robotics spending in 2026, the first ten-billion-yuan vertical market for physical AI, according to industry analysts cited by Humanoid Daily.

The contrast with Western peers

US humanoid news this month has been dominated by financing rounds and management hires (Apptronik built a "dream team" of executives, Meta acquired Assured Robot Intelligence). China's news is dominated by purchase orders.

Pudu Robotics, EngineAI, and Galaxea between them raised over $640 million in private capital in April. Now State Grid puts almost a billion dollars on top of that pile, but as revenue, not as venture funding. That is a different signal for the supply chain.

Sources: Humanoid Daily

Sources: Humanoid Daily

Last December, Meta announced a $2 billion acquisition of Manus, a Singapore-headquartered AI agent startup founded by Chinese engineer Xiao Hong. On Monday, four months later, the foreign investment security review office under China's National Development and Reform Commission ordered both sides to withdraw the deal.

It is the first known use of the foreign investment security review mechanism introduced in late 2020. Meta has indicated it will comply, according to multiple reports.

Why this matters now

Manus was launched in March 2025 and was hailed by Chinese state media as "the next DeepSeek." Its co-founders, CEO Xiao Hong and chief scientist Ji Yichao, raised a $75 million Series B led by Benchmark Capital in April 2025 at a $500 million valuation. In June, the company relocated its HQ to Singapore via a new entity called Butterfly Effect, then laid off 80 of its 120 China-based core technical staff.

Meta announced the acquisition in late December after only a few weeks of due diligence. Neither Meta nor Manus sought Chinese regulatory approval for the deal or the relocation, according to five sources cited by Reuters.

Why "Singapore washing" did not work

Beijing's framing is that Manus's technology, talent, and data remain Chinese, regardless of where the holding company sits. According to Bloomberg and CNBC, the decision to block the deal was elevated beyond economic regulators to China's National Security Commission, the body chaired by Xi Jinping.

Xiao Hong and Ji Yichao have reportedly been barred from leaving China since March, after being summoned to Beijing for talks with regulators. China's state-backed Global Times wrote that the issue is not where Manus is registered, but "the extent of its technological, talent and data links with China."

The mechanical problem

The acquisition is already done. Manus engineers moved into Meta's Singapore offices and were granted Meta corporate accounts. Meta and Manus held joint working sessions for months, according to a Meta employee cited by SCMP. Several lawyers told the Hong Kong paper that unwinding the integration will be "time-consuming, complex, and difficult."

Han Shen Lin, China country director at The Asia Group in Shanghai, summed up the read in plain terms: "Beijing effectively drew a bright red line that Chinese AI talent and technology are not for sale to American companies, full stop."

The next test sits on the calendar. Trump is scheduled to visit Beijing in mid-May. Mark Zuckerberg may try to put the Manus case on that agenda. Beijing has just signaled how that conversation will go.

Sources: Yicai Global, CNBC, SCMP, DealStreet Asia, Bloomberg

Sources: Yicai Global, CNBC, SCMP, DealStreet Asia, Bloomberg

China's CATL, the world's largest EV battery maker, sold 62.4 million new H-shares at HK$628.20 each on Tuesday, raising HK$39.2 billion ($5 billion).

It is the largest equity offering in Hong Kong this year and the second largest globally in 2026. The price marked the bottom of the marketed range and a 7% discount to Monday's close of HK$675.50.

The Details

The order book filled within an hour of launch, with more than 150 institutions participating, including hedge funds, sovereign wealth funds, and existing CATL shareholders.

Use of proceeds: global factory expansion, the development of CATL's zero-carbon business, R&D, and general working capital, according to the stock-exchange filing.

Despite the strong demand, the stock dropped on the dilution. CATL's Hong Kong shares opened down 6.7% on Tuesday and fell as much as 7.6% to HK$624 by midday. CNBC's headline put the move at over 8%. The Hong Kong listing has still risen about 137% from the HK$263 IPO price in May 2025.

Why CATL keeps tapping HK

CATL held a 38.1% share of the global EV battery market in the first ten months of 2025, according to South Korean research firm SNE Research, ranking it number one worldwide. Its customer list includes Tesla, BMW, Volkswagen, Xiaomi, and Nio.

The placement is the biggest single capital raise in Hong Kong since CATL's own $5.25 billion listing in May 2025. Last week, a Sinopec unit sold 8.5 million CATL Hong Kong shares for about $770 million, cashing in on the rally.

H-shares are trading at roughly a 35% premium to the A-shares with very low free float, said Dickie Wong of uSMART Securities. "The placement should help improve liquidity and attract more long-term international investors and index funds," he added, while flagging that valuations look rich.

The macro tailwind

The fundraising rides a separate wave. Chinese exports of solar products, batteries, and electric vehicles hit record highs in March, according to energy think tank Ember, as the Iran war oil shock accelerates the global pivot to clean energy. CATL is the dominant supplier on that supply chain.

Winston Ma, executive director of the Global Public Investment Funds Forum, called the timing "a perfect wave" of surging stock price, supply-side fossil-fuel shock, and a Hong Kong market hungry again for heavyweight tech leadership.

Sources: Nikkei, Caixin, CnEVPost, DealStreet Asia, CNBC, Business Times SG

Sources: Nikkei, Caixin, CnEVPost, DealStreet Asia, CNBC, Business Times SG