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

The bottleneck for running large AI models has never been processing power alone. It has been memory, specifically how fast data can move between storage and the processor doing the work. South Korea's SK hynix just started mass-producing a memory module designed to attack that problem directly.

The product is a 192GB SOCAMM2 module built on SK hynix's sixth-generation 10nm-class LPDDR5X DRAM. The target platform: Nvidia's Vera Rubin, the successor to the Blackwell architecture that currently powers most of the world's AI data centers.

The specs

The SOCAMM2 form factor was built specifically for AI accelerators, not adapted from existing server memory. SK hynix says it delivers more than double the bandwidth of conventional RDIMM modules and over 75% better power efficiency. For data center operators paying enormous electricity bills to run AI workloads, that efficiency gain translates directly into lower cost per inference.

Kim Joo-sun, president of SK hynix's AI infrastructure division, put it plainly: "192GB SOCAMM2 sets a new standard for AI memory performance."

Why it matters

SK hynix is the world's second-largest memory chipmaker and already Nvidia's primary supplier of HBM (High Bandwidth Memory), the specialized chips stacked inside every H100 and B200 GPU. That relationship has turned SK hynix into one of the biggest beneficiaries of the AI boom, with its stock price roughly tripling since early 2023.

Moving into SOCAMM2 production for Vera Rubin deepens the dependency in both directions. Nvidia needs SK hynix to deliver enough memory at the right specs. SK hynix needs Nvidia to keep winning the AI accelerator market. For now, both sides of that bet look solid.

Sources: Yonhap News

China's central bank kept its benchmark lending rates unchanged on Monday, the 11th straight month without a move. The one-year loan prime rate stays at 3.0%, the five-year mortgage rate at 3.5%.

The Details

The decision comes after Q1 GDP hit 5.0%, accelerating from 4.5% in the prior quarter and landing at the top of Beijing's 4.5-5.0% target range. That kind of growth removes any urgency to cut rates.

Factory-gate prices: Turned positive for the first time in over three years, climbing 0.5% in March. The deflation scare that dominated 2024 and early 2025 is fading.

Consumer inflation: Jumped to 1.3% in February (biggest increase in three years) before easing to 1% in March.

Energy costs: Surging global oil prices from the Iran conflict are pushing up import costs, giving the PBOC even less incentive to ease.

UBS chief China economist Yu Song expects a "wait-and-see" approach. "Rising inflation reduces the PBOC's incentive to cut policy rates. The government may also need time to assess the impact of external uncertainties amid the Middle East conflict."

Why 5% changes the conversation

Beijing set 4.5-5.0% as its growth target for 2026, the least ambitious since the 1990s. Hitting the top end in Q1 means policymakers can afford to hold rates steady while inflation ticks up and the Iran war clouds the outlook.

The PBOC signaled it would maintain a "supportive" and "moderately loose" monetary stance this year. Central bank governor Pan Gongsheng warned at the IMF meeting last week that geopolitical tensions and trade barriers are weighing on global growth and urged deeper international policy coordination.

Finance minister Lan Fo'an reiterated Beijing's push to expand domestic demand and boost consumption. The subtext: China wants to grow through spending at home rather than relying on exports that are slowing under pressure from energy disruptions and trade friction.

For markets, the hold means no fresh stimulus catalyst in the near term. The next move likely depends on whether Q2 growth holds above 4.5% and whether oil prices stabilize after the Hormuz ceasefire.

Sources: CNBC, PBOC

For years, China was where multinational pharma companies went to manufacture cheaply. The molecules were invented in Boston or Basel, tested in European clinics, and assembled in Chinese factories. That sequence is flipping.

The numbers tell the story

CSPC Pharmaceutical has signed out-licensing agreements worth up to $18.5 billion. RemeGen has deals totaling $5.6 billion. Haisco Pharmaceutical struck a $745 million agreement with AbbVie for pain drug molecules, giving AbbVie the rights to develop, manufacture, and sell outside China.

These are not contract manufacturing deals. Chinese companies are licensing their own proprietary molecules to Western pharma giants. The value chain position has shifted from production to invention.

The AI accelerant

AI is compressing the drug discovery timeline. Identifying candidate molecules, predicting protein interactions, running virtual screening on millions of compounds. Tasks that used to take years of wet lab work now take weeks of compute. China's deep bench of chemistry talent, combined with large patient datasets and aggressive AI adoption, has created a pipeline that global pharma cannot ignore.

Why now

The global pharmaceutical industry is facing a patent cliff. Blockbuster drugs worth tens of billions in annual revenue are losing exclusivity over the next three to five years. Pharma companies need to replenish their pipelines fast, and Chinese biotech firms are sitting on molecules that can fill the gap.

Tony Ren, head of Asia Healthcare Research at Macquarie Capital, framed it directly: "China is already a large player in the global drug value chain, and its role could become larger over the next three to five years."

The transition from low-cost manufacturer to innovation exporter will not happen overnight. Regulatory hurdles, IP concerns, and geopolitical friction all remain. But when AbbVie writes a $745 million check for Chinese-developed pain molecules, the conversation has already moved past theory.

Sources: SCMP, Macquarie Capital

Japan's Sumitomo Mitsui Banking Corp and U.S. asset manager Neuberger Berman are teaming up on a private credit fund to finance Japanese companies making acquisitions. Target size: 100 billion yen ($627 million) within the first year, with plans to scale to 500 billion yen ($3.1 billion) over five years.

Launch is expected in June 2026.

The Details

The structure splits neatly. Neuberger raises capital from institutional investors, identifies M&A opportunities, and structures the lending. SMBC brings its banking relationships and deal flow across Japanese corporates.

Japan's M&A market hit a record 38 trillion yen in 2025, an 86% jump year on year. The number of deals also reached an all-time high. Yet more than 80% of listed Japanese company fundraising still comes from traditional bank borrowing.

That gap between surging deal activity and old-fashioned financing is exactly what the fund targets. Japanese companies pursuing acquisitions need flexible capital that moves faster than a syndicated bank loan, with terms structured around the specific deal rather than the borrower's general credit profile.

SMBC is also considering a separate fund with Nippon Life Insurance. Other Japanese banks have noticed the same opportunity: Aozora Bank and Resona Bank launched similar private credit vehicles in 2025.

The bigger picture for APAC private credit

Private credit across Asia-Pacific is entering what SC Lowy CEO Michel Lowy calls a "structural expansion." SMBC's own APAC Real Estate Credit strategy has already secured $165 million.

The logic is straightforward. Japan's corporate sector is consolidating. Companies are buying rivals, spinning off divisions, and restructuring ownership, all of which require financing that sits between equity and traditional bank debt. With deal volume at record highs and banks still lending the old way, private credit fills a gap that keeps widening.

At $627 million, the initial fund is modest by global standards. The five-year target of $3.1 billion is the number worth watching. If Japanese M&A volume holds anywhere near 2025 levels, the demand for acquisition financing will outpace what the fund can supply.

Sources: DealStreetAsia, SMBC

Every autonomous vehicle today runs two separate systems to understand the world around it. LiDAR measures distance. Cameras capture color. A third layer of software tries to merge both into a single picture. Hesai, the world's largest LiDAR maker by volume, just eliminated the middleman.

What the chip does

Picasso is the first chip to fuse color perception and distance measurement at the hardware level. Instead of stitching together data from separate sensors after the fact, a single chip produces colored 3D point clouds in real time. Up to 4,320 laser channels. No fusion algorithm required.

For autonomous driving and robotics, this matters because the multi-sensor approach has always been the weak link. Cameras can see a red traffic light but struggle with depth. LiDAR knows exactly how far away an object is but has no idea what color it is. The standard fix, running both systems in parallel and merging their outputs in software, adds latency, hardware cost, and failure points.

Why Hesai

Hesai shipped more LiDAR units in 2025 than any other company on earth. The Picasso chip turns that volume advantage into a platform advantage. If one chip replaces what used to require a camera module, a LiDAR unit, and a fusion processor, the bill of materials for an autonomous vehicle drops significantly.

The timing tracks with a broader shift in China's robotics and AV sector. Companies like AGIBOT, Momenta, and WeRide are scaling deployments that depend on cheaper, faster perception hardware. A single-chip solution from the market leader gives those deployments a simpler stack to build on.

Hesai has not disclosed pricing or production timelines for Picasso. But the direction is clear: fewer components, less software glue, lower cost per vehicle.

Sources: CNEVPost, Hesai

South Korea's DeepX is betting that the next wave of AI will not run in data centers. The startup, founded in 2018 by former Apple chip designer Kim Lok-won, makes processors for robots, autonomous vehicles, and factory systems. Its pitch: the DX-M1 chip delivers 20x better power efficiency than Nvidia's Jetson Orin at roughly one-tenth the cost.

Average power draw: 2 to 3 watts.

The Details

DeepX calls its target market "Physical AI," machines that process sensor data and make decisions in the real world rather than on a cloud server. The DX-M1 is manufactured at Samsung's foundry with yields above 90%.

The software layer matters as much as the silicon. DXNN, DeepX's proprietary framework, lets developers take models trained on Nvidia GPUs and run them directly on DeepX chips. DX-Newton, the company's hardware interface, is compatible with Nvidia's Isaac ROS ecosystem. In practice, that means customers can switch from Nvidia to DeepX without rewriting their software stack.

Partners: Advantech, Dell, and Raspberry Pi on the hardware side. Ultralytics and Baidu on software.

Revenue target: roughly $40 million in 2026, with $25 million from product sales. DeepX is planning a domestic Korean IPO, though no timeline has been announced.

Next-gen DX-M2

The successor chip, DX-M2, will use Samsung's 2nm process and deliver up to 80 TOPS at under 5 watts. Mass production is scheduled for 2027.

The cost equation for edge AI

Nvidia dominates AI training and cloud inference. But edge devices, robots on factory floors, drones, delivery vehicles, need chips that run on battery power in dusty environments, not $10,000 modules designed for server racks. DeepX is one of several startups targeting this gap, but few have the combination of Samsung foundry access, Nvidia software compatibility, and hardware partners already shipping.

Kim puts the opportunity in simple terms: "The physical AI market could grow more than threefold over the next five years." At 2 to 3 watts per chip, the math favors whoever can deliver usable AI performance at the lowest power budget. That race is still early.

Sources: The Investor, DeepX

For two decades, ZTE made its money selling base stations and network equipment to China's three telecom operators. That business is plateauing. The operators are spending less on traditional infrastructure: combined 2026 capex is down 8% to RMB 262.6 billion ($38.4 billion). User growth has flattened, and base station coverage across China is near saturation.

The money is moving somewhere else.

Where the capex is going

All three operators, China Mobile, China Telecom, and China Unicom, have made computing infrastructure their top spending priority for 2026. Combined investment: over RMB 80.8 billion ($11.8 billion), with each operator posting double-digit growth in this category.

The global comparison makes the scale clearer. Amazon, Google, Meta, and Microsoft invested $383 billion in AI infrastructure in 2025. S&P Global projects $635 billion for 2026. IDC forecasts global AI investment will exceed $1.2 trillion annually by 2029.

China's own AI usage has grown accordingly. Daily AI token calls across the country jumped from 100 billion to 140 trillion over the past two years. That is a roughly 1,000x increase in two years.

ZTE's bet

ZTE's EVP Xie Junshi described the shift in engineering terms: "Like telecommunications, AI is fundamentally a complex engineering science." The company is repositioning from pure connectivity to connectivity plus computing power, selling the server racks, cooling systems, and networking gear that AI data centers require.

The competition among suppliers is shifting too. When every operator buys similar hardware, differentiation comes down to total cost of ownership, not specs. ZTE's pitch is that its telecom-grade reliability and existing operator relationships give it an edge over pure IT infrastructure vendors.

The transformation is still early. Telecom equipment remains the majority of ZTE's revenue. But the direction is clear: the customers ZTE already has are spending less on what ZTE traditionally sold and more on what ZTE wants to sell next.

Sources: KrASIA, ZTE Investor Relations

Hong Kong's IPO market is pulling in international capital again. Two recent listings raised a combined $2.1 billion with strong orderbooks from across Asia, North America, and Europe.

The Details

Muyuan Foods, China's largest pig breeder, raised HK$10.7 billion (roughly $1.4 billion). The cornerstone investors tell the story of where the demand came from: Thailand's Charoen Pokphand Foods and Singapore's Wilmar International together accounted for about half the deal. A Chinese pig company, priced in Hong Kong, anchored by Southeast Asian food giants.

Biren Technology raised HK$5.58 billion (about $720 million), making it the first mainland Chinese GPU developer to list in Hong Kong. Cornerstone investors included UBS, Eastspring Investments, and Lion Global Investors. Orders came in from North America, Europe, and the rest of Asia.

HKEX CEO Bonnie Chan said the exchange is seeing "very good momentum on supply and demand sides." Her framing: Hong Kong is "building bridges and connecting Asia's capital markets."

Why global funds are showing up

The rebalancing thesis is straightforward. Global investors spent the last two years underweight on China-linked assets. Valuations have compressed. The IPO pipeline in Hong Kong is filling up with companies that would have listed in New York three years ago.

For Biren, the GPU angle matters. After U.S. export controls cut Chinese chip firms off from Nasdaq, Hong Kong becomes the default listing venue for China's semiconductor sector. Biren is the first mover. The pipeline behind it includes several other AI chip and cloud computing companies eyeing HKEX.

The investor mix is the real signal. Cornerstone allocations from UBS, Eastspring, and Lion Global are institutional bets, not retail speculation. And Muyuan's anchoring by CP Foods and Wilmar shows that Asian strategic capital is flowing through Hong Kong, not around it.

Two deals do not make a trend. But $2.1 billion raised with meaningful international participation, not mainland money recycling through Hong Kong, suggests the market is open for business again. The test comes with the next wave of filings over the summer.

Sources: SCMP, Bloomberg, HKEX

In Nanchang, China, humanoid robots now assemble tablets around the clock. AGIBOT and Longcheer Technology, one of the world's largest ODMs for consumer electronics, have deployed G2 humanoid robots on a full production line. The setup runs 24/7 with zero human intervention.

The numbers

Cycle time: 19 to 20 seconds per unit. Output: 310 units per hour. Success rate: above 99%. The robots have logged over 140 continuous hours so far, with downtime losses below 4%.

Perhaps the most telling detail: integrating the robots into Longcheer's existing factory took 36 hours. Not months of retooling. Not a new facility. Thirty-six hours inside a running plant.

How it works

Each G2 robot uses what AGIBOT calls "Physical AI," processing decisions on-device rather than relying on cloud instructions. That means the robot reacts to what it sees and touches in real time, without waiting for a server response. For repetitive assembly tasks on a production line, local intelligence removes the latency that would slow everything down.

AGIBOT shipped its 10,000th robot last month. The company plans to scale the Longcheer deployment to 100 units by Q3 2026.

Where this fits

Humanoid robots have spent the last three years in demo mode. Walking on stage, shaking hands with executives, doing choreographed warehouse tasks for YouTube clips. AGIBOT's deployment is different because the output is measurable: real products, real throughput, real error rates.

Dr. Yao Maoqing, AGIBOT's founder, put it plainly: "2026 marks the beginning of large-scale deployment for embodied intelligence."

Whether that holds depends on how fast other factories adopt. But 310 tablets per hour with a 99% success rate is a hard number to argue with.

Sources: Humanoids Daily, AGIBOT

China's March exports grew just 2.5% year on year, the weakest reading in five months and below analyst forecasts. Imports told the opposite story: +27.8%, the strongest since November 2021.

Trade surplus: $51.13 billion.

The Details

The export slowdown comes as the Iran conflict adds fresh uncertainty to global supply chains, particularly for AI-driven tech goods and energy logistics. War in the Middle East tends to disrupt two things simultaneously: shipping lanes and commodity prices. Both matter for China's export machine.

On the energy side, natural gas imports fell 10.7% and crude oil imports dropped 2.8%. Refined oil exports, however, jumped 20.5% month on month. China's refineries are processing more crude domestically and shipping the finished product.

South Korea tells an interesting subplot. Korean exports to China surged 62.4% in March, driven almost entirely by semiconductors. The chip cycle is running hot between these two economies even as broader trade patterns cool.

The energy math

The energy trade paints a split picture. China is buying less raw fuel but exporting more refined product. That is a conscious strategy: move up the value chain, sell gasoline and diesel instead of importing it. With Middle Eastern supply routes under pressure from the Iran conflict, domestic refining capacity becomes a strategic asset.

Trump's planned visit

President Trump is expected to visit China in May for a meeting with Xi Jinping. Analysts anticipate deals around agricultural goods and aircraft parts. Movement on Taiwan or structural trade issues is not expected.

The import surge deserves scrutiny. A 28% jump rarely comes from organic demand alone. Front-loading before anticipated tariff hikes, commodity restocking, and base effects from a weak March 2025 all likely contribute. The export number is harder to explain away. A five-month low while global demand for Chinese electronics remains strong suggests friction is building somewhere in the pipeline, whether from logistics disruptions, shifting orders, or both.

Sources: Reuters, Bloomberg