This set of numbers is like a mirror, clearly reflecting how China’s high-end manufacturing is responding to the worldwide AI boom. Let’s break it down in a simple, straightforward way.

1. Export Side: AI Demand Accelerates High-End Product Exports

       Exports simply mean Chinese-made goods being sold overseas. In June, the standout story is that high-end manufacturing and high-tech products took center stage.

• Electromechanical products (machinery, electrical equipment, electronics, and industrial components) grew 34% year-on-year and 8% month-on-month, accounting for nearly half of total exports.
• Integrated circuits (chips — the “brain” inside every electronic device) saw export value surge 122%, effectively doubling.
• High-tech products rose 52%.
• Automobile exports jumped nearly 73%, with new energy vehicles (NEVs) performing especially well.
• Rare earth exports also more than doubled in value.

       The main driver is the explosive global demand for AI infrastructure. Countries are racing to build data centers and train large AI models, which require massive amounts of servers, advanced chips, and sophisticated electronic components. As the world’s largest electronics manufacturing base, China naturally became a key supplier.

       On the auto side, strong NEV exports show that more countries are accepting and even preferring Chinese new energy vehicles for their quality and value.

       Where are these exports heading? The United States remains the largest single market at $43.4 billion (up ~14%). However, the fastest growth came from exports to Taiwan region (+44%) and South Korea (+43%) — increases closely tied to the AI chip and semiconductor supply chain.

2. Import Side: Domestic Tech Buildout Drives Component Demand

       Imports rose 36% year-on-year. Electromechanical products grew 47% and high-tech products grew 57%, with both still rising month-on-month.

       This directly reflects China’s rapid expansion of AI computing power and smart hardware assembly lines. The country needs to import large volumes of core components and production equipment to support its own high-end manufacturing upgrade. In short, China is both “selling out” high-end goods and “buying in” critical parts — building a complete industrial chain from both ends.

       Energy and bulk commodities showed a different pattern. Crude oil import volume fell year-on-year, while natural gas imports rose but at higher prices. Grain and soybean imports continued to increase, partly to replenish national reserves and company inventories.

3. What These Numbers Really Tell Us

       The clearest signal from the data is that global technological restructuring — especially the AI wave — is delivering real benefits to China’s high-end manufacturing sector. China’s stable and resilient supply chains stand out even more clearly against rising global geopolitical uncertainty. The country is accelerating its shift toward AI hardware, chips, new energy vehicles, and other high-tech, high-value areas — what policymakers call “new quality productive forces.”

       At the same time, as a major energy importer, China faces external disruptions and needs to plan ahead: diversify supply routes (both sea and land pipelines) and maintain bottom-line options such as coal-to-oil and coal-to-gas technologies for extreme scenarios to safeguard energy security.

       Additionally, volatile international energy prices could create imported inflation pressure, directly affecting domestic companies’ production costs, profits, and hiring. It’s worth watching how the central bank and other policymakers respond in the coming months.

       Trade data is never just cold numbers — it’s the economy’s “weather report.” June’s figures highlight both the upgrading potential of Chinese manufacturing and the need to balance global opportunities with solid security foundations.

       But here’s the important clarification: the photo showing a drone carrying two people is AI-generated. It’s simply too dangerous and physically impossible for current drones to safely lift two adults. Real rescues that happened involved lifting just one person, and only in truly life-or-death situations where someone was about to be swept away.

       The real story took place on July 7 in Yunbiao Town. A truck driver was trapped on top of an oil tanker in the middle of the flood. A rescuer in an orange-red suit flew in on a drone, landed on the vehicle, and then both were lifted out together. The whole operation took less than 20 minutes.

       This wasn’t some hobby drone from the park. It was a coordinated effort by ordinary pilots from across China who rushed to the disaster zone on their own.

1. An Old Dam Couldn’t Withstand the Storm

       The disaster started with Typhoon Misaak. Between July 5–6, parts of Nanning and Guigang in Guangxi saw torrential to extremely heavy rain. Binyang County recorded 713 mm in 24 hours, and Shitang Town in Hengzhou saw 565 mm — basically a full year’s rainfall for many northern regions dumped in a single day.

       On the morning of July 6, Liulan Reservoir in Hengzhou experienced overtopping. This 60-plus-year-old homogeneous earth dam (built in 1958–1960 by compacting soil) simply couldn’t hold back the water. Once water flowed over the top, it rapidly eroded the back side of the dam, and the structure gave way.

       By July 9, the floods had caused 39 deaths and 9 people missing. Roads and bridges were destroyed. Even assault boats sometimes couldn’t reach the trapped areas because the current was too strong.

2. Drone Pilots Showed Up on Their Own — Hundreds of Aircraft

       At this critical moment, a special group of people drove in from all over the country.

       Pilots came from Guangxi, Sichuan, Guizhou, Hunan, Guangdong, and even as far as Heilongjiang. They loaded their own drones and generators into their vehicles, paid for their own gas, food, and lodging, and asked for zero payment or subsidies. They just headed straight to Hengzhou.

       Because everyone came spontaneously, exact numbers are hard to track, but reports mentioned WeChat groups with hundreds of members and several hundred drones operating on site.

       By July 8, these drones had delivered more than 1,000 tons of relief supplies. At one distribution point in Dutian Village, five drones worked non-stop around the clock. In Yunbiao Town, single drones covered a 4–5 km radius, reaching multiple trapped villages.

       They couldn’t always pull people out, but they could drop water, instant porridge, and instant noodles — giving trapped residents something to hold onto.

3. DJI Stepped Up: Free Repairs for All Rescue Damage

       One of the most heartwarming details came on July 8. DJI Agriculture issued a notice: any plant-protection or cargo drone used in the Guangxi rescue that got water damage, crashed, or had parts broken during the operation would receive completely free repairs — no labor fees, no parts charges, no inspection costs.

       Normally, water damage or crashes are not covered under warranty. But this time DJI said: just provide your flight logs from the rescue period and report the drone’s serial number, and authorized stores would handle it immediately. Local DJI stores even switched to 24-hour operation to service the pilots overnight.

       For many pilots, a good drone plus accessories and a vehicle can cost 200,000–300,000 RMB. DJI’s policy removed the biggest financial fear — “what if I crash my expensive machine?” — and gave them the confidence to go all in.

4. These Aren’t Toy Drones — They’re Serious Work Machines

       The drones used in the rescue are very different from the small consumer drones people fly in parks for fun.

      These are professional plant-protection drones and cargo drones.

• Plant-protection drones (T series):
  Built for agriculture — spraying pesticides and spreading fertilizer. The common T200 model can carry dozens of kilograms per flight.
• Cargo drones (FlyCart series):
  Designed specifically for lifting and transporting goods. The FC200 can lift up to 200 kg by itself, and multiple units can work together to lift 600 kg.
• Hybrid VTOL / compound-wing drones:
  These take off and land vertically like a multi-rotor drone but fly long distances like a fixed-wing aircraft. Some were used in the disaster zone as temporary flying 4G/5G base stations to restore mobile signals.

      These machines normally work in farmland. This time they went straight into emergency rescue.

5. China’s Drone Industry Grew 100× in Just Ten Years

       Most people don’t realize how big this industry has become.

       Around 2015, DJI’s agricultural drone shipments were only about 2,000 units per year. By 2020 it reached 40,000 units. In 2025 it hit 200,000 units annually. In just ten years, yearly sales grew roughly 100 times.

       By 2025, China had around 500,000 of these agricultural and cargo drones operating nationwide. They covered more than 3.2 billion mu (about 213 million hectares) of farmland in a single year. Roughly one-third of China’s cultivated land now relies on drones for pesticide application.

       DJI holds about 55% of the domestic plant-protection drone market and around 59% globally. These drones are now used in over 100 countries, with especially fast growth in places like Brazil and Mexico.

6. A Brand-New “Atomized” Job: The Agricultural Version of Didi

      This is an entirely new type of work.

       Most pilots are independent individuals or very small teams — no big company, no boss, no formal employment. They connect loosely through WeChat groups and order platforms (one popular one is called “Nongtian Guanjia” or Farmland Steward — basically the agricultural version of Didi). Farmers place orders on an app for spraying or transport, and the platform assigns the job to the nearest qualified pilot.

       Pilots must hold a CAAC drone pilot license issued by China’s Civil Aviation Administration. There are three levels: Visual Line of Sight operator, Beyond Visual Line of Sight Captain (the main one used for real work), and Instructor. Getting certified costs money — roughly 4,000+ RMB for the basic license and 8,000–12,000 RMB for the captain level.

       A complete setup (drone + vehicle + fuel generator + chargers) usually costs 200,000–300,000 RMB. Many pilots took loans to buy their equipment.

       Earnings: Spraying pesticide typically pays about 20 RMB per mu. For cargo lifting, a drone can move around 10 tons in a good day, generating roughly 2,400 RMB in gross revenue before fuel (~300 RMB) and helper labor (~300 RMB). Net daily income can reach over 1,000 RMB under good conditions. It’s real work, though — weather, order volume, and equipment wear all matter. Most pilots say they can break even in about two years, but the market is gradually becoming more competitive, similar to what happened with ride-hailing.

       These pilots are also very good at shooting and editing videos. They post daily on Douyin (Chinese TikTok), which helps them promote their services and sometimes recruit students.

7. Why Did Drones Stand Out So Much This Time?

       Professional rescue teams, government, and military are still the main force in disaster response. So why did the drones get so much attention?

Two main reasons:

       First, the pilots are naturally skilled at content creation. Drone footage has great visual angles, and the videos look dramatic and hopeful — perfect for going viral.

       Second, after a disaster, people need both facts and hope. The drone pilots delivered positive, concrete stories of technology and ordinary people taking action, which felt especially valuable when everything else felt heavy.

8. The Truly Moving Part

       When you connect all the pieces: a group of young people who returned to rural areas to make a living spent 200,000–300,000 RMB (often with loans) on drone equipment. On normal days they post videos, take orders, spray fields, and transport fruit — basically running a small business one job at a time.

       They are scattered, independent operators. Then the flood hit. These strangers, with no central organization, used just their phones and a few WeChat groups to gather into a fleet of over a hundred drones within two or three days.

       No one was directing a grand plan from above. There was no strict hierarchy. It was simply the kindness of ordinary individuals combined with China’s world-leading drone technology and existing digital platforms that allowed them to connect and act so quickly.

       In the past, we often assumed that big things require big, tightly organized systems with top-down commands. What we saw here was something different: countless independent people, normally connected loosely through social media and service platforms, suddenly becoming a capable rescue force when it mattered most.

       The organization was loose. The hearts were kind. Technology did something remarkable — not just lifting a person out of danger, but amplifying and connecting the goodwill of thousands of ordinary people for the first time.

       Life has to go on. Stories like this remind us that even in difficult moments, there are more sources of warmth and possibility than we sometimes realize.

       He said, “Back then, people would walk in with phones to sell all the time. I didn’t even have to ask — you could tell most of them weren’t theirs. These days? I might not see someone like that for months.”

       He described the old customers: hats pulled low, eyes darting around, phone with no box or receipt, straight to the point — “Boss, how much is this worth?” Everyone knew the score, and the boss played along.

       But things have flipped in recent years. It’s not just that fewer people try to sell shady phones. Even when someone finds a lost phone on the street, they usually try to track down the owner and return it.

       Foreigners visiting China often say the same thing surprises them most: packages left at doors stay untouched, takeout left downstairs doesn’t get stolen, and if you drop your phone in a mall, there’s a real chance you’ll get it back.

So they ask: Has China run out of thieves?

       Not exactly. The thieves didn’t suddenly vanish. “Being a thief” simply stopped being a job worth doing.

1. Back Then: Stealing Was a Real “Business”

       Twenty years ago, stealing a wallet could mean thousands in cash you could spend the same day.

       Stealing a phone? Take it somewhere to flash and unlock it, then sell it to a second-hand dealer — instant income.

       The main skill you needed wasn’t technical ability. It was just being fast on your feet. If you didn’t get caught red-handed, the risk afterward was pretty low.

2. Now: Phones Have Become “Hot Potatoes”

       Steal a phone today and the moment it powers on, the owner can track its location.

       Trying to flash or unlock it? Modern phones have much stronger protections, so success rates have dropped sharply.

       Even if you strip it for parts, the money you make often isn’t worth the risk of getting caught — and serving years in prison.

       What used to bring in serious cash now might only be worth a few hundred dollars in parts.

3. Cash Has Disappeared — Stealing a Wallet Barely Pays

       These days a lot of young people leave the house with just their phone. Many don’t even carry a wallet once a year.

        If you do steal one, it probably contains an ID card and a couple of bank cards whose passwords were forgotten long ago.

        After all that effort, you might walk away with only twenty bucks in actual cash.

4. Surveillance Has Closed Almost Every Escape Route

       Back then, a street might have had cameras only outside the bank.

       Now? Cameras are everywhere — apartment entrances, elevators, malls, subways, parking lots, convenience stores, even many residential doorways.

        People joke that there are no more blind spots… because even the blind spots might have an old guy filming with his phone.

        Lose something and it often gets posted in the neighborhood WeChat group within minutes. Neighbors jump in to help look, and recovery rates have gone way up.

5. It’s Not That People Suddenly Got “Better” — They Just Ran the Numbers

       A lot of people assume Chinese society became more moral overnight.

       But most folks aren’t wrestling between good and evil every day. They’re quietly doing the math:

       How much can I actually make from this? What are the chances I get caught? If I do, how many years will it cost me afterward?

       In the past the numbers made sense, so some people did it.

       Now the costs have crept up — dense cameras, accurate tracking, hard-to-sell stolen goods, heavier penalties — while legitimate ways to earn money have multiplied: food delivery, ride-hailing, factory work. The pay might not be amazing, but you can sleep without looking over your shoulder.

       When someone thinks about reaching out, the calculation usually finishes in their head first.

       This one job might not even make that much money.

       But if you get caught, the next several years could be a total loss.

       So the hand naturally stays in the pocket.

6. What Foreigners Are Actually Seeing

       The safety foreigners notice in China isn’t just because there are fewer thieves.

       It’s because an entire society gradually turned “stealing” from a workable way to make a living into a consistently losing proposition.

       Good systems don’t always work by catching more bad guys every day.

        Sometimes they succeed by making the bad choice stop looking attractive to most people.

       When enough individuals quietly decide “Yeah… this isn’t worth it,” the result feels like safety.

On June 22, Japanese company Sakana AI released Fugu (meaning “pufferfish”). It scored 73.7 on the SWE-bench Pro coding benchmark, beating Anthropic’s Claude Opus 4.8 (69.2) and OpenAI’s GPT-5.5 (58.6). Social media quickly filled with excitement that Japan had finally arrived as a serious AI contender.

Before we celebrate, let’s slow down and look at what actually happened. Here’s a clear breakdown.

1. Who is Sakana AI?

Sakana AI was founded in Tokyo in 2023. Unlike labs that train massive models from scratch, Sakana specializes in combining and orchestrating existing models.

The company has three founders with complementary strengths:

• CEO David Ha: Chinese heritage, born in Hong Kong. He studied at the University of Toronto and earned his PhD from the University of Tokyo. He previously worked on neural networks and evolutionary algorithms at Google Brain and served as research lead at Stability AI.
• CTO Léon Jones: British engineer and one of the eight authors of the landmark “Attention Is All You Need” Transformer paper. He spent 12 years at Google and drives the company’s technical direction.
• COO Ito: Law graduate from the University of Tokyo and NYU, with prior experience at Japan’s Ministry of Foreign Affairs. He brings government connections and experience scaling Japanese tech companies globally.

It’s a blend of strong technical talent and deep roots in the Japanese market.

2. Sakana’s three main products so far

• Their Japanese large language model (released March 24), paired with a free Sakana Chat interface.
• An autonomous agent (released June 15) that can run unsupervised for about 8 hours and generate 60–100 page strategic reports.
• Fugu (released June 22) — the multi-model orchestration system behind the recent benchmark results.

3. Fugu is not a single model — it’s a system

Fugu’s core is a “Loop Agent” workflow. Instead of building one all-powerful model, it routes tasks across multiple existing models in a repeating loop:

Task arrives → one model breaks it down → another executes the work → another reviews the output → if the result isn’t good enough, the loop repeats until it meets the standard.

Sakana claims Fugu partially surpasses their own previously announced Mythos and Fable models (these were announced by Sakana; there are no independent third-party public benchmarks available, and access is currently restricted — non-US citizens may not be able to use them). It also outperformed Claude Opus 4.8 and GPT-5.5 on SWE-bench Pro.

4. The important detail: Fugu doesn’t disclose which base models it uses

Fugu doesn’t reveal exactly which models it calls internally. It only confirms it is not using its own latest Fable 5 and Mythos 5 versions (for competitive reasons and due to regulatory constraints).

It is very likely combining several strong publicly available models — for example, using Claude Opus 4.8 for planning, GPT-5.5 or DeepSeek for execution, and looping back for verification. This “mix and improve” approach helps explain the strong benchmark numbers.

5. Does Sakana have any models of its own?

Yes, but they are not frontier base models.

Inside the Fugu system is a lightweight “conductor” model fine-tuned from Qwen 2.5 7B that mainly handles task routing and loop management.

Their earlier Japanese LLM was fine-tuned on DeepSeek V3.1 Terminus and also drew from the Llama 3.1 405B open-source base. Its main strength is localization: it significantly lowered refusal rates on Japan-related historical and political topics (from around 72% in the base model to nearly zero).

In short, Sakana excels at adapting existing models for the Japanese market rather than pushing the absolute frontier of base-model intelligence.

6. Why this doesn’t create a lasting moat

Model orchestration itself is not a high barrier to entry. On June 12, OpenRouter — the largest model aggregation platform — launched a very similar feature called Fusion. It can combine up to 8 models on the same query and fuse their answers with a judge model. They claim it can approach top-tier quality at roughly half the cost without needing the single strongest model.

OpenRouter reported that about 75% of the performance gain came from the fusion step, with only 25% coming from model diversity.

This shows that multi-model orchestration is an open engineering technique available to anyone — not a unique advantage for Sakana.

The real competitive moat comes from owning frontier base models + controlling token pricing and access. Those currently sit with OpenAI, Anthropic, Google, and similar labs. They decide when new models are released, run their own internal orchestration systems (already through many iterations), and set the prices everyone else pays.

Companies like Sakana can only use publicly released models and have little control over costs. The big labs can adjust pricing, limits, or even restrict API access at any time. This situation closely resembles the early internet advertising industry: many smaller companies built clever optimization tools, but the traffic and pricing power stayed with Google and Meta. Most were eventually squeezed or acquired.

7. What’s likely next for Sakana AI?

Sakana is unlikely to disappear, but it probably won’t become a global AI leader on its own.

A more realistic path is acquisition — potentially by Google Japan, OpenAI or Anthropic’s Japan teams, or another player focused on the Japanese market.

For Sakana, being acquired could actually be a positive outcome. Buyers would likely value three things:

• The elite technical team (including a co-author of the Transformer paper).

• Strong access to Japanese government and corporate relationships.

• Proven ability to localize models for Japanese users and sensitive topics.

Japan has a history of nurturing domestic champions. If Sakana ends up deeply embedded in the Japanese market rather than trying to outcompete the global leaders head-on, that could be a sustainable position.

8.Quick takeaways

• For founders: 

Pure orchestration / Loop Agent products are hard to scale long-term without owning a base model, proprietary data, or a unique local/enterprise position. The best realistic outcome is often acquisition by a larger player. Treat orchestration as a useful capability, not the entire business bet.

• For regular users: 

When possible, stick with models and services from the original providers. Third-party orchestration systems can only work with already-released models from others and usually don’t disclose exactly which models they’re combining. This makes consistency and reliability harder to verify.

• For investors and creators: 

Ask two questions first: Does the company control its own base model? Does it control its own pricing power? For orchestration-focused companies, valuation should reflect realistic acquisition potential rather than assuming they will become the next foundational model leader. Benchmarks can be assembled; real moats have to be owned.

9. Bottom line

Sakana AI’s Fugu release looks like a breakthrough for Japanese AI on the surface. When you look closer, it’s a smart orchestration company that leveraged existing resources to deliver strong results.

It underscores a key reality in today’s AI landscape: the companies that own frontier base models and pricing power hold the real leverage. Orchestration techniques are clever and practically useful, but on their own they don’t create durable competitive advantages.

For Sakana, the most probable future is becoming a meaningful player in Japan’s domestic AI ecosystem — likely through partnership or acquisition — rather than suddenly emerging as a peer to the global leaders.

What do you think about model orchestration approaches like this? Smart short-term tactic or long-term limitation? Share your thoughts below.

 Recently, US Commerce Secretary Howard Lutnick raised concerns in a meeting with ASML executives: Is there an ASML EUV lithography machine — or at least key components — that may have bypassed export controls and reached mainland China?

Bloomberg reported that US officials say they have evidence, but have not made it public.

ASML’s response was clear: No complete EUV machine is in China.

After the meeting, ASML circulated a document in Washington with key figures:

• There are currently 314 EUV machines worldwide
• Most are in Taiwan, South Korea, and the US (TSMC, Samsung, SK Hynix, Intel)
• None are in mainland China
• Another 26 have been decommissioned and scrapped — also not in China

ASML stressed that an EUV machine weighs 180 tons and has about 100,000 parts. Transport, installation, and maintenance require their engineers. The machines stay connected to headquarters, allowing remote monitoring and the ability to shut them down if needed. They say every machine is fully accounted for.

1. The two sides seem to be answering slightly different questions

The US appears to be asking whether China has gained EUV-level capability or access to critical components. ASML answered that no complete ASML EUV system has gone to China.

It feels like they’re talking past each other.

2. Quick explainer: EUV vs DUV

EUV (extreme ultraviolet) uses a much shorter wavelength, allowing finer patterns on chips — essential for key layers in advanced nodes. DUV (deep ultraviolet) has a longer wavelength, so making fine features requires multiple exposures.

Huawei and SMIC currently rely on DUV multi-patterning to produce 7nm chips. But more exposures mean accumulated errors and lower yields. EUV’s advantage is achieving complex patterns with fewer exposures, improving both speed and yield.

Even with EUV, chips are still built layer by layer — it’s not a single “flash” process.

3. Where do China’s 7nm chips come from?

Mainly three sources:

• Stockpiles ordered by Huawei before tighter restrictions kicked in

• Chips ordered through a related entity (Sophgo, linked to Bitmain) from TSMC — TSMC later terminated the relationship after discovering the end use

• SMIC producing them domestically using existing DUV machines with multi-patterning techniques

TSMC itself used DUV multi-exposure for 7nm in the early days. Once EUV became available and more cost-effective (better yield and speed), it shifted. Some process know-how likely spread through various channels.

Still, using DUV for 7nm requires dozens of exposures — yields are not great.

4. Does China have EUV capability?

A few layers to consider:

• Basic principles: 

China operates many DUV machines and understands lithography workflows well. ASML has a large software R&D center in Shenzhen, so the fundamentals are familiar.

• Core components: 

This is the tough part. Zeiss precision mirrors, the EUV light source, and high-precision stages are extremely difficult to replicate at the required accuracy.

However, these machines need regular maintenance and part replacements. It’s possible that some critical components could have entered circulation during servicing. ASML can check major modules via the machine’s software, but tracking every single small part would be extremely costly and technically challenging.

So ASML can confidently say “no whole machines,” but hasn’t fully closed the door on component-level questions.

5. The real barrier is the certification system

You can’t just build something and use it. Every part in an ASML machine goes through rigorous engineering certification. Think of the C919 airliner — some cables must be original certified parts, not because they can’t be made, but because they haven’t passed the required validation process.

The certification involves many detailed tests and standards that aren’t fully public. Without participating in the system, it’s hard to know exactly where the bottlenecks are or how to improve.

6. To fully block China would require “three cuts”

• Prevent participation in the supply chain and R&D from the start

• Strictly stop critical components from reaching China

• Use the certification system as a gatekeeper

The second cut is very hard to enforce. Supply chains are deeply intertwined. China was once ASML’s largest market (around 30% of sales in 2024, mostly DUV). ASML has over 1,400 suppliers in Asia, and some materials inevitably involve China.

7. China’s practical advantage: low-cost, fast iteration

Western and Japanese/Korean companies face high costs for each trial — a few failures can hurt badly. China benefits from lower costs in labor, facilities, energy, and materials, plus a complete domestic supply chain. With policy support, it can run multiple test cycles faster and cheaper.

This is similar to China’s high-speed rail story: start with imported systems, test and localize step by step, and reach very high domestic content.

8. Why is the US pressing ASML now?

It may be connected to Huawei’s recent process optimization announcement (sometimes called the “Tao Ding Law” in discussions). The changes are mostly clever reordering of steps — similar to what TSMC did years ago — but they signal that pushing to even more advanced nodes will eventually require EUV-level capability.

The US may have spotted specific supply-chain or technical details worth asking about. ASML is threading a careful line: addressing US concerns while not wanting to lose the Chinese market (still over 20% of sales in 2025).

9. Bottom line

Based on available information, China appears to have developed some form of working EUV capability — not by acquiring complete machines, but likely through reverse-engineering similar systems and incorporating critical components obtained via maintenance, replacement, or decommissioned equipment.

This gets machines running, but it’s still a long way from stable, high-yield mass production of cutting-edge 5nm or 3nm chips. Cost and efficiency remain real hurdles.

The US wants to maintain a technological edge, ASML is balancing commercial realities, and China continues investing and iterating. In a deeply interconnected global supply chain, completely cutting off one side comes at a high cost for everyone.

This story is still unfolding — worth keeping an eye on.