I spent ten days in Norway this summer. What follows are reflections from my time there on Oslo, the Vikings, and WWII.
Oslo Vibes
“This place isn’t perfect Jordan,” a civil servant told me, “please tell me you won’t make that your angle.” I then asked him what the worst neighborhood in Oslo is, walked there, and felt it was nicer than half of Manhattan.
The first few days of 19 hours of sunlight in 72-degree weather were an unparalleled endorphin rush, but by day six I felt a little strung out.
Servicepeople regardless of your race start conversations in Norwegian so as to not make immigrants feel unwelcome.
I played some pickup sand volleyball in one of the thousand Oslo parks with a Kurdish culture affinity club. No-one on my team could tell me how to say “nice serve” in Kurdish but when some Kendrick came on their speakers, they all sang along to “certified Loverboy, certified pedophile.”
Chinese EV showrooms dotted Oslo, with Nio taking plum position on the main street right outside parliament. The salesman there said vibes are mostly good, though every few weeks someone walks in just to say “we don’t like Chinese cars here.” The XPENG 小鹏 saleswoman unprompted told me, “we are Chinese but a private company not owned by the government like BYD. Also, Volkswagen owns 5 percent and Norwegian oil fund owns some of us too.”
Norway up until the 70s was one of the biggest Israel supporters. Their two Labor parties both ran their countries for decades, and living on a kibbutz was a thing Norwegian lefties did. But Norwegian soldiers saw some shit as peacekeepers in Lebanon in the 80s, everyone got really invested in the Oslo Peace Process and felt burned by the Israelis in the subsequent decades. “We were a colonized country too, you know. First the Danes then the Swedes…”
Thanks presumably to oil wealth guilt, Norway might be the country most into ESG. The government in early June officially recognized Palestine but Parliament decisively voted down a push to make the Oil Fund divest from all companies with ties to Israel. They did recently sell $70m of Caterpillar stock…? The ratio of pride to Palestinian flags was maybe 5:1.
Haaretz recently ran a feature on rising antisemitism in Norway which convinced me I didn’t want to move there. For an illustrative excerpt on what happened when a group of Jews tried to join an International Women’s Day protest to raise awareness of Hamas. They got approval to join, and on parade day this happened:
The hostile reaction manifested almost immediately. Initially, the group was refused entry to the event and had to prove that they had the organizers' authorization to participate. "One of the organizers went on shouting and cursing, and then took one of our signs and threw it on the ground," Nilsen recalls. "After the police made sure he couldn't get close to us, more and more organizers told us that our message conflicted with the messages of the event.
"They looked at us with hatred and disgust and started to shout that we were Zionists and fascists. Then the crowd joined in with slogans and rhythmic chanting that we were already used to, like 'Murderers,' 'No to Zionists in our streets' and 'From the river to the sea, Palestine shall be free.'"
They avoided getting into a direct confrontation, Nilsen relates, "and we instructed our group not to scatter and not to respond. But when the atmosphere heated up, some of the other demonstrators – Norwegian men and women of my age – approached the members of the group very closely and whispered into their ear things like 'child murderer' and skadedyr' ['parasites' in Norwegian].
"I've had anti-Israeli calls shouted at me in the past," Nilsen continues. "But this time it was very different. The hatred came from people I thought we shared basic values with. The feeling was that we were being canceled as human beings. We weren't women and men – we were the embodiment of evil."
Parks midday on a Monday were packed. There’s an abundance of minigolf. Workdays in winter start very early so people can get some sunlight outside the office in the afternoon.
Norwegian youth wear the most boring clothes I’ve ever seen in a city. The one signature that stood out were these rainbow-tinted athletic glasses. A few years ago, a comedian made a hit song about the top brand which features a yodel.
Norway had the highest ratio of American to local music I’ve ever seen in a Spotify Top 50. The vast majority of what modern Norwegian hip hop, pop, and indie I came across was flat.
At first I thought there was some adverse selection going on where the best artists try to make it in English, but an arts and culture newspaper editor told me that actually that the cool thing nowadays is to sing in the local language. The Swedes have figured this out…what gives, Norway?
The closest to okay top Norwegian act I came across was Karpe, a rap duo of a Hindu and Muslim second generation immigrants. Electronic music was much stronger. I quite liked this mix and was told they do jazz well too.
Vikings
After flipping through a handful of intro to Vikings books, Children of Ash and Elmstood out for its writing and breadth. It an excellent portrait of the Vikings which brought the terror as well as the humanity to the culture. For instance, I quite liked this discursion into Viking bread.
Some more good writing:
And this:
This list of sea-king names was amazing:
The sagas were also surprisingly accessible and make for great audio books. The Poetic Edda would be my bet for an entry point.
But let’s not forget, the Vikings were actually horrible. This account of a king’s burial by a travelling Arab diplomat in the 900s is one of the most terrifying primary sources I’ve ever come across.
Sexual violence trigger warning.
Modern Norwegian History
Aside from non-fiction on Vikings and Hitler in Norway, the only book-length title I came across telling the history of modern Norway was The Norwegian Exception: Norway’s Liberal Democracy since 1814. I found its thesis hysterical: it’s been incredibly lucky. Its neighbors Sweden, Denmark, and Russia never invaded. The touchiest moment came in 1905 with Sweden…I’m sorry but I can’t help at laughing at the nationalist chest-puffing in Scandanavia.
But ultimately, good call by Norway conceding on the great reindeer dispute of 1905.
Other lucky turns: Norway’s time under Nazi Germany was the easiest ride of any country that got conquered in WWII (good book the occupation here). The country should get some credit for not having a civil war, fumbling the bag when it comes to exploiting the boom in global trade in the late 19th century, successfully leveraging water power to industrialize in the early 20th, and of course making the most out of its oil riches.
But by the 70s, they somehow they became the party of no fun.
WWII
Aside from Vikings, you also have a number of incredibly detailed but not particularly engaging books on Hitler’s invasion. Here’s the case for caring:
The most interesting bits I found were on the strategic level, where before Germany made its move the UK was also dancing around a pre-emptive invasion primarily to secure iron ore. At one point, France pitched the UK to come into the Winter War on the side of the Finns, doing the enormously idiotic move of putting them directly in conflict with the USSR.
Can’t pass on another opportunity to clown on Chamberlain.
Photos
Oslo is big on public art and every other statue was naked. City Hall had some particularly suggestive murals.
年轻一些的时候，甚至人到中年，经常忽视生活中这些善意，甚至用恶意去看世界，以为是犀利或深刻。年龄会改变人。像得克萨斯演员Tom Lee Jones演绎的得克萨斯故事中说的一样：“Age will flatten a man”（No Country for Old Men）。大意是说，“年纪会磨去人的棱角”，尤其是恶意的棱角吧。觉得年纪大了些，变化之一就是，知道学习体会周围人和陌生人表现的善意和体贴，开始珍惜这些善意和体贴。以前不在意的一些事，现在觉得宝贵了。
And a job post! ChinaTalk is hiring for a dedicated China AI lab analyst. Chinese fluency and a technical background are required. Apply here!
Okay, Trump — Your Turn
Jordan Schneider: We have new regulations with significant gaps [discussed in depth in part 1 of our conversation], and a new president arriving in four weeks. What should Trump and his team do on chips? And what do you think they will do?
Greg Allen: Marco Rubio, our presumptive Secretary of State, has consistently criticized the Biden administration’s export-control packages as too lenient, citing numerous loopholes and oversights. While the Commerce Department leads on dual-use technology export controls, the State Department participates in the interagency decision process.
Rubio’s passion for addressing Chinese technology threats could make him an influential voice in this arena. Similarly, the incoming national security advisor, Mike Waltz, prioritizes Chinese technology competition. The Biden administration established this new approach to the Foreign Direct Product Rule — a tool now available to the US government. The Trump administration might wield this tool quite differently.
Jordan Schneider: Let’s revisit our strategic premises, particularly regarding allies and partners. Trump’s negotiation style with allies differs markedly. The irony would be if Trump confronts allies over minor issues like Mexican auto imports or Canadian timber while overlooking semiconductor manufacturing equipment — the EU’s primary export to China and Japan’s second-largest.
If Trump takes an aggressive, unilateral approach, allies might accept semiconductor restrictions while focusing on larger concerns like NATO’s stability or US troops in Okinawa. The impact on industry follows similar logic — these restrictions won’t collapse American, Dutch, or Japanese economies.
The crucial question becomes, “Do we abandon end-use controls for a nationwide approach?”
Should we implement straightforward restrictions on sub-300mm semiconductor equipment exports to China, eliminate servicing allowances, and replace 200-page rulebooks with five-page directives?
Greg Allen: The Trump administration initiated our modern semiconductor export control approach — from chip-level restrictions with ZTE to the Foreign Direct Product Rule affecting Huawei and TSMC, and equipment export controls involving Dutch EUV machine licensing. The question is whether they’ll follow this strategy to its logical conclusion.
No apology to China would dissuade their pursuit of domestic self-sufficiency and indigenization. They’re fully committed to this strategy regardless of any potential trade deals with Trump. The distinction lies between appearing tough and implementing effective policies.
Regarding countrywide export controls: they’ve proven unambiguously most effective among our policy iterations. The current 200-plus pages of regulations create enormous complexity for future negotiations. Simplicity benefits both companies and allies in understanding these policies. While I appreciate the nuanced logic behind these complex distinctions, countrywide controls offer valuable simplicity.
Dylan Patel: The real question is exactly what Greg said: “How tough will they be on China?” While they initiated these measures, they ultimately relented with ZTE. They didn’t follow through completely, allowing ZTE to survive and continue growing. The core question remains.
Banning 300-millimeter equipment seems like an extreme measure. Perhaps they’re just accelerating the tightening of restrictions. Most people presume they’ll take a tougher stance — they’ll certainly appear tougher, but the extent remains uncertain. If they were to ban all 300-millimeter equipment, it would completely halt the Chinese equipment industry, though such a drastic step seems unlikely.
Jordan Schneider: Different question. If you had half an hour with Howard Lutnick to pitch the right export control policy, what would your key points be?
Dylan Patel: First, don’t listen to tool company lobbyists — they’re motivated to maintain loopholes that allow them to continue selling for another year, worth over $5 billion to them.
Regarding tools being multipurpose: should we maintain the 14-nanometer logic threshold? Even above that, China has achieved significant indigenization in their military equipment, which the US lacks. Is that the right boundary? Moving it to 28 nanometers would eliminate many dual-purpose equipment issues. At 14 nanometers, some 20-nanometer equipment might work for 7-nanometer applications.
We must consider China’s breakthrough innovation capabilities. They’re developing interesting technologies beyond EUV. We could restrict these areas — for example, Zeiss lenses to China face minimal restrictions. Looking up the supply chain is crucial because even if China achieves breakthrough innovation in tools, they’d need to replicate entire companies like Zeiss and others across the industry.
Understanding the primary goal is essential. If it’s slowing China’s AI chip development to limit their economic and military projection power over the next decade, there’s much more to address beyond AI chips, though they remain the primary focus. The strategy should be tactful — ban subcomponents first, then tools at a lesser level, followed by chips at an even lesser level. This framework still needs refinement.
South Korea presents a crucial consideration, particularly regarding Samsung and SK hynix’s large Chinese facilities. We need their alliance while preventing IP transfer from their Chinese operations. Perhaps CHIPS Act 2.0 could provide significant support to Samsung and SK hynix in the US.
The diplomatic approach with South Korea requires more finesse than with the Netherlands. Dutch companies only make tools and rely heavily on US supply chains — while Korean manufacturers like CMS rank seventh globally in tool production. Their fabs lead in certain areas with significant Chinese capacity. We can’t simply impose blanket bans without considering the implications for Samsung.
Closing loopholes seems straightforward, but the strategic objectives and precise targets require careful consideration.
The current strategy resembles a jigsaw puzzle. Give a hundred-piece puzzle to an eight-year-old, and they’ll complete it. Remove one piece — they’ll still figure it out. Take away ten pieces — it becomes much harder. Remove fifty pieces — they can’t finish it. Remove all the edges — they’re completely stuck.
Right now, the strategy involves removing just a few puzzle pieces.
Greg Allen: And they’re doing it one at a time, giving China time to stockpile.
Jordan Schneider: Not to mention announcing it in Reuters six months before removing the puzzle piece — saying nothing of listening to Gina Raimondo’s phone calls. It’s all publicly available outside of paywalls.
Dylan Patel:This strategy is clearly failing. They remove a few puzzle pieces, but China responds by stockpiling equipment, accumulating HBM, buying subsystems, and dedicating significant engineering resources to solve each banned component.
Take high-aspect ratio etchers for 3D NAND: because the ban was telegraphed, they purchased substantial Lam Research equipment beforehand, including years of spare parts. They positioned new tools beside foreign equipment, analyzed the data from both, and now YMTC is close to developing domestic high aspect ratio etchers. The quality might not match Lam Research, but it’s progress. This happened because the 2022 restrictions for 3D NAND only removed one puzzle piece.
The key insight is that you need a shotgun approach, not a scalpel. If you precisely target one linchpin technology, they’ll solve it with their substantial engineering talent, capital, and industrial base. A shotgun approach increases both cost and time requirements — if you force them to simultaneously solve ten different technologies, splitting their engineering resources, they’ll advance more slowly and fall further behind in AI development.
Jordan Schneider: The irony here is fascinating:
If you sell them the complete puzzle, they won’t learn to manufacture pieces — there’s no incentive.
With a shotgun approach, they might decide it’s too challenging and redirect resources to other sectors like EV batteries.
However, America’s current approach of leaving enough scaffolding actually creates the perfect industrial-policy scenario. Companies typically avoid researching existing technologies when ROI is low, but the Swiss-cheese nature of restrictions over the past two years keeps them in the game, pushing indigenization further than if the US had either implemented dramatic FDPR in 2022 or continued selling everything.
Greg Allen: Say you and your spouse are choosing where to build your house: you’ve selected the neighborhood, but are still debating which side of the street. The dumbest thing you could do is compromise and build your house in the middle of the street. You can make logically consistent arguments for selling almost everything or almost nothing to China. The illogical approach is telegraphing your intention to restrict China while leaving numerous loopholes that undermine the strategy’s effectiveness.
These policies emerge from political compromises, which can be problematic. However, the “sell everything” scenario wouldn’t have ended well either. We sold everything regarding solar manufacturing equipment, and China now dominates that industry. The same happened with electric vehicles. Chinese policy documents and industry patterns don’t support the hypothesis that unrestricted semiconductor sales would have yielded positive outcomes. At this point, we’re committed to the export control strategy — we need to implement it effectively.
Jordan Schneider: Let’s create an alternate history. Up until October 2022, we sold everything to China. Huawei controlled one-third of global market share while Apple struggled in China. Meanwhile, SMIC was approaching competition levels with Intel, TSMC, and Samsung...
Dylan Patel: In this alternate history, Huawei had unlimited purchasing power until the Trump administration implemented restrictions. Huawei became TSMC’s largest customer and dominated Apple in the Chinese phone market. They emerged as the world’s largest phone manufacturer — not quite as profitable as Apple, but they were getting there. They dominated global telecom equipment markets, only facing resistance in regions where we explicitly banned their equipment due to security concerns, despite their technical superiority.
When companies have unrestricted purchasing power, they overtake industries. Take SMIC, for instance. With unlimited access to resources, they achieved 7-nanometer technology independently. Even though they could access TSMC’s 7-nanometer technology since 2018, SMIC still developed their own capabilities and found a market for it.
Their capacity today would be significantly larger without restrictions. Consider NAURA before the October 7, 2022, restrictions. Why did they maintain hundreds of millions in revenue when Applied Materials and Lam Research could sell freely to China? Because China’s industrial policy focuses on replication and building domestic supply chains. In an unrestricted scenario, it’s like giving them the complete puzzle, which they then recreate independently. Now, we’re only withholding one piece, yet they’re still determined to complete the puzzle themselves.
‘We Must Decide’
Jordan Schneider: Final thoughts — what’s the “America First” argument for investing in domestic semiconductor industry while restricting China’s semiconductor development?
Dylan Patel: Making American chips great again requires more than just the current CHIPS Act. $50 billion barely scratches the surface — Intel alone spends $20 billion annually on R&D, plus additional capital expenditure. The current allocation represents less than one year of spending, with Intel receiving under $10 billion spread across multiple years.
The renewable energy subsidies in the Inflation Reduction Act represents about the same cost as securing even 5% domestic market share in chips. The semiconductor industry, where we currently hold significant market share, requires proportionally less investment. Industrial policy must be implemented before we lose our competitive edge.
To maintain at least 20% market share in memory, advanced logic, and other sectors, we need to act now. The cost increases dramatically if we wait five years. Tariffs alone won’t relocate chip manufacturing since the focus should be on end systems and servers. Manufacturing servers should happen in places like Vietnam and Mexico.
We need industrial policy that encourages significant capacity development. Should TSMC allocate 15-20% of their leading-edge capacity here, or should we aim for 30-40%? This goal is achievable with modest additional investment relative to government spending. Companies like Samsung, SK hynix, STMicroelectronics, and Infineon should be manufacturing in the US.
The CHIPS Act focuses primarily on leading-edge technology. We need expanded funding for both leading-edge and trailing-edge technologies to counter China’s dominance in the latter. Between 2022 and 2025, China’s IGBT [insulated-gate bipolar transistor] capacity growth exceeds the world’s existing capacity. While their yields may initially be lower, they’re positioning to control 50% of global capacity in power semiconductors. This creates significant supply chain security concerns that require strategic industrial policy rather than blanket restrictions.
Greg Allen: My question regarding everything you said is that Donald Trump considers himself “tariff man” and loves tariffs. The current tariffs on Chinese semiconductors apply only at the chip level when shipped as standalone items. While it would be complex to apply tariffs to the component value of chips in finished goods, it’s not impossible.
I’ve been wondering if the Trump administration might say that they don’t want what Trump has called “corporate welfare” through the CHIPS Act. Instead of industrial policy through subsidies, they may prefer industrial policy through tariffs. The current tariffs on Chinese semiconductors aren’t effective, but a different approach to tariffs might work. Though I’m not certain this is what they’ll pursue, it seems consistent with their messaging.
Dylan Patel: The question is, “Are you going to tariff electronic systems manufactured in China? Are you going to tariff 90% of iPhones?”
Greg Allen: We’re entirely speculating here, but I think they would say if an iPhone contains Chinese chips, the tariff applies based on the value of those Chinese chips. We’re always tariffing chips, whether they arrive in a box labeled “chips” or in telecommunications equipment.
Dylan Patel: Presumably it would be tiered — Chinese chips at a 500% tariff and Taiwan chips at a 10% tariff.
Greg Allen: Exactly. All this Chinese legacy buildout we’ve discussed — some of which might be advanced node production disguised as legacy node — could become the industrial equivalent of those ghost apartment buildings in China. If there’s no end market for these Chinese semiconductors, their industrial policy would be a disaster. They would have built a bridge of subsidies to nowhere. While I haven’t heard from Howard Lutnick or others in the Trump administration that this is their planned policy, I could see this approach being attractive.
Dylan Patel: But if you want to prevent China from gaining global market share in trailing chips outside of China, the primary task is moving electronic manufacturing out of China. The US market share for most products — excluding high-end AI servers — is only about 30% to 40%. For AI servers, it’s around 70%. We can dictate policy on AI servers, assuming we resolve the data center shortage, which requires significant regulatory changes.
Greg Allen: You’d have to persuade Europe and Japan to participate.
Dylan Patel: Exactly. Otherwise, why wouldn’t Xiaomi phones — which hold 20% global market share — and other Chinese phone makers like OPPO simply use Chinese RF chips, power management ICs, and antennas? They clearly will, unless we can move both manufacturing and vendors out of China.
Consumer goods, especially phones, are dominated by China. For laptops, you’d need to convince Dell and HP — through their ODMs [original design manufactures] like Compal — to completely relocate to southeast Asia, India, or elsewhere. A tariff on chip value made in China doesn’t solve this issue.
Since we’re speculating about the Trump administration’s approach, why not be more heavy-handed? We could tariff everything shipped from China, with lesser tariffs on Taiwan and southeast Asia. This would make moving out of China a massive cost saver — perhaps not enough to justify Mexico, but definitely southeast Asia.
Greg Allen: We’re at a point in the story where the Biden administration has assessed the policy toolbox created by the first Trump administration. Now we’ll see how a second Trump administration utilizes the toolbox Biden’s team has created. While some people in DC — certainly not me — may be tired of the semiconductor and AI great power competition narrative, I don’t think it’s going anywhere. This will remain a significant part of geopolitical competition and a key focus for the Trump administration.
Jordan Schneider: I got one more riff.
The intellectual- and execution-level challenges the Biden administration encountered with export controls exemplify broader Democratic Party challenges. There’s a tendency to believe they can devise the perfect algorithm that balances all competing interests. They think with solving enough integrals, extensive legal review, and track changes on docs, they’ll reach the optimal solution.
This pattern emerged with the Inflation Reduction Act’s lengthy development, the CHIPS Act’s extended negotiations, the periodic reassessment of Ukraine arms distribution, and these export controls. The problem is that, if you can’t make tough strategic decisions upfront and execute them — accepting that not everyone will be happy — you end up in limbo. You achieve worse results by trying to moderately satisfy five variables instead of maximizing the two most critical ones.
Greg Allen: Another way to put it: faster and good enough is almost always better than slower and theoretically perfect.
Jordan Schneider: As a new American dad returning from paternity leave, I’ve been exercising by birthright by reading Civil War history. There’s an excellent quote from Colonel James Rusling’s memoir about how Grant made decisions.
The irony is that October 2022 really felt like a decision point. Jake Sullivan gave a dramatic speech stating we needed to stay as far ahead of China as possible in critical strategic emerging technologies. A month later, ChatGPT emerged, clearly demonstrating AI as the critical emerging strategic technology. They were onto something, but now we’re left with this muddle.
ChinaTalk is a reader-supported publication. To receive new posts and support our work, consider becoming a free or paid subscriber.
The global race to build humanoid robots is heating up, and Beijing aims to dominate the industry by 2027. Our deep dive today explores:
Why it’s worth it to build humanoid robots instead of the strictly industrial robots we covered in part 1 of our robot series
How leading Chinese players stack up against Tesla and Boston Dynamics and where China is still reliant on western technology
The challenges of data acquisition for LLMs vs for humanoid robots
Why the Chinese auto industry is key to humanoid robot success
Human toddlers, on average, take 17 tumbles and toddle 2,368 steps each and every houras they learn how to walk. By the age of two, children make walking look easy. But make no mistake — after a century of research, neuroscientists still don’t fully understand how the human brain learns to execute such a wide array of complex physical activities.
Natural selection has refined the human form over billions of years. And yet, companies around the world are now betting that they can artificially imitate the human body on a much shorter timescale, to create AI-driven general-purpose bi-manual, bipedal robots.
The question is, will this marriage of AI and robotics (also known as ‘embodied intelligence’ 具身智能) produce viable offspring?
Chinese edition of Philip K. Dick’s novel, Do Androids Dream of Electric Sheep? Source.
The allure of humanoid robots
Generative AI and humanoid robotics seem like a perfect match. If combined successfully, they could replicate optimal human performance on a massive scale.
By mimicking human form, humanoid robots can operate in working environments originally designed for people (eg. mine shafts). By mimicking human behavior, humanoids can integrate more easily into human social and emotional contexts (e.g. waiting tables or modeling clothing).
In the face of pervasive labor shortages, Goldman Sachs stuck a finger in the wind and guessed thatthe global market for humanoid robots could be worth US$38 billion in a decade.1 With such a large potential payoff, western firms like Tesla, Boston Dynamics, Figure AI, and Apptronik are making huge investments in humanoid robot development. But at leastone third of that global market value will come from imminent Chinese demand for humanoids — and the CCP wants to make sure those industrial androids are home grown.
In 2023, China’s Ministry of Industry and Information Technology proclaimedthat China would aim to be the world’s top producer of cutting-edge humanoid robots by 2027. Models by Chinese firms like AGIBOT, Astribot, and Galbot threaten to outcompete Tesla’s Optimus bot, thanks in part to Chinese advantages in supply-chain integration and mass production.
Given China’s shrinking labor force, general-purpose bipedal robots have clear appeal. Of course, humanoids could revolutionizeindustries characterized by dirty, dangerous and demeaning work,2 such as agriculture, construction, manufacturing, mining, or transportation.
But humanoids could also be valuable in the service sector. As personal assistants, personalized tutors, and caregivers for the elderly and children, humanoids could automate rote daily monitoring tasks while offering companionship. Maybe one day they could even provide entertainment as street performers.
Finally, China’s existing prowess in industrial automation could serve as an additional motivator. With the power to directly access, operate, and repair existing automation and computer systems, humanoids could unlock a number of creative multiplier effects that we analog humans haven’t even imagined yet.
So, what’s needed for these sci-fi fantasies to become day-to-day reality?
A multi-functional humanoid robot requires advancements in both hardware and software. Beijing’s Embodied Intelligent Robot Action Plan 具身智能机器人行动计划 breaks this down into three core technologies: robotic body components (the “limbs” 肢体), motion control and balance (the “cerebellum” 小脑), and AI (the “brain” 大脑). Designing every part comes with trade-offs — complexity, power systems, weight, and size each influence cost, durability, stability, and control.
Hardware for the “limbs” 肢体 needs to mimic the complex array of joint and muscle movements possible in a human body. In robotics, the concept of “degrees of freedom” (DOF) refers to the number of independently controllable joints on a robot. A basic robotic arm might have three DOFs (forward/back, left/right, up/down). In contrast, a functional humanoid robot might require a staggering 28 DOFs just for its limbs (3-DOF hip or shoulder, 3-DOF ankle or wrist, one-DOF knee or elbow). Robot hands are a major challenge — human hands have at least 27 DOFs, a difficult target for hand-sized hardware to achieve. Recently, Shanghai-based Fourier launched a model with 12-DOF hands; Tesla’s Optimus is meant to upgrade to 22-DOF hands by the end of 2024. A robot that can manipulate objects at the level of an experienced human worker on the factory floor remains elusive.
Humanoid robots at the 2024 World AI Conference in Shanghai. Source. The lady robot in the front row was photoshopped out in Xinhua’s report on this conference.
For safety, these humanoid bodies need to be stable, accurate, and reliable. The physical world tends to be less forgiving of mistakes than the digital world—one wrong move could cause a seventy-kilogram metallic mass to crush a nearby object or human, so no room for error. Achieving coordination, balance, and posture in the robotic “cerebellum” 小脑 requires complex autonomous control systems that integrate sensory inputs into motor outputs. And, with all that out of the way, these robot bodies are still not fit for commercial use unless they can run reliably without frequent maintenance.
Meanwhile, the AI-powered “brain” 大脑 needs to substitute for human thinking and behavior.
In contrast to internet AI (those that only operate online), embodied AI learns by interacting with a physical environment. Such AI-powered robots need to be able to continuously monitor, process, and quickly respond to massive quantities of sensory input in real-time. Can the robot pick any object from a messy pile, shift it to the correct orientation, identify it, and transport it to the right place, all without damaging it? These are the kinds of questions that robot “brain” developers are asking.
Moreover, one of the key targets for a general-purpose robot brain is emergent behavior — defined as a robot’s ability to perform actions not present in its training data, such as catching an unexpected falling object. Robots have yet to master the “commonsense knowledge” to handle everyday environmental variations that humans take for granted. Even something as simple as pouring tea into a mug has countless “edge cases” that would challenge a robot. What if the mug is upside down? What if the mug is already full? What if the mug accidentally falls? Composed of many actions in a sequence, such seemingly simple tasks have long time horizons with many opportunities for errors to compound.
Techniques like end-to-end neural networks can help companies to research, develop, and iterate their products more quickly.3 But creating humanoid robots fundamentally requires big science — that means big datasets, investment budgets, talent pools, and teams of collaborators. The time has not yet come for them to break into reality.
But if not now, when?
Where data might come from
As with many AI applications, many researchers argue that enormous training datasets are the key to develop a general-purpose robot. But obtaining this data is not easy.
There is no internet-equivalent that can spin up a data flywheel for AI+robotics. Comprehensive robotic datasets often require sensory, motion, environmental, interaction, social, and task-specific data. That requires a lot of time, money, and coordination. The diversity of robot shapes and sizes, along with the variety of environments in which they can be deployed, complicates data collection further.
With tools like NVIDIA’s Isaac Sim, Researchers can generate synthetic data and run virtual simulations to train and test their humanoid models. These methods are increasingly advanced and safer than real-world operations, but synthetic datasets still risk producing results that are incomplete, biased, inaccurate, or ungeneralizable. Ultimately, before deployment, a humanoid robot must be trained and tested in real environments.
But where?
Automotive industry, meet Optimus
The automotive industry — in China and elsewhere — is full of problems that humanoid robots could help solve.
Manufacturers are grappling with the global EV reckoning, a fiercely competitive export market, and supply chain uncertainty. Meanwhile, consumers' tastes have grown more complicated than ever, as demonstrated by the rising popularity of built-to-order models.
The dwindling automotive workforce isn’t enough to handle these challenges. In China, government data forecasts a 1.03 million shortage of talent in the new electric vehicle industry by 2025.
The answer lies in the fact that car manufacturers can provide data and training environments that robot designers desperately need.
Factories and warehouses are “behind-the scenes” use cases in which a general-purpose robot can train and prove value without high costs of failure.
Manufacturing facilities already have structure and safeguards, and are only occupied by people with specific safety and hazard training.
Vehicle manufacturing is an especially good fit for training and testing humanoids. Standardized, process-oriented tasks like handling, sorting, welding, assembly, and quality inspection are perfect activities to help robots accumulate training data and build task libraries. Auto manufacturing factories also provide the physical ingredients for a humanoid training gym — varied terrain and dynamic elements from which robotics can learn in a relatively safe and controlled space. Through these “factory internships,” humanoids can perform relatively simple tasks to collect data, learn and generalize, and show practical value for broader commercialization.
Now, Chinese car manufacturers can preorder humanoid prototypes from Shenzhen-based robot manufacturer UBTech — presumably at steep discounts. UBTech’s plan is simple: achieve general-purpose commercialization by first rolling out humanoids in the auto industry and then expanding horizontally into consumer electronics and other industries. UBTech has reportedly already received intention orders for over 500 units from a slew of Chinese automakers. The humanoid collaboration club now includes SOEs like Dongfeng Liuzhou Motor and FAW, the privately-owned Geely Automobile, publicly-listed EV makers Zeekr and Nio, and the multinational joint venture FAW-Volkswagen (which produces VW and Audi cars for the Chinese market).
Similar strategies are taking shape outside of China as well. Figure AI’s first commercial partnership involves deploying robots in BMW’s South Carolina facility. Apptronik is sending its 160-pound bipedal Apollo bot to Mercedes-Benz’s facilities in Hungary, where the company has faced a sustained labor shortage. Toyota is investing in in-house R&D for humanoid robots and partnering with Boston Dynamics. The automotive sector is the largest source of new robot installations in all of North America, with many partnerships going beyond training to include collaboration in the eventual production and use of humanoids.
According to an anonymous industry insider, once humanoids are viable for factory applications, consumer applications could follow within two to five years.
Given that a robust manufacturing sector is critical for national defense, the auto industry’s adoption of humanoid robots could have far-reaching geopolitical implications.
Moreover, because of the unique difficulties associated with high-quality training data for humanoid robots, any entity with high-quality, proprietary, real-world data locks in an immense incumbent advantage.
China’s Strengths and Weaknesses
While data is limited, it appears Chinese humanoid models are behind the global cutting edge in a few areas.
A report from the US-China Economic and Security Commission finds that Chinese firms are on par with the US regarding robot weight, height, and speed, but lagging on key sensor technologies.
A recent Goldman Sachs analysis reveals that while both global and Chinese entities are proficient in AI “navigation,” technology is still lacking in “manipulation” and “interaction” abilities, with China slightly behind international competitors in these areas.
For a few specific hardware components — such as planetary roller screws and sensors —- China’s domestic companies seem to encounter bottlenecks not faced by their global counterparts. Chinese humanoid companies also rely on US-based NVIDIA for processing units and software. Nevertheless, some hardware suppliers like Shanghai KGG, humanoid manufacturers like Kepler Robotics, and AI companies like Huawei have made attempts to help the industry move towards localization.
However, when it comes to the inputs for humanoid robots, China is competitive thanks to its low-cost and manufacturing advantages. By taking a “fast-follower, rapid scaling” strategy, Chinese companies may become global leaders in humanoid manufacturing, even while relying on foreign innovations.
In China’s fragmented landscape of over 3,400 robotic startups, there are a few players who might become leaders in innovation as well. Two firms worth highlighting:
Unitree Robotics: producing both quadrupeds and humanoids, this company has been responsible for flashy displays at the Super Bowl and the Winter Olympics. Its H1 humanoid demonstrates dynamic motion capabilities, including a record speed of 7.38 miles per hour. The robot will be rolled out at a price around USD $90,000, comparable to models by Tesla and Boston Robotics.
Fourier Intelligence: specializing in medical and rehabilitation robots, this firm started its GRx series of general-purpose bipedal bots in 2023. GR-2, the latest edition, offers 53 degrees of freedom, longer battery life, and a streamlined design. Not yet commercialized, the G-2 is compatible with open-source software like MuJoCo and NVIDIA’s Isaac Lab for further robotic development.
Bolstered by government support, more advancements in Chinese robotics are likely forthcoming. Consider the timeline of Chinese electric vehicles: after prioritizing EVs in national economic policy throughout the 2010s, China is now recognized as a global leader in 2024. If Beijing follows through on its recent commitment to humanoid robotics, it’s not unreasonable to imagine significant strides in the next decade.
ChinaTalk is a reader-supported publication. To receive new posts and support our work, consider becoming a free or paid subscriber.
Zooming Out
Humanoid robots are a highly visible way for China to demonstrate its progress in robotics relative to the rest of the world. However, it’s important to remember that improvements in embodied intelligence — and the enhanced supply chains that follow — will come with knock-on effects that impact the rest of the economy. For example, improvements in humanoid technology could spill over to unlock new alternative robotic forms. After all,the human body does have its flaws.
Regardless of form, AI-driven robots raise thorny safety and ethical questions. A robot in the real world can collect all sorts of data — biometrics, building layouts, social behavioral data, continuous streams of audio and video, and more. Few regulations currently exist to safeguard transparency and personal privacy in the face of this increasingly robotic future. The IEEE’s study group to develop a roadmap for standards has only just started, and is set to release findings next year.
The Taiwanese cover of Isaac Asimov’s I, Robot. Source.
In the scenario that China dominates the humanoid robot market, some US politicians worry that the presence of Chinese-made robots on US soil could threaten national security (much like the FBI’s concerns about drones made by DJI).
There are concerns about unregulated humanoids in China as well. At present, no official safety standards exist for humanoid design. A group of Shanghai industry leaders published China’s first-ever governance guidelines for humanoid robots in July 2024, in which they floated a number of proposals including risk controls, emergency response systems, consensus-based regulation, ethical use training, and global collaboration guidelines for humanoids.
With industry leaders so eager to make their Asimovian fantasies a reality, ethical concerns are likely to arise within the next decade. The US and China have an opportunity to shape global governance frameworks together.
Admittedly, this will be difficult given geopolitical tensions. But without cross-border collaboration, firms can lobby against ethical regulations on the basis that international competitors won’t be bound by the same standards.
Behind fantasies of robot bartenders, technology is steadily advancing as hardware and software meet biology, neuroscience, and psychology. Our society is not prepared.
Still curious about humanoid robots? ChinaTalk has you covered. Write your questions in the comment section below for a chance to get insight from an anonymous industry expert coming soon to ChinaTalk.
Goldman Sachs’ original projection was US$6 billion by 2035. They recently revised it up to US$38 billion citing rapid advances in AI, reduced component costs, demand push factors (eg labor shortages), and broader, deeper supply chains.
Intuitively, end-to-end learning refers to the process of “training a single model to perform a task from raw input to final output, without any intermediate steps or feature engineering.” Tesla has implemented end-to-end neural networks in the development of their full self-driving features.
Login
書衣偵探華斯比评论: 100年，接了个龙
