a16z: The United States cannot lose the Bots competition.

Source: Martin Casado, a16z partner; Anne Neuberger, former Deputy National Security Advisor, currently a senior advisor at a16z; Translated by: Golden Finance

China Achieves the Impossible

AI is reshaping global soft and hard power. Notably, the United States has taken the lead in soft power. Leading LLMs are trained on Western texts, and American companies still dominate the global training and inference space, maintaining a leading position in the overall token generation market share competition.

But as it stands, China is far ahead in the hard power of AI—robotics technology. As AI continues to make incredible strides, we are beginning to see intelligence embedded in the physical world—ultimately leading to the birth of general-purpose robots capable of performing various tasks across fields from manufacturing to services and defense. This will redefine every aspect of our society and reshape daily life. The country betting on this future is China, not the United States.

In the decade since the Chinese government announced the "Made in China 2025" strategy, the per capita robot density of Chinese enterprises has far exceeded that of other regions in the world. In 2021, China surpassed the United States, and in 2024 it will surpass Japan and Germany, which are known for their automation, and will soon outpace the remaining competitors, Singapore and South Korea. China has rapidly become a global powerhouse in robotics. Fully autonomous "dark factories" like Xiaomi, the smartphone and automotive manufacturer, operate in complete darkness without human supervision.

China has successfully achieved what we once thought was impossible. Just ten years ago, we laughed at the notion that "China can copy, but cannot innovate," and later revised it to "they can innovate, but cannot manufacture high-precision upstream tools." Perhaps we should not be so complacent, as Chinese companies have outperformed their counterparts in various industries—from solar photovoltaics (where competitors outside China have been almost completely defeated) to 5G (the global deployment of 5G has been a tremendous success for China's leading domestic company, Huawei). Today, the same pattern is unfolding in the field of robotics. China has established a strategy to dominate strategic industries and has utilized this strategy to become a superpower in robotics.

Today, local Chinese companies are capable of designing and manufacturing precision components such as harmonic reducers, with competitive quality and lower prices. And—most importantly—they are co-located with customers in manufacturing superclusters. This is what should truly scare the West. So many manufacturers, assemblers, and customers of robotic tools are gathering at nodes like Shenzhen or Shanghai, and it is in this way that new combinatorial use cases are discovered, manufacturing sequences are optimized around new potentials, and companies are able to develop advanced process knowledge that the West completely misunderstands. In a few years, Chinese companies will produce components that we cannot replicate—not only at a low cost but at an extremely low cost. This resembles the past. In the 1970s, Japan shocked the world with Toyota's lean production methods, just-in-time inventory, and the philosophy of continuous improvement to eliminate waste. Initially overlooked, Japanese automakers had surpassed American and European giants by the 1980s, reshaping the global automotive industry. If we do not take action to prevent this, it will be another Toyota moment, but on a much larger scale.

If we do not take action quickly, the United States will struggle to catch up: we are entering a period of compound improvement that could make China's advantages nearly insurmountable. Like LLMs, training advanced robotic systems requires internet-scale pre-training data, along with reinforcement learning to train general strategies—these strategies need to be able to reason in environments where perception and tasks exhibit various biases. As data from real-world deployments comes online, countries with more robots will gain a flywheel effect; more deployments mean more high-quality data, which will ensure further deployments. The U.S. has not completely fallen behind, and our leading position in AI software will continue: American companies like (World Labs) are at the forefront of building cutting-edge models that enable robots to reason in three-dimensional space. However, as these capabilities mature, actions in the real world—from routing wire harnesses through chassis paths in electronics assembly to simple laundry—will unlock the economic and strategic potential of general robotic technology.

Micron Tolerance

To understand the achievements China has made in recent years, let's talk about the harmonic reducer—a seemingly simple manufacturing part, but very difficult to produce.

A harmonic reducer is a type of gear system that resembles the shoulder or elbow in a sleeve. They transmit rotational energy from one end (usually powered by a motor providing high-speed power) to a low-speed, high-torque gear device. Its working principle is based on slightly offset inner and outer gear rings, equipped with a rotating elliptical component on the inside. When driven by a motor, it generates waveforms that slowly drive the outer sleeve with a high gear ratio and high torque—suitable for various robotic applications, including humanoid robots.

The challenge of manufacturing these tools lies in their sensitivity to small deformations during the processing and operation processes. To ensure they can function properly, they must be manufactured with micron-level precision at low cost. When these sleeves are linked together to form a system with multiple degrees of freedom, such as multiple joints on robotic fingers, hands, or limbs, higher precision is required. Achieving the strength and flexibility of human hands at a reasonable cost requires true excellence in manufacturing.

The precision required to manufacture harmonic reducers far exceeds the capabilities of most mechanical workshops. Historically, production has been dominated by highly specialized German and Japanese manufacturers: Japan's Sumitomo and the German-Japanese joint venture Harmonic Drive are the two major players in the field, accounting for a combined 95% of the global market share. However, in recent years, they have faced increasingly fierce competition from new entrants in China. A company called Green Harmonic, located in Suzhou near Shanghai, produces harmonic reducers that can compete with Sumitomo and Harmonic Drive's products in performance, but at prices that are about 30% to 50% cheaper. Green Harmonic currently holds over 30% of the market share in China and is soon turning its sights overseas. In the coming years, we can foresee that companies like Harmonic Drive will face their 'Toyota moment,' which will have significant strategic implications: countless Chinese enterprises will transform low-cost, reliable manufacturing into global market share and ultimately drive competitors out of the market.

Harmonic reducers are just a typical example of the hardware stack of robots. Building a fully functional robot requires a variety of other small components—precision bearings that ensure smooth rotation of joints, custom printed circuit boards for transferring power and signals between subsystems, dedicated connectors that maintain reliable communication in high-vibration environments, miniature encoders that provide millimeter-level precision position feedback, force-sensitive resistors embedded in fingertips for fine manipulation, inertial measurement units that can accurately track angle changes, servo motors with complex current control algorithms, shielding layers to prevent electromagnetic interference between tightly packaged electronic devices, thermal interface materials for high-performance processors to dissipate heat, and countless fasteners, washers, and protective enclosures, all of which are designed to withstand mechanical stress during actual operation. Each component must be carefully selected, considering not only its individual performance characteristics but also how it integrates with the broader system: a single point of failure can render a complex robot completely inoperable.

From controller manufacturers Siasun and Estun to torque sensor manufacturer AVIC Electromechanical, Chinese companies are rapidly entering the robotics market and beginning to capture market shares of various components of the system. These companies, along with countless others, form a mature and complete ecosystem that enables Chinese enterprises to locally source nearly an entire set of robots—both domestically in China and within large industrial clusters like Shenzhen.

Today, domestic manufacturers and their suppliers in China provide all the components needed to realize the dream of robotics, continuously learning from each other. The Chinese startup Yushu Robotics has captured global attention with its affordable and highly advanced robots—agile and LLM-integrated robotic dogs priced as low as $1,600, and humanoid robots priced at $5,900. These costs will continue to decrease; robotic dogs will also become increasingly powerful, with stronger capabilities.

How the West Failed in the Contest of "Open vs. Closed"

How has China so rapidly integrated with the global forefront? Like other sectors of China's industrial sector, national subsidies have played a facilitating role. Many Chinese cities offer full tax exemptions on R&D expenses, generous subsidies, and favorable corporate income tax rates to robotics companies. Although China's national support for other strategic industries, such as electric vehicles or batteries, is far greater than that for the robotics industry, this accumulated economic aid undoubtedly contributes to the development of domestic industries and keeps barriers for external entrants in the Chinese market.

But focusing only on these subsidies overlooks the most important part. The vigorous development of the robotics industry in China benefits from the unique advantages provided by the Chinese market. The scale of China's industrial sector offers a variety of workshop scenarios to train and reinforce robots, while the densely packed industrial clusters around Shanghai or Shenzhen provide endless opportunities for co-location and close cooperation across all links of the supply chain—from component suppliers like Green Harmonic to end users like BYD. Chinese factories quickly adopt new technologies and benefit from ultra-high-speed digital infrastructure: they are now among the most connected factories in the world, equipped with high-definition cameras and precision sensors that collect data through low-latency 5G connections. Beneath this core level of Chinese industry, there is also a large pool of low-cost, high-skilled labor that can produce complex hardware at costs and speeds that are almost unmatched anywhere else in the world.

Most importantly, all of this happens in a vibrant, competitive, and open ecosystem: new enterprises and new technologies are constantly emerging, while regulatory barriers are far less than in other countries. The natural selection of market forces quickly eliminates uncompetitive businesses, allowing companies like Yushu Technology to rise and join the ranks of global top enterprises. Chinese entrepreneurs may enjoy national support in strategic areas, but more importantly, they benefit from the rapid innovation and fierce competition of a highly dynamic open market.

The reason why the United States was able to win key industries in the twentieth century is not much different from China's winning strategy today: a vibrant open market, sometimes coupled with strategic government support, created a number of groundbreaking American companies, unleashing tremendous economic value—from Boeing, Lockheed Martin to IBM and Intel. But now, we are undermining our historic advantages. Not only have we failed to thoughtfully update the regulations established to meet past challenges—before we face the prospect of global competition with China—but some policymakers are also busy adding new barriers, such as administrative requirements on AI startups or proposing restrictive copyright measures at both federal and state levels. China is doing everything it can to ensure it wins the industries of the future. The United States, however, is not doing the same.

We know the ending of it all

When Marc Andreessen wrote the phrase "software is eating the world", few realized to what extent all businesses would become software companies and all organizations would turn into software interfaces. Today, the same thing is happening in the real world: robotics will thoroughly penetrate all manufacturing industries and every corner of the real world.

However, the transformation driven by the internet occurs against the backdrop of the government recognizing the benefits of innovation and encouraging a relaxed regulatory regime. Today, these changes will play out in the heavily regulated workplaces and factory floors of the United States. From drones to sidewalk delivery robots, we implement a "license-first" regulatory system across all sectors, meaning any company interested in robots must spend time and money dealing with lawyers, applying for permits, and enduring endless regulatory approvals. For more established companies, agreements reached with labor unions make automation extremely difficult: dockworker unions have been fighting against automation for years, leading to U.S. ports becoming some of the slowest, least automated, and least efficient ports in the world. In short, we are not prepared to win.

Given that it is extremely difficult for businesses to introduce any form of automation, it is not surprising that many sectors of the U.S. economy are defined more by cost disease and rising prices than by productivity gains. From 1970 to 2020, labor productivity in the construction industry declined by more than 30%. Even in manufacturing, which can typically withstand cost disease, U.S. labor productivity has also declined over the past 15 years.

The United States does not have "black factories." The closest is Tesla's Nevada Gigafactory, which has an automation level of 90%. No other major manufacturer can compete with it. Due to our highly stringent regulatory approach limiting the demand and supply of robots—startups find it difficult to manufacture robots, and enterprises find it hard to purchase them—the influence of the U.S. robotics industry on a global scale is minimal. The U.S. does not have a self-sufficient robot supply chain. There are some robotics companies in the U.S., but they almost always rely on Chinese suppliers: products labeled as "Made in America" still use core components sourced from China. Even components procured from allied countries inevitably depend on China—for instance, the CMOS image sensors dominated by Japan's Sony not only rely on neodymium and dysprosium from Chinese refineries but also on lenses produced by Chinese companies like Sunny Optical.

The development trajectory is evident. Chinese companies are destined to dominate one segmented market after another and create oversupply until all competitors' production becomes unprofitable. Foreign suppliers are feeling competitive pressure from China; some suppliers have been directly acquired by Chinese companies. KUKA, a large industrial robot manufacturer based in Germany with a history dating back to 1898 in Bavaria, was acquired by a Chinese company in 2016, and the German government only changed its foreign direct investment policy after the acquisition was approved. However, there are also some signs of recovery—thanks to Taiwan's TSMC, the United States maintains an advantage in advanced logic chips, with Nvidia's Orin being the most popular advanced robotics chip. But Yu Shu's G1 (one of the most advanced humanoid robots on the market) does not use Nvidia chips. It uses the Chinese-made Rockchip RK3588.

What do we need to do

It is now time for the United States to take robotics technology seriously. With the emergence of generalized physical intelligence and robotics becoming a major force reshaping countless key industries, the absence of a robotics industry will become a critical weakness in U.S. strategy. This is not much different from our reliance on importing all key network and telecommunications equipment from Huawei.

In the domestic market, we need to stop constraining our robotics industry. The market discipline of competing with Chinese suppliers can have a positive impact on American companies: steel sharpens steel. But for American companies to realize these benefits and truly engage in competition, they need the freedom to experiment and iterate, so they can genuinely compete with their Chinese counterparts. We need to shift from "permission-first" to "permissionless."

Even if we readjust regulatory policies to focus on promoting innovation, we still cannot catch up with China in a timely manner. The ecosystem is not built in a day. However, with our domestic production capacity, along with the capacities of allies like Germany, Japan, and South Korea, we can still form an almost complete and powerful robotics supply chain, and these weaknesses can be identified and addressed.

As Chinese companies seize market share, replacing existing firms in Europe and Asia, our foothold is being wasted. The United States and its allies need to coordinate our interests, identify our existing effective resources, and allow our startups to thrive.

This has sounded the alarm for us. The U.S. leadership has been boldly advancing in defining and defending the AI software stack; now, it must work hand in hand with its most important allies to lay the groundwork for a defensible AI robotic stack. There is still time, but not much left. The most critical challenge in the AI race is about to arrive. We must join the competition.