At electronica China 2026, NXP Semiconductors took "Pioneering Trusted Intelligent Edge Systems" as its theme, unveiling a comprehensive lineup of system-level solutions across three major scenarios: smart mobility, industrial and robotics, and smart living. These solutions include UWB seamless access, Bluetooth precise ranging, and wireless charging for the mass-produced Li Auto L9 Livis; active diagnostics for wireless GenZ BMS; radar edge intelligence that pushes AI perception and decision-making down to the radar edge; the combination of S32N79 and Ara240 building a "central brain" for smart vehicles; and the S32K5 platform showcasing edge AI deployment and MRAM high-performance storage. Additionally, there are closed-loop control for robotic arms based on LeRobot and i.MX 95, and I3C bus dexterous hand solutions; as well as AI medical human-machine interfaces, continuous glucose monitoring patches, and single-chip hearing aid solutions based on the i.MX 95.
Photo: NXP Booth at electronica China 2026
These demos are not merely a list of isolated technologies, but rather cross-sections of the same architectural philosophy applied across different physical scenarios.
NXP's judgment is: Cloud AI has changed "how we think," while Physical AI will change "how we live." Over the next 10 to 15 years, sectors such as industrial, automotive, and consumer electronics will converge. A car is an "agent on four wheels," and a humanoid robot is an "agent on two legs," with their underlying technological capabilities being essentially identical.
The "Incompromisable Triangle" of Physical AI
As agents step into the physical world, they will be with us at all times, interacting with our families and children.
As Mr. Liu Xiaohe, Executive Vice President of NXP Semiconductors and General Manager of NXP China, stated: "If the competition in Cloud AI is about who is smarter, then when it enters the real world, the competition in Physical AI is about who is more reliable. This is because it will be with us at all times, interacting with our families and our children."
Photo: Mr. Liu Xiaohe, Executive Vice President of NXP and General Manager of NXP China
Mr. Liu summarized the three incompromisable prerequisites for Physical AI:
First, energy efficiency. Most Physical AI relies on battery or renewable energy power supplies. Computing power should not be measured solely in TOPS, but in "TOPS per watt";
Second, ultra-low latency. When asking ChatGPT a question, it does not matter whether the response delay is 100 milliseconds or 200 milliseconds; however, if there is a 10-millisecond error in a car's braking system or a robotic arm's locking mechanism, it could lead to unacceptable risks;
Third, safety determinism. Current large language models are essentially probabilistic mechanisms with inherent randomness; whereas when humans interact with the physical world, they require deterministic and predictable mechanisms. Errors are unacceptable in Physical AI.
Only by substantively resolving these three elements can Physical AI become truly "reliable" enough to be accepted by millions of households, and the underlying economic value can be genuinely unlocked.
Cracking "Moravec's Paradox": The Answer from 500 Million Years of Evolution
In 1988, Professor Moravec from Carnegie Mellon University proposed a famous paradox: it is easy for robots to do things that humans find strenuous, such as playing chess or proving theorems, but for robots, tasks that humans find simple, like picking up an egg or catching a ball, are exceedingly difficult.
Mr. Liu stated: "The current mainstream solution in the industry is to stack computing power, which means larger models and larger chips, collecting all data to the cloud for processing. However, NXP has provided an alternative path: seeking answers from biology."
500 million years of natural selection have already provided the optimal solution. The agents of vertebrates are not centralized, but hierarchically distributed. Taking humans as an example, the neural axis consists of three parts:
- Cerebral cortex: Approximately 16 billion neurons, with a power consumption of only 9 watts. It is responsible for planning, learning, and decision-making, with a response speed in the 300-millisecond range, featuring creativity and probabilistic nature.
- Cerebellum: Approximately 69 billion neurons, with a power consumption of only 2 watts. It is responsible for motor coordination and control, with a response speed in the 10-50 millisecond range, featuring high real-time performance and semi-determinism.
- Spinal cord: Approximately 69 million neurons, with a power consumption of only 5 watts. It is responsible for reflex-level safety protection, with a response in the tens of milliseconds range, capable of bypassing the brain to act directly. Because it has a highly robust functional safety mechanism, it can protect itself without the brain's approval.
The three layers of intelligence each perform their respective duties, completing the most complex tasks with the least amount of energy. The total power consumption of the entire human nervous system is about 20 watts, yet it supports 2,200 kilocalories of complex activities per day. This is precisely the evolutionary direction that humanoid robots, constrained by battery power, must take.
Through 75 years of accumulation, NXP can organically integrate all its products and technological capabilities in fields such as microprocessors, coprocessors, network latency products, transmission products, automotive functional safety products, and industrial safety products into the aforementioned hierarchical system architecture.
Robots are the most typical embodiment of this architecture, but the form is not limited to this. Drones, AGVs, and automobiles are essentially morphological variations of Physical AI entities when solving different problems. This also explains why automakers are starting to build robots today and industrial enterprises are crossing over to build cars, because the underlying E/E architectures, energy management architectures, and intelligence classification architectures are interconnected, and core technological capabilities can be reused at scale.
75 Years of Technological DNA × 40 Years of Deep Cultivation in China, Efficiently Empowering Physical AI
NXP's systemic capabilities are not built in a day, but are the intersection of 75 years of technological lineage and 40 years of deep cultivation in China.
Dual DNA Fusion. NXP was formed by the integration of two semiconductor companies with different technological DNA. Philips entered the semiconductor industry in 1953 and launched the first single-chip radio in 1983, subsequently accumulating profound high-performance analog capabilities in RF, security, and radar; Motorola established its semiconductor laboratory in 1949, and during the Apollo 11 moon landing in 1969, it was Motorola's transceiver that transmitted the astronaut's first words. Later, it launched the epoch-making MC68020—the first 32-bit MPU, laying the foundation for the DNA of automotive processors and functional safety. In 2016, the two companies merged, and the "dual DNA" of analog RF × automotive processors was thus integrated.
Notably, in 2025, NXP made three more strategic moves: acquiring Aviva Links (automotive SerDes connectivity), Kinara (edge-side NPU), and TTTech Auto (software-defined vehicle middleware), completing a powerful puzzle piece for the Physical AI era.
40 Years Rooted in China. In 1986, Philips Semiconductors, the predecessor of NXP, established its first office in Shanghai; in 1992, Motorola built its first semiconductor factory in China in Tianjin. Four decades later, NXP has expanded its footprint to 14 cities across the country, with over 6,000 employees in China, serving more than 6,000 customers, and officially established the NXP China Business Unit in 2025.
What does localization truly mean? Mr. Liu Xiaohe gave a clear answer: True localization is absolutely not just about conducting single sales or production in China, but about implementing local product definition and development capabilities. Only by jointly defining and developing products with local customers can we match the needs and development speed of Chinese customers, and together achieve the goal of "In China, for the world."
Currently, NXP has approximately 1,600 embedded engineers in China, having cumulatively developed over 200 products over the past decades; it has set up 6 major R&D centers and 1 world-class assembly and test factory; recently, it has even specially allocated over 100 design personnel with 16nm advanced automotive process development capabilities to the China Business Unit.
The localized product roadmap is also clearly visible: the high-performance automotive processor S32G2 has been mass-produced in 2026; networking products will be mass-produced in 2027; the next-generation edge node MCU is being jointly designed and developed, with mass production expected in 2028; in the electrification field, battery management system products are currently in the design phase, also expected to be mass-produced in 2028. Subsequently, more products will be implemented in the mode of "local production" or "local design + production."
The 75-year technological DNA defines "what can be done," and the 40-year deep cultivation in China defines "for whom to do it and how to do it." The intersection of the two is precisely the confidence with which NXP is betting on the Physical AI era.
40 Years: A New Starting Point
The competition in the Physical AI era is no longer a contest of single-chip parameters, but a battle of system-level architectural capabilities. Whoever can organically integrate perception, computing, connectivity, security, and power into a reliable, low-power, ultra-low latency hierarchical system will win the next decade. NXP is betting on this transformation with its 75 years of technological accumulation and 40 years of roots in China.
And this, perhaps, is just the starting point of the Physical AI marathon.