The next automotive semiconductor cycle: stability, AI, and a change in architecture

In 2025, a typical passenger car contains roughly 900 semiconductor devices, of which 93% are built on legacy nodes. The share of leading-edge nodes continues to grow as computing and memory needs for ADAS, infotainment, and centralized architectures increase.

In today’s automotive snapshot, Pierrick Boulay, Senior Analyst, Automotive Semiconductors at Yole Group, explores the next automotive semiconductor cycle, where supply stabilization meets AI-driven architectural shifts. Drawing on Yole Group’s extensive automotive collections, teardown tracks, and market & technology reports, this analysis connects device mix, node evolution, memory dynamics, and OEM strategies to the broader transformation toward software-defined vehicles. A must-read to understand how legacy nodes, advanced compute, and AI workloads will shape the next phase of automotive electronics.

The automotive sector suffered heavily during the 2020–2022 semiconductor shortages because more than two-thirds of its chips run on mature processes (180–40 nm). Yole Group’s data shows that a single vehicle integrates more than 700 power and analog components, typically manufactured on 350–40 nm nodes. By 2025, substantial foundry investments, particularly in 90 nm, 55 nm, and 40 nm lines in the United States, Europe, Japan, and Southeast Asia, greatly reduced the bottlenecks that once hindered production. Automotive power semiconductors, MEMS, magnetic sensors, and display drivers now benefit from improved availability and predictable lead times.

While advanced-node devices still account for only about 7% of components per car, their strategic weight is dramatically higher. Yole Group’s breakdown shows that ADAS application processors (APUs), produced on nodes as advanced as 5 nm, are now among the most expensive silicon elements in the vehicle, with prices that can exceed $100 for a single device. The adoption of advanced ADAS platforms, typically for level 2+ and level 2++ autonomy, across multiple OEMs has enabled silicon suppliers such as Nvidia, Qualcomm, and Mobileye to increase shipments, driving performance gains and cost efficiencies.

Along with computing needs, memory needs follow the same direction. With more than 90 MCU devices for powertrain, connectivity, and dashboard computing, the electronics architecture is steadily shifting toward domain and zone controllers, requiring more memory for data storage and real-time processing. This consolidation is a necessary step toward software-defined vehicles (SDVs), enabling OTA updates, more consistent cybersecurity, and future scalability.

Recent political scrutiny and operational restrictions involving Nexperia underscore how geopolitics can destabilize even the mature-node layers of the automotive semiconductor ecosystem. The implications include:

Given that 93% of automotive chips rely on mature nodes, any disruption, regulatory, trade-related, or investment-related, can propagate rapidly across global production.

Even with a stable supply, the automotive industry remains exposed to mature-node capacity constraints because it relies on technologies that the broader semiconductor industry considers low-margin. Foundries continue to invest heavily in 3 nm, 2 nm, and eventually sub-2 nm nodes for consumer electronics and AI compute, while 180 nm to 40 nm capacity grows only incrementally. Automotive OEMs must therefore secure long-term agreements and co-investments to guarantee access to these mature technologies.

Until recently, global DRAM and NAND suppliers signaled upward pricing trends due to:

For automotive OEMs, this poses two problems:

Automotive volumes of tens of millions of units are dwarfed by those of the cloud and PC markets. If AI servers consume a disproportionate share of the DRAM/NAND supply, automotive buyers will face higher input costs and longer lead times.

Some OEMs have followed the trend, initiated by Tesla before 2020, to develop their own SoCs for ADAS with the goal of reaching higher levels of automation and having better control over system integration. The latest OEMs following this trend are Volkswagen, with its “In China, for China” strategy to deliver the L3 system, and Rivian, with its first-generation Rivian Autonomy Processor based on 5nm process nodes, targeting L4 autonomous driving systems. These OEMs are the latest examples to date, but others are expected to follow to iterate faster, optimize AI workloads, and reduce dependency on merchant silicon roadmaps.

With the development of SoCs requiring ever-increasing computing power, chiplets could be an interesting solution to avoid large and costly devices. Indeed, by mixing small devices made with different process nodes, the chiplet approach could be the right solution to meet an ever-increasing demand for computing power while limiting the cost increase of such processors. Chiplets are already in use in other industries such as consumer electronics and high-performance computing, though they are not expected in automotive in 2026 and should enter the market around 2027-2028.

Stay tuned on Yole Group website and LinkedIn for future insights!

Pierrick Boulay is Principal Analyst, Automotive Semiconductors at Yole Group.

He works in the fields of solid-state lighting and lighting systems, carrying out technical, economic, and marketing analyses. In addition, he leads the automotive activities within the company.

Pierrick has authored several reports and custom analyses on topics such as automotive lighting, LiDAR, sensing for ADAS vehicles, and VCSELs.

Prior to Yole Group, Pierrick has worked in several companies where he developed his knowledge of lighting and automotive. In the past, he has primarily worked in R&D departments on LED lighting applications.

Pierrick holds a master of science in electronics at ESEO (Angers, France).


Source: www.yolegroup.com