AI chip demand squeezes automotive manufacturing: a quietly unfolding supply chain crisis

Event: Auto Industry Joins Forces with Multiple Sectors to Sound Alarm

In June 2026, a cross-industry alliance composed of the Alliance for Automotive Innovation, the National Retail Federation, the Medical Device Manufacturers Association, and the Internet & Television Association jointly sent a letter to the U.S. Department of the Treasury and the Department of Commerce, warning that the demand for memory chips from AI data centers is seriously distorting the market. Automakers, retailers, medical device manufacturers, and telecom operators are all facing pressure from chip shortages and rising costs, which could lead to declining auto production, rising prices, and a repeat of the pandemic-era supply chain crisis.

This is not alarmist talk. According to Reuters, the alliance pointed out that global memory chip production capacity is being rapidly consumed by AI data centers, especially high-bandwidth memory (HBM) and DRAM. Major memory manufacturers such as Samsung Electronics, SK Hynix, and Micron are increasingly allocating capacity toward high-margin AI products, leaving fewer mature-process memory chips for sectors like automotive and consumer electronics. Analysts predict that in 2026, the global smartphone market will experience its largest annual decline ever, partly due to memory chip shortages.

Background: The "Memory Black Hole" Under the AI Arms Race

Since ChatGPT ignited generative AI in 2022, global tech giants have been frenetically building AI data centers. Training and inferencing large models require massive amounts of HBM and high-speed DRAM, which consume more power and offer higher bandwidth than ordinary memory, but also have higher unit costs. Taking the NVIDIA H100/H200 GPU as an example, each GPU requires 6–8 HBM3 chips, and HBM manufacturing processes are complex with limited capacity. TrendForce data shows that in 2025, HBM will account for over 20% of total DRAM capacity, and this share is expected to approach 30% by 2026.

Meanwhile, demand for automotive chips is growing structurally. Smart cockpits, advanced driver-assistance systems (ADAS), and powertrain electrification all require more chips. An average traditional internal combustion engine vehicle uses 500–600 chips, while a high-end electric vehicle requires over 2,000. Among these, control units (ECUs), sensors, power semiconductors, and memory chips are indispensable. However, automotive chips have high reliability requirements, long certification cycles, and mostly use mature process nodes (28nm and above), with profit margins far lower than AI chips.

In-depth Analysis

Technology Impact: The Dilemma of Technology Roadmaps

The technology roadmap for memory chips is diverging. AI requires 3D-stacked HBM and ultra-high-speed DDR5/LPDDR5X, while the automotive industry mainly needs LPDDR4, DDR3/LPDDR3, and even older SDRAM (for basic control). The automotive sector has long relied on mature, stable processes. Switching to new-generation memory not only requires redesign but also mandates passing automotive-grade certifications (AEC-Q100, ISO 26262). As AI chips seize advanced process capacity, automakers are forced to compete with consumer electronics for limited legacy process capacity.From a wafer fabrication perspective, memory chips (especially DRAM) are standardized products, making it difficult for foundries (such as SK Hynix’s Wuxi plant and Samsung’s Pyeongtaek plant) to adjust their product mix without increasing capital expenditure. Once the profit margin of HBM exceeds that of ordinary DRAM by several times, manufacturers naturally shift capacity toward HBM. This “virtuous cycle” benefits AI, but it spells disaster for traditional sectors such as automotive, industrial, and medical.

Supply Chain Impact: Who Benefits? Who Loses?

The upstream part of the supply chain (equipment and materials) is unaffected, and actually benefits from memory manufacturers expanding HBM capacity. ASML’s EUV lithography machines, Applied Materials’ thin-film deposition equipment, and Lam Research’s etching tools are all being purchased in large volumes. However, the midstream (chip design) and downstream (end-product manufacturing) segments of the chain show a sharp divide:

• Beneficiaries: AI chip design companies (e.g., NVIDIA, AMD, Google) and cloud service providers (AWS, Microsoft, Meta). They are able to secure sufficient HBM and DRAM supply, keeping AI server shipments on track. Memory manufacturers (Samsung, SK Hynix, Micron) are posting record profits.
• Losers: Automotive OEMs (Toyota, GM, Ford, Tesla), Tier 1 suppliers (Bosch, Continental, Aptiv), as well as consumer electronics (Apple, Xiaomi), medical equipment (Medtronic, Siemens Healthineers), and telecom equipment (Ericsson, Nokia) manufacturers. These companies either have to pay higher prices or face shortages.

The coalition noted in its letter that memory chip prices have risen for 12 consecutive months, with some categories seeing increases of over 80%. These costs are ultimately passed on to consumers. The automotive industry is especially vulnerable because a single vehicle often uses multiple types of memory chips, and a shortage of just one can delay the delivery of the entire vehicle.

Competitive Landscape: The AI vs. Automotive Resource Battle

In the past, the main competitor for automotive chips was consumer electronics. But AI has changed the rules. The AI chip market (GPUs, ASICs, FPGAs) exceeded $200 billion in 2025, while the automotive chip market is around $60 billion. Capital and capacity naturally flow toward high-margin areas.

In the memory space, HBM capacity from Samsung and SK Hynix has been locked in for years by NVIDIA and AMD in advance; Micron is also fully shifting toward DDR5 and HBM. Automotive customers are left to use only the remaining capacity. Automotive memory chip suppliers such as Macronix, Winbond, and GigaDevice, though focused on niche markets, also face tight upstream foundry capacity.

Automotive chip design companies (such as Renesas, NXP, Infineon) themselves are already facing shortages of MCUs and SoCs, but now memory chip bottlenecks have been added to the mix. They may have to seek long-term agreements (LTAs) with memory manufacturers, though the price discounts are far less favorable than before.

Regional Implications: Reshaping the Global Semiconductor Landscape- United States: The alliance is directly pressuring the federal government to use CHIPS Act funding to encourage memory manufacturers to expand production of legacy chips. However, the CHIPS Act has already allocated tens of billions of dollars for advanced process nodes, and whether to shift direction still requires congressional debate. - Taiwan, China: Although TSMC does not produce memory, there is a severe shortage of advanced packaging capacity (CoWoS) for AI chips, indirectly affecting the supporting capabilities for automotive SoCs. - South Korea: Samsung and SK Hynix have become "winners," but they also face public criticism for sacrificing the domestic automotive industry (Hyundai-Kia) to meet global AI demand. The South Korean government may require memory manufacturers to reserve some capacity for key industries such as automotive. - Japan: Rapidus is advancing its 2nm process, but the memory shortage actually benefits Japan's strengths in automotive analog chips and power semiconductors (e.g., Renesas, Murata). - Europe: The European automotive industry is deeply concerned. Stellantis, Volkswagen, and others are lobbying the EU to include memory chips in the "strategic reserves" list of the European Chips Act.

Investment Perspective: Bifurcation in Capital Markets

The investment logic has become clear: AI chip-related companies (NVIDIA, Samsung, SK Hynix, Applied Materials, etc.) continue to see expanding P/E ratios, while automotive chip companies (Renesas, NXP, Texas Instruments) face valuation pressure. Investors worry about declining auto sales and chip costs eroding profits. The memory price index (DXI) keeps rising, but automotive stocks fall due to supply risk.

In the long run, if the automotive industry cannot solve the memory supply issue, it may accelerate self-developed chips (e.g., Tesla's Dojo D1, Mobileye EyeQ) or shift to designs more resilient to shortages (e.g., highly integrated SoCs to reduce material types).

Long-Term Outlook: 3-5 Years of Pain and Structural Changes

By 2029, as dedicated AI chip production lines mature and new global fabs come online (especially Samsung, Micron, and Kioxia's plans), memory capacity may ease. However, structural contradictions will not disappear: the growth rate of compute and storage consumed by AI far outpaces that of traditional industries.

The automotive industry may be forced to accept the reality of "persistent chip supply tension," raise inventory levels, and even build its own memory chip reserves. At the same time, automotive design will accelerate the transition to "software-defined vehicles," reducing reliance on physical chips through OTA updates.

Another possibility is geopolitical intervention: The US and EU may use export controls or investment guidance to force some memory capacity to serve "critical infrastructure" sectors such as automotive. But the efficiency logic of the market economy and the security logic will remain in long-term tension.

Industry Chain Analysis: Supply-Demand Tear Between AI Chips and Automotive Chips### Upstream: Equipment and Materials Unaffected. Orders for ASML, Applied Materials, and Tokyo Electron continue to grow, but the automotive industry has no bargaining power.

Midstream: Memory Manufacturing (Samsung, SK Hynix, Micron) Production capacity is shifting towards HBM. By 2026, HBM will account for more than one-third of the three companies' DRAM capacity. The share of DDR4/LPDDR4 capacity available for automotive use is declining, leading to price increases of 50%-100%.

Downstream: Automotive Tier 1 and OEMs - Procurement costs surge: A high-end electric vehicle uses approximately 10-20 memory chips (including LPDDR4, DDR3, eMMC), with total cost rising from $30 to over $60. - Delivery delays: Due to memory shortages, some ECUs cannot be completed in production, causing entire vehicle assembly lines to halt. - Terminal price increases: General Motors and Ford have hinted at price hikes in the second half of 2026.

Indirect Impact: Advanced Packaging (TSMC CoWoS) Automotive SoCs often adopt heterogeneous integration and require advanced packaging. However, CoWoS capacity is monopolized by AI customers like NVIDIA and AMD, leading to extended packaging times for automotive chips. This further worsens the availability of automotive chips.