Gasgoo Munich- Prices for automotive-grade memory chips have surged roughly 180% over the past three months, according to CCTV Finance.
Explosive demand for AI servers has soaked up nearly all the world's incremental memory chip production. The three major manufacturers have shifted 70% to 80% of their advanced process capacity to HBM and high-end server DDR5 memory, leaving automotive and consumer electronics sectors facing a severe supply squeeze. With the auto industry accounting for only about 3% of the global DRAM market, it sits at the bottom of the allocation queue and is feeling the impact acutely.
Image source: 699pic
Citing TrendForce and multiple industry reports, the per-vehicle memory cost for mid-to-high-end models has jumped from $40–$90 to $90–$220 due to capacity being siphoned off by AI. High-end intelligent models have even broken the $500 mark.
Qualcomm Executive Vice President Nakul Duggal was blunt at a recent media briefing: "The industry is indeed facing severe challenges with memory pricing over the next 12 to 18 months." He noted that massive amounts of supply are being displaced by data center demand, and supply conditions won't improve until that demand growth slows.
AI Feasts, Auto Starves?
To understand this price surge, one must first look at the logic behind memory chip capacity allocation.
Samsung, SK hynix, and Micron collectively control more than 90% of the global DRAM market. Their logic is consistent: prioritize high-margin, high-price product lines. AI customers are willing to sign multi-year agreements, pay premiums, and lock in capacity early. Consequently, the three manufacturers have tilted most of their advanced process capacity toward high-profit products like HBM and high-end server DDR5 memory.
Crucially, as a key component for AI training and inference, HBM production has a significant magnifying effect on capacity usage compared to traditional DRAM. According to SemiEngineering, HBM consumes about three times as many wafers per bit as traditional DRAM. As next-generation products like HBM3E and HBM4 move into mass production, this ratio is set to rise further.
This means the continued tilt of wafer capacity toward HBM will inevitably squeeze the supply of standard DRAM for end-uses like automobiles and smartphones, creating a clear zero-sum dynamic.
Currently, the dominant DRAM specification for the automotive industry remains LP4/4X (LPDDR4/LPDDR4X), the fourth-generation low-power memory. The cockpits and autonomous driving systems of the vast majority of intelligent models on sale are designed and mass-produced based on this standard. However, LP4/4X is not on the priority list for high-margin products protected by manufacturers, leaving it facing severe structural displacement.
Even as new factories come online over the next two years, priority will still go to replenishing HBM and high-performance server memory lines. TrendForce data indicates that significant new capacity won't arrive until late 2027 or 2028.
Even as supply gets squeezed, demand for automotive memory is growing sharply. With large AI models moving into vehicles, both in-cabin AI interaction and complex autonomous driving calculations require rigid, high-capacity, high-bandwidth memory. Traditional cars needed only a few gigabytes of memory, but current intelligent vehicles have generally jumped to 16GB–32GB, with flagship models reaching 64GB–128GB. Industry consensus predicts that per-vehicle memory demand will exceed 300GB once L4 autonomous driving becomes widespread.
Image source: Huaban
Caught in this pincer movement between supply and demand, the auto industry faces exactly what HARMAN recently highlighted: the combination of LP4/4X supply uncertainty and rapidly rising prices is exacerbating vehicle costs and delivery risks.
Predictions from multiple organizations—including Edgewater Resource Estimates, Samsung, and TrendForce—suggest LP4/4X will shift from "expensive but available" in 2026 to "expensive and scarce" in 2027, with potential discontinuation looming around 2028. HARMAN was blunt: the industry must plan ahead and prepare fully.
Defend the Base, Find New Growth
With upstream capacity locked down by AI and the dominant LP4/4X spec facing retirement, this double squeeze on the supply side is forcing automakers to actively seek a way out.
For automakers that have already locked in long-term agreements, the rigidity of those contracts provides the most direct short-term buffer. Prices and volumes were fixed at signing, meaning original terms hold even if the market fluctuates wildly.
An insider at a leading domestic automaker told Gasgoo that their company locked in prices and volumes at the time of purchase. "Any price hike would constitute a supplier breach—the price is locked, and we might be buying hundreds of thousands of units at once. A temporary price adjustment is a breach, and the damages would far exceed those in the consumer electronics sector. If supply chain issues cause losses on our production lines, the penalties are even higher."
He added that companies also balance costs through other components. "Generally, the automotive industry doesn't adjust prices temporarily due to short-term market changes. Even with significant volatility, we try to maintain the original price."
But this protection isn't indefinite—contracts eventually expire for renewal. At that point, automakers will have to renegotiate at prevailing market rates. The truly tricky part is that even if automakers accept higher prices, they may face the predicament of having nothing to buy.
Automakers may lock in prices, but the real risk lies in supply allocation.
According to investor documents disclosed by SanDisk, it has signed 3- to 5-year Long Term Agreements (LTAs) with multiple customers, locking in more than a third of its bit production for fiscal 2027.
The commercial logic of memory manufacturers is highly consistent: during capacity shortage cycles, they prioritize binding supply relationships with customers willing to commit to long-term purchase volumes.
This means that even as new capacity comes online in 2027, a significant portion of that incremental supply has already been pre-allocated to the contract pools of AI customers. Once the majority of manufacturers' capacity is locked down by AI long-term agreements, the supply quota left for other sectors will be compressed to even lower levels.
From this perspective, innovation on the technology side becomes particularly critical.
An intelligent driving technology expert at a prominent domestic supplier was blunt in an interview with Gasgoo: "We can't control the rise in memory chip prices. What we can do is find the optimal technical solution within given constraints—for example, optimizing the model structure so it runs on 32GB of memory, eliminating the need for 64GB."
Li Peizhi, Vice President of R&D for HARMAN China's Smart Cockpit division, recently stated that for automakers, high storage costs and supply volatility have become a dual risk. While OEMs can currently reduce dependence on LP4/4X by enhancing supply chain resilience, looking to the long term, restructuring the vehicle E/E (electrical/electronic) architecture is imperative.
The fusion of cockpit and driving functions is a key direction. Under traditional distributed architectures, the cockpit and ADAS each occupy independent DRAM (main memory), easily leading to data duplication and redundant processing. Cabin-driving fusion, however, uses a single SoC and shared memory to enable data reuse. This improves storage efficiency at the source and reduces DRAM usage, thereby lowering system costs.
Nakul Duggal made a similar point in an interview: "Integrating cockpit and intelligent driving functions into a single system-on-chip significantly reduces the customer's total cost of ownership. From a bill-of-materials perspective, memory also benefits because there is no longer a need to equip separate dedicated memory for two different chips."
HARMAN emphasized that automakers should upgrade their E/E architectures as soon as possible to ensure long-term stability. Regarding storage, the future will focus on LP5 and single-chip cabin-driving fusion, supplemented by local substitution and stockpiling.
Image source: Changxin Memory Technologies
This leads to the path of domestic substitution. Breakthroughs by Changxin Memory Technologies in memory, and progress by GigaDevice and Puya Semiconductor in flash memory and other niche markets, are providing alternative options. Objectively speaking, however, current domestic automotive-grade production capacity still struggles to fully fill the gap for high-end products, leaving the market heavily reliant on imports in the short term.
Overall, buying time with long-term contracts, reducing consumption through architecture, and building backups via domestic substitution form a strategic triad—a complete matrix for automakers to navigate the cycle of rising memory prices.
Bracing for Another Year of Pain
Synthesizing information from all sides, a clear judgment has emerged: the tangible impact of rising memory prices on the automotive industry will last for at least another 12 to 18 months.
Nakul Duggal acknowledged that rising memory prices are a major challenge for the entire industry. "We are in a period of memory supply premiums where massive amounts of capacity are occupied by data centers. But I believe that within the next 12 to 18 months, overall supply conditions will improve and pricing is expected to return to more reasonable levels, though the present is certainly a complex period."
On the supply side, new capacity won't make a substantial contribution to the market until 2027 at the earliest. Citing TrendForce, a research report from China Merchants Securities predicts there is still room for high-capacity flash memory prices to rise by 60% to 65% in the second half of 2026, with SLC NAND expected to increase by 70% to 75%.
On the demand side, the evolution of AI from generative models to intelligent agents requires the coordinated operation of various memory types, further driving up overall consumption.
What does this timeline mean for automakers?
Speaking at a media briefing following the GX launch, XPENG CEO He Xiaopeng was candid about the difficulties of pricing new vehicles. "Making cars is truly painful; any price increase affects us." He revealed that the money XPENG saved through technological innovation was "largely returned to our partners in the memory and lithium carbonate businesses."
These words articulate a predicament many automakers are reluctant to voice openly. Demand for automotive memory is inherently growing, and the requirements for memory bandwidth and capacity in high-end autonomous driving are continuously climbing. Even if unit prices remained flat, per-vehicle memory costs would still be rising.
At the same time, raw materials for power batteries are also climbing.
An auto executive at the 2026 Chongqing Auto Forum did the math: memory chips rose from 20 RMB to 100 RMB; lithium carbonate jumped from 80,000 RMB per ton a year ago to 180,000 RMB; and lithium iron phosphate climbed from just over 10,000 RMB per pack to more than 25,000 RMB. On average, that adds 15,000 to 20,000 RMB to the cost of a single vehicle.
That calculation covers only raw materials, excluding the hidden costs of supply chain management, inventory turnover, and capacity allocation. Deutsche Bank estimates that rising memory costs will push the selling price of ordinary vehicles up by $150 to $300, with high-end autonomous vehicles seeing increases of $400 to $600.
The squeeze on automakers is evident. Pricing strategies have been disrupted by rising costs, profit margins are being compressed layer by layer, yet price sensitivity in the terminal market hasn't eased. Finding a balance between "protecting profits" and "protecting market share" has become a real dilemma facing every automaker.
Moving forward, automakers will indeed have to tough it out. But "toughing it out" doesn't mean waiting passively. Long-term contracts have bought time, architectural innovation has reduced consumption, and domestic substitution has provided backups. Companies that complete technical optimization and supply chain restructuring during this period will gain greater initiative in the next cycle.
Perhaps the biggest variable brought by this round of price hikes lies right here: it is forcing the entire industry to re-examine its dependence on memory resources and seek more efficient usage at the architectural level. The endgame of rising memory prices may not be a return to previous price levels, but rather a forced evolution for the industry.
