The Last Byte: The DRAM Shortage Auto Industry Never Saw Coming

Text "chip shortage" on a chip of an electronic board. Concept on world production shortage of microchips and semiconductors.

getty

Imagine you’re driving down a highway at night, across a desolate stretch—let's say Highway 22E in Alaska, 200 miles from the nearest town, with the temperature outside a shocking -40°F—when the dashboard of your brand-new SUV flickers and dies. No warning, no sign of anything wrong with the car. The issue: faulty DRAM. Confused? Don’t be. You’re probably one of the many people who don’t know what DRAM is. Modern cars run on dozens of computers, but what is more obscure is that those computers rely on DRAM: dynamic random-access memory, a fast, volatile memory that holds active data while a processor works, which, it turns out, is as essential to your vehicle's electronics as fuel is to its engine. DRAM is probably something only the most avid car enthusiasts have heard of, yet a global shortage of it is quietly becoming one of the auto industry's most disruptive—and underreported—crises.

Today’s vehicles are becoming increasingly smart, and the DRAM chip powering the vehicle's computer is at the heart of it all. DRAM (Dynamic Random-Access Memory) is a type of memory chip that temporarily stores data as it is processed—like a place to keep and use information in real time, rather than a place to store it long-term. Modern smart cars can have up to 150 microprocessors that control everything, right from infotainment and advanced driver assistance systems to ECUs and vehicle telematics. And as they grow smarter—think software-defined vehicles, high-end EVs and autonomous vehicles—their reliance on DRAM increases significantly.

So, what’s brought DRAM into the limelight? To put it simply, a few things—cars becoming smarter, the increasing rise of AI and the phasing out of older generation memory. While the first is a natural progression and would have probably been accounted for, it’s the second and third that together are becoming the main causes of the crisis. The world has moved to something I’m beginning to call ‘AI-in-Everything’—a place or a time where everyone is using AI for everything. And I do mean everything—right from creating sophisticated codes and running an army of AI agents to asking mundane questions and editing your favorite image. The result? An unprecedented demand for AI and the data centers to process the compute. And while this may seem unrelated to the automotive industry, the reality is far from so. Major chip manufacturers are increasingly supplying wafers to more lucrative data centers rather than the automotive industry, creating an alarming shortage of DRAM. This is driven by higher prices hyper scalers are willing to pay, higher margins for the manufacturers and more importantly higher volumes.

In fact, the leading suppliers of DRAM—Samsung, SK Hynix, Micron and others — have started reallocating capacity toward AI data centers, where margins are 3-5x higher than in the automotive industry, creating a supply chain nightmare that seems to be unresolvable (Samsung, SK Hynix and Micron control about 95% of the global DRAM). Even more alarming is the fact that by the end of this year, data centers are projected to consume almost 70% of all the memory produced globally. And of the remaining 30%, the automotive industry accounts for barely 5%. For a chip manufacturer, choosing between Nvidia and a Ford (or any OEM) has become a no-brainer.

To understand why this is alarming and not just another supply-demand imbalance that will eventually recover, you need to understand the economics of modern memory fabrication. High Bandwidth Memory, the DRAM variant used in AI accelerators and data center GPUs, consumes three times the wafer capacity per gigabyte compared to standard LPDDR4. Its manufacturing yield hovers around 50 percent—meaning fabs discard half their output. And yet HBM is dramatically more profitable than automotive DRAM. Automotive DRAM, at the volumes and price points the industry has historically demanded, cannot compete with HBM.

Chip shortages have been seen before—almost everyone will remember the 2021 COVID-19 shortage, which caused price fluctuations in cars, smartphones and appliances. And though that probably affected the automotive industry the worst, the biggest difference between the 2021 shortage and the current crisis is that previous shortages were not caused by reallocation. Previous shortages could be termed as supply shocks or imbalances between demand and production capacity that eventually corrected. The forthcoming crisis cannot qualify as a supply shock. It is a technology discontinuation driven by a structural, rather than cyclical, shift in the economic incentives for memory fabrication, making a reversal less likely.

This is further compounded by the fact that the current DDR4/LPDDR4 RAM—used in over 80% of vehicles—is being phased out. This is not a distant forecast. Prices in the market have already risen by 70-100%. And by 2028, these parts will be unavailable at any price, from any supplier, anywhere in the world. The discontinuation of DDR4 and LPDDR4 is driven by the inescapable economics of where memory fabs are allocating their wafer capacity. And those economics have permanently shifted. Making a bad situation even worse, 8 major automakers—Ford, General Motors, Stellantis, Volkswagen Group, Hyundai, Nissan, Rivian and Volvo—all depend on Qualcomm’s SA8295P chip, which supports only LPDDR4 memory. These automakers are inexorably walking toward the edge of the same cliff, at similar speeds and with similar migration timelines ahead. The fact that the SA8295P is an excellent chip is irrelevant. Its memory interface is the constraint, and it's immovable without a platform change.

So, the next question that arises: why can’t the industry shift to LPDDR5? Unfortunately, DRAM generations are not pin-compatible. Moving from LPDDR4 to LPDDR5 requires a new system-on-chip, a new printed circuit board design and a full automotive safety recertification under ISO 26262. End-to-end, this process takes a minimum of 12 months and more typically 24. There is a silver lining, if it can be called that. Not all automakers are on that cliff, walking toward their doom. Three manufacturers have begun migrating to LPDDR5 platforms: Tesla, which has historically maintained the industry's shortest silicon-to-vehicle cycle; Volvo, through the EX90 program; and General Motors, through its Ultra Cruise platform. Everyone else remains locked to legacy memory. The degree of exposure varies, and the most vulnerable positions are held by Volkswagen, Toyota and Honda.

While Western OEMs face a supply discontinuation with no easy exit, Chinese manufacturers have assembled a vertically integrated alternative that makes them largely immune to the 2028 cliff. CXMT, China's domestic DRAM manufacturer, offers supply at roughly one-third the price Western OEMs pay in the open market. Chinese OEMs are not subject to the same allocation economics that are squeezing automotive DRAM globally—they have a domestically captive supply. More significantly, their next-generation ADAS chips—XPeng's Turing platform, Nio's Shenji chip and Huawei's Ascend architecture—are already native LPDDR5.

What’s the solution, then? Automakers seem to have quite a few. Buffer inventory, one of these solutions, provides only time. It cannot provide supply after 2028 because there is no supply to buy at any price thereafter. NCNR contracts can lock pricing and allocations, but cannot create memory that fabs are no longer producing. The only strategy with a durable outcome is accelerated LPDDR5 migration: new SoC selection, PCB redesign, safety recertification and full production validation, completed before the legacy supply collapses. Every other approach just slows down the inevitable.

The companies that understand this distinction—deadline, not cycle—are making moves right now. They are locking in supply. They are building relationships that go beyond transactional. And they are treating memory not as a commodity to be squeezed on price, but as a strategic asset that determines whether their production lines run or sit idle.

Sure, the future of driving is intelligent, connected and autonomous—but only if we can solve the memory crisis before the road runs out. It is already narrowing. And it does not care whether you're ready.

Written together with Benny Daniel, Senior Vice President, Markets and Markets.