GM Warns AI Memory Costs Are Repricing Cars as BYD Raises Driver-Assistance Prices

General Motors has warned that rising memory expenses will add multibillion-dollar pressure to its costs, according to new tom hardware coverage. The warning turns an AI infrastructure problem into a direct concern for car buyers.

GM now expects average new-vehicle prices to increase by 0.3% this year. That reverses its earlier projection that prices would remain stable or become slightly more affordable. In China, BYD has reportedly raised prices for separately sold driver-assistance features by 20%.

The common pressure point is memory. AI data centers are consuming growing quantities of advanced chips, while modern vehicles need more DRAM and storage for infotainment, connectivity, and advanced driver-assistance systems. Automakers must compete for supply while meeting stricter qualification requirements than consumer electronics manufacturers.

This is not simply another component shortage. Carmakers have spent years adding software-driven features while presenting them as a route to safer, smarter, and more valuable vehicles. Memory inflation is now exposing the cost of that strategy.

Tom Hardware Connects GM’s Warning to the AI Memory Shortage

The automotive impact has moved from a supply-chain forecast into automaker pricing and financial guidance.

GM Chief Financial Officer Paul Jacobson discussed the cost pressure during the company’s second-quarter earnings call. He identified rising component expenses, particularly DRAM, as a major contributor.

DRAM, or dynamic random-access memory, temporarily holds information needed by active software. A vehicle uses it when rendering dashboard graphics, combining sensor inputs, running navigation, or processing driver-assistance functions.

The company’s expected cost increase is not limited to memory. However, Jacobson’s remarks identify vehicle components and DRAM as important drivers. That distinction matters because it connects GM’s guidance to a wider semiconductor allocation problem.

GM also changed its pricing forecast. The company previously expected vehicle prices to remain flat or decline slightly. It now anticipates a 0.3% increase, according to the memory shortage report.

That percentage looks modest beside the price of a new vehicle. Across a large production base, however, even a small average change can affect affordability, incentives, dealer negotiations, and competitive positioning.

The warning arrives as GM reports stronger operating expectations. Its quarterly results included increased full-year guidance, which gives the company some room to absorb higher input costs.

That room does not eliminate the pressure. It creates a strategic choice between accepting lower margins, reducing incentives, changing vehicle content, or passing part of the increase to customers.

BYD’s response presents the same problem through a different business model. Rather than changing the reported base price of every vehicle, it has reportedly increased the charge for certain driver-assistance capabilities by 20%.

That approach makes the cost more visible. It also places greater pressure on customers to decide whether software-supported driving functions justify the added expense.

Hyundai has taken another route. The South Korean automaker has called for stronger domestic semiconductor supply, according to the same reporting. Its response treats memory availability as an industrial resilience issue instead of only a purchasing problem.

The three approaches reveal the emerging playbook. GM adjusts financial and pricing expectations. BYD changes the cost of a feature package. Hyundai seeks deeper local supply support.

None of those responses expands memory production immediately. They distribute the shortage’s effects among manufacturers, suppliers, shareholders, and buyers.

The critical change is therefore not a single price adjustment. Automakers are acknowledging that memory has become a material constraint on the software-defined vehicle strategy.

Cars Now Consume Memory Like Computing Platforms

Modern vehicles need far more memory because their defining features increasingly depend on continuously running software.

A basic car once needed electronics for engine control, braking, airbags, and limited cabin functions. Current models often combine multiple displays, cameras, radar sensors, navigation, voice interfaces, connectivity services, and over-the-air updates.

Each function produces or consumes data. Infotainment systems load maps, media interfaces, phone integrations, and animated graphics. Driver-assistance systems process streams from cameras and other sensors to identify lanes, vehicles, pedestrians, and road boundaries.

Advanced driver-assistance systems, commonly called ADAS, help with tasks such as emergency braking, lane control, parking, or adaptive cruise control. They do not automatically make a vehicle autonomous.

These workloads require several types of memory. DRAM supports active computation, while NAND flash retains software, maps, vehicle data, and downloaded updates after power is removed.

Micron previously estimated that an average vehicle used 90GB of combined DRAM and NAND in 2023. It projected that average requirement would reach 278GB by 2026, while some premium vehicles could require up to 2TB.

Those figures explain why vehicle production can suffer even when each car uses much less memory than an AI server. Carmakers build at scale, and manufacturers cannot substitute every missing automotive component with a consumer-grade equivalent.

The workload also varies by subsystem. According to figures cited by tom hardware, infotainment can require between 4GB and 16GB. ADAS and safety systems can require between 8GB and 32GB.

Centralized computing systems often begin around 16GB and can reach 64GB or more. These systems consolidate work previously divided among many electronic control units.

Centralization can simplify software development and reduce some wiring complexity. It also concentrates demand around high-capacity, high-performance components. A shortage affecting one central platform can therefore interrupt several vehicle functions at once.

The shift toward local AI adds another layer. An in-car assistant that processes speech, vehicle information, or navigation requests without constantly contacting a cloud service needs stored models and working memory.

Tom hardware reports that some automotive AI systems require at least 256GB of memory. Micron has projected that vehicles designed for Level 4 autonomy could require at least 300GB of RAM.

Level 4 autonomy describes a system capable of driving without human control inside defined operating conditions. It remains different from unrestricted autonomy on every road and in every weather condition.

The exact requirements will vary by design. Automakers can compress models, limit features, divide work between the car and the cloud, or use specialized accelerators with different memory architectures.

The direction is still clear. More capable interfaces and driver-assistance systems increase memory demand, even before fully autonomous vehicles reach mass adoption.

Consumers experience that transition through visible features. A larger display, responsive camera view, detailed navigation system, or voice assistant appears to be software. Underneath, each feature depends on physical computing capacity.

That dependence creates an uncomfortable reversal. Software was supposed to let automakers add value after manufacturing a vehicle. The hardware needed to support that software is now raising the vehicle’s initial cost.

The problem becomes sharper when features are bundled. Buyers who want a particular safety function might also receive a larger infotainment system, more cameras, and a more capable central computer.

Automakers can redesign those bundles, but development cycles are long. A model entering production today was engineered around component choices and performance targets established years earlier.

Removing memory is also not equivalent to deleting an optional mobile application. Engineers must confirm that every required function still meets response, reliability, and safety targets.

This is why the shortage threatens more than premium entertainment systems. Memory supports the operating environment in which essential vehicle software runs.

AI Infrastructure Is Winning the Supply Contest

The main conflict is between rapidly expanding AI infrastructure and automotive programs that cannot change components quickly.

AI servers use high-bandwidth memory, or HBM, to move large quantities of data between processors and memory at high speed. That capability is essential for training and running large models.

Automotive systems do not generally use the same memory configurations as flagship AI accelerators. Yet the markets remain connected through manufacturers, fabrication capacity, packaging resources, capital spending, and product priorities.

Memory companies direct investment toward products with the strongest demand and returns. When AI customers commit to enormous volumes, suppliers have a strong reason to prioritize server-oriented capacity.

That does not mean an HBM chip was diverted directly from a dashboard. The mechanism is broader. AI demand influences which production lines receive investment, which mature products remain attractive, and how quickly manufacturers expand other categories.

Automakers face a weaker position because their volumes are large but their qualification requirements are demanding. They also expect long-term availability across extended vehicle programs.

A consumer-device company can sometimes redesign a product around another part within a relatively short cycle. A vehicle manufacturer must account for harsh temperatures, vibration, electromagnetic interference, functional safety, and a long operating life.

The AEC-Q100 standard defines stress-test qualification procedures for packaged integrated circuits used in automotive applications. Qualification covers failure mechanisms rather than simple feature compatibility.

A replacement memory component must fit electrical, thermal, mechanical, software, and production requirements. Safety-related systems can require additional analysis and validation.

That makes the automotive supply chain less flexible than a product specification might suggest. Two chips can offer similar capacity while remaining unsuitable substitutes inside a validated vehicle platform.

The qualification burden also slows emergency responses. Tom hardware notes that automotive memory can require months or years of validation before vehicle use.

This difference separates the current shortage from ordinary purchasing friction. Automakers cannot solve it only by accepting a higher bid from an unfamiliar supplier.

During the pandemic-era semiconductor shortage, car companies learned how one unavailable component could stop an assembly line. Manufacturers removed selected features, changed production schedules, and prioritized higher-margin models.

That historical comparison is useful, but today’s pressure has a different demand source. Pandemic disruptions combined factory interruptions, logistics problems, and a rebound in electronics purchasing.

The latest constraint is tied to sustained AI infrastructure expansion and growing memory content across product categories. It is therefore harder to treat as a brief transportation bottleneck.

AI customers and automakers also operate on different planning rhythms. Cloud companies can expand capacity through large, concentrated projects. Vehicle programs must coordinate suppliers across many models, factories, regions, and regulatory regimes.

Scale does not automatically give the automotive industry priority. The decisive question is which customers offer suppliers the strongest combination of margins, commitments, and strategic growth.

This places automakers in an awkward position. They need more memory precisely when memory manufacturers see greater opportunities elsewhere.

Long-term supply agreements can reduce exposure, but they introduce other risks. A carmaker that commits at elevated costs could remain locked into unfavorable terms if supply later improves.

Dual sourcing can help, yet qualifying a second supplier takes time. It also requires engineering resources that manufacturers might prefer to spend on new vehicles and software.

Vertical integration offers another possible response. BYD has developed more internal semiconductor capabilities, while other automakers are deepening relationships with chip designers and manufacturers.

Internal chip design does not remove the need for fabrication or memory supply. It can give an automaker more control over architecture, procurement, and workload efficiency.

The contest is ultimately about allocation. AI infrastructure buyers can justify aggressive spending because compute capacity directly supports their services. Automakers must recover higher component costs in an intensely competitive vehicle market.

That imbalance explains why an AI memory shortage can reach a dealership without any generative AI feature appearing in the purchased car.

BYD’s 20% Increase Exposes the Cost Versus Capability Tradeoff

BYD’s reported increase makes the cost of software-driven vehicle capability more visible than GM’s broader pricing forecast.

BYD sells certain advanced driver-assistance functions separately from the vehicle. Raising that charge by 20% connects hardware inflation to a specific customer decision.

This is strategically different from distributing component costs across every model. A separate feature price preserves the advertised base vehicle position while asking interested buyers to fund more of the underlying computing expense.

It can also test demand. If adoption remains stable, BYD gains evidence that customers value the capability despite its higher cost. If adoption falls, the company learns where willingness to pay begins to weaken.

However, the increase does not prove that memory alone accounts for every part of the change. Driver-assistance pricing can reflect hardware, software development, support, data processing, competitive strategy, and expected margins.

The reported timing makes memory pressure a credible factor. Readers should still avoid treating the 20% change as a direct measurement of DRAM inflation.

This is the article’s main uncertainty. GM has identified rising components, especially DRAM, as a significant cost driver. Public reporting does not provide a complete model-by-model breakdown showing how much memory changes each vehicle’s final transaction price.

Automakers also have several ways to absorb or obscure the increase. They can adjust incentives, alter trim availability, renegotiate supplier contracts, or shift production toward models with stronger margins.

A higher list price is only one outcome. Reduced discounts can raise the amount a customer pays without changing the vehicle’s official starting figure.

Feature removal is another possibility. During the earlier chip shortage, some manufacturers shipped vehicles without selected convenience functions or delayed their activation.

That response becomes more complicated when memory supports multiple applications. Removing a single feature might not reduce capacity enough to justify a costly redesign.

Automakers could instead offer simpler computing platforms on entry-level models. Such a move would create a clearer distinction between transportation-focused vehicles and software-heavy premium products.

That choice carries competitive risk. Buyers increasingly compare vehicles through displays, camera quality, mobile integration, and driver-assistance features. A simpler system can look dated even when it performs core driving functions reliably.

Safety adds another complication. Consumers may accept fewer entertainment features, but emergency braking and blind-spot monitoring carry a different value.

Not every ADAS feature requires the same memory or computing resources. Automakers should therefore distinguish foundational safety functions from expensive automation packages when making cost decisions.

BYD’s increase arrives amid intense competition in China’s vehicle market. Manufacturers have used frequent launches, technology packages, and price adjustments to capture attention and defend volume.

Charging more for driver assistance risks weakening a selling point. Absorbing the cost risks weakening profitability. That is the cost-versus-capability conflict in its clearest form.

GM faces the same conflict through a wider North American portfolio. It sells vehicles with different margins, electronics architectures, and customer expectations.

A memory increase that is manageable on a premium vehicle can become more difficult on an entry-level model. Lower-priced vehicles provide less room to absorb component inflation without affecting demand.

The impact also differs between electric and combustion vehicles. Both use digital systems, but many electric vehicles were designed around centralized computing, large displays, and connected software from the start.

That does not mean every electric vehicle contains more memory than every combustion model. Equipment level and platform design matter more than the powertrain label alone.

The larger point is that software-defined vehicles make memory a strategic input. A software-defined vehicle uses centralized computing and updatable software to manage functions that once depended mainly on fixed hardware modules.

The term can sound abstract. The BYD adjustment turns it into a concrete purchase decision: pay more for added computational capability, or accept a more limited feature set.

Consumer resistance remains a genuine constraint. Comments under the tom hardware report include calls for simpler vehicles and fewer connected services.

Those reactions are anecdotal, not representative market data. They still reveal a tension that automakers should not dismiss.

Customers might value emergency intervention systems while rejecting subscriptions, tracking services, or elaborate infotainment. Treating every digital feature as equally desirable can produce the wrong response to higher memory costs.

Automakers that separate essential safety, optional convenience, and advanced automation more clearly could manage the shortage with less customer frustration.

The companies that preserve useful functions with efficient hardware will hold an advantage. Simply adding memory is easier than optimizing software, but shortages make efficiency economically valuable again.

Why Automotive Memory Cannot Be Replaced Overnight

Qualification rules turn a broad memory shortage into a slow-moving production constraint.

A vehicle operates through temperature swings, vibration, electrical noise, moisture exposure, and years of repeated use. Components must remain dependable throughout those conditions.

Automotive memory therefore belongs to a different purchasing category from desktop RAM. Capacity and speed matter, but they are only part of the requirement.

AEC-Q100 qualification subjects integrated circuits to defined stress tests. Suppliers and automakers also apply their own validation, manufacturing, documentation, and quality processes.

Renesas describes AEC-Q100 as a failure-mechanism-based qualification standard. That framing matters because the goal is not simply to confirm that a chip works when installed.

Engineers need evidence about how packaging, fabrication, and materials behave under stress. A failure in an entertainment device is inconvenient. A failure affecting a vehicle control system can carry safety consequences.

Some automotive memory must operate across wide temperature ranges. It also needs stable long-term supply because car platforms and replacement-part obligations extend beyond typical consumer-product cycles.

Infineon describes automotive flash designed for extreme temperatures and long availability. Its documentation also connects external memory with safety-related automotive applications.

This long lifecycle limits rapid substitution. A newer chip might offer better performance, but adopting it can require hardware changes, new software work, and renewed testing.

An automaker dealing with shortages must choose among several imperfect responses. It can pay more for qualified inventory, redesign a module, reduce production, delay a vehicle, or change the features offered.

Paying more is usually the fastest response when inventory exists. It protects production but transfers pressure to margins and vehicle pricing.

Redesigning can improve long-term resilience. It requires engineering time and may introduce fresh supply dependencies.

Reducing production prevents incomplete vehicles from accumulating. It also lowers sales volume and can leave factories underused.

Feature changes can conserve scarce components. They risk disappointing customers and complicating dealer communication.

The best response depends on how long the shortage lasts. A short spike favors purchasing flexibility and temporary margin pressure. A multi-year constraint favors architectural changes and new supplier relationships.

Current signals do not offer a simple duration estimate. Memory manufacturers are expanding capacity, but new fabrication and packaging investments take time.

Demand is also moving. AI models, accelerators, and server architectures continue to change, so suppliers must decide which memory technologies deserve the most capital.

Automotive demand is moving in the same direction. More cameras, richer displays, centralized computers, and AI interfaces increase the required memory per vehicle.

This means added supply can be consumed by added content. The market can expand without immediately restoring the purchasing conditions automakers previously expected.

There is also a mismatch between leading and mature memory products. AI accelerators draw attention to HBM, while many vehicles depend on established DRAM and flash families.

When manufacturers shift investment toward newer products, mature automotive parts can tighten even without direct competition for identical chips.

Legacy supply can become especially vulnerable when production volumes no longer justify continued manufacturing. Automakers then face redesign work for components embedded in long-lived platforms.

This mechanism helps explain why the shortage can persist unevenly. One automaker might have secured adequate supply, while another struggles with a specific qualified part.

It also explains why a universal vehicle-price forecast would be misleading. Exposure depends on contracts, architectures, supplier concentration, inventories, and the feature content of each model.

GM’s disclosed pressure provides one visible benchmark. It does not establish that every automaker will face the same cost increase.

BYD’s feature adjustment provides another signal. It does not establish that every driver-assistance package will rise by the same percentage.

The evidence supports a broader conclusion: automotive memory has become harder to treat as a minor line item. It now influences vehicle design, production planning, and pricing strategy.

Three Signals Will Show Whether Car Prices Keep Rising

The next phase depends on automaker disclosures, feature decisions, and evidence that qualified memory supply is catching demand.

The first signal is upcoming earnings commentary from GM and other major automakers. Investors should watch for changes in component-cost guidance, incentive spending, production schedules, and average vehicle pricing.

If more companies identify DRAM or storage as a material pressure, the shortage has spread beyond isolated procurement problems. If commentary becomes less severe, supply agreements or easing demand might be containing the effect.

The wording will matter. A general reference to inflation provides less information than a specific discussion of memory, infotainment modules, or ADAS hardware.

The second signal is how automakers package digital features. BYD’s reported 20% adjustment creates an early test of whether customers will pay more for advanced assistance.

Watch for similar increases, new optional packages, reduced standard equipment, or wider separation between basic safety features and premium automation.

If adoption remains strong after price changes, manufacturers will have more freedom to pass along hardware costs. If buyers reject the packages, companies will face pressure to absorb costs or simplify systems.

Software efficiency should also become visible here. Automakers that deliver comparable functions with lower memory requirements can protect both margins and affordability.

The third signal is qualified automotive memory availability. Announcements about capacity matter, but lead times and validated product supply matter more.

A new memory investment does not immediately produce components that automakers can install. The output must match the required technology, pass qualification, and reach module suppliers in useful volume.

Shorter lead times would weaken the case for sustained vehicle-price pressure. Continued tight allocation through 2027 would strengthen it and encourage deeper redesigns.

Buyers should watch actual vehicle configurations, not just semiconductor headlines. The first visible effects might appear through smaller discounts, different trim mixes, delayed features, or reduced availability.

Developers working on automotive software should care for a different reason. Memory budgets that once looked generous can become commercial constraints.

Efficient models, careful data handling, and predictable resource use can influence whether a feature survives a cost review. Optimization is no longer only a performance exercise.

Enterprise technology buyers should also recognize the spillover. AI infrastructure purchasing affects component allocation across unrelated products, from vehicles to industrial systems.

That does not make AI investment inherently wasteful. It means the real cost of infrastructure expansion reaches beyond server procurement.

Tom hardware has captured an early consumer-facing consequence of that shift. The decisive question is whether GM and BYD are isolated cases or the first visible examples of a wider repricing cycle.

Over the next quarter, compare automaker guidance, feature-package changes, and qualified memory lead times. Together, those signals will show whether the industry can absorb the shortage or must redesign around it.

Car buyers should ask a practical question before choosing a technology package: which features deliver lasting safety or utility, and which mainly add complexity? That distinction will matter more if memory remains scarce.