By ATN Editorial Team
Guest Insights piece by Russell Ruben, Technical Segment Marketing Director, Sandisk: why automotive flash storage has become the quiet foundation of AI-powered vehicles.
We’ve seen successive waves of computing reshape the world: from mainframes to PCs, from the Internet to mobile. Each leap multiplied the number of users, use cases, and the volume of data being created. Now, as we enter the AI-powered era, the growth curve steepens again. And nowhere is this more visible than in today’s L2+ and higher autonomous vehicles, a category of cars with enhanced automation features, such as advanced driver-assistance systems (ADAS), that help improve vehicle safety and convenience.
These modern L2+ and higher vehicles are effectively rolling supercomputers. Packed with high-resolution cameras, LiDAR, radar, sonar, GPS, and other sensors, they capture a torrent of information every second. A typical AV sensor suite can produce raw data ranging from hundreds of megabytes to several gigabytes per second. Over a day of driving, that can amount to two terabytes from a single vehicle.
Every byte of that data must be processed and accessed in real time. Relying on the cloud alone isn’t just impractical, it’s uncertain. To operate reliably, vehicles need next-generation, edge-ready storage embedded directly in the vehicle.
Enter automotive-grade flash.
Data centres on wheels
Today’s vehicles bear little resemblance to cars built even a decade ago. No longer just mechanical machines, they are dense clusters of software, sensors, and high-speed electronics. The conveniences drivers now expect (blind-spot alerts, automatic braking, adaptive cruise control, lane keeping assist, and traffic jam assist) depend on constant data capture and millisecond-level decision-making across networks of ultrasonic sensors, cameras, and onboard AI. These features work only because the car is continuously collecting, processing, and interpreting its environment.
Across the globe, L2+ cars, which use telecommunications to link with in-car devices, other vehicles, and the cloud, are rapidly becoming the norm. By 2030, it is estimated that 95 percent of new vehicles sold globally will have embedded connectivity.
As vehicles advance toward full autonomy, their data footprint grows exponentially. Self-driving systems require complete, 360-degree environmental understanding in varying conditions: rain, glare, snow, darkness, complex urban environments. By 2035, robotaxi fleets are projected to operate at scale in 40 to 80 major cities around the world, and autonomous commercial trucks could represent nearly 30 percent of new U.S. truck sales.
AI onboard
At the centre of this data are AI models running directly inside the vehicle. Think of them as the car’s brain, constantly interpreting the world, deciding how to respond, and then taking action. In advanced driver assistance systems, these models process streams of information continuously from cameras, radar, and other sensors to understand what’s happening around the car: where the lanes are, how fast nearby vehicles are moving, and whether a pedestrian or cyclist is in its path. This real-time intelligence is what helps power features like automatic emergency braking, lane-keeping assist, blind-spot monitoring, and adaptive cruise control.
In fully autonomous vehicles, the data challenge grows even more complex. AI models must not only perceive the road, but continuously assess the health of its own systems. They monitor whether sensors are blocked, whether LiDAR signals are degraded by fog or dust, or whether calibrations are drifting over time. When something goes wrong, the vehicle has to judge whether it’s safe to continue, or whether it should slow down and pull over.
In modern cars, data moves not only within onboard AI, but back and forth to the cloud in a learning loop. When a vehicle encounters situations it doesn’t yet understand, those events are captured and sent to the cloud, where they are used to retrain and improve the AI models. Updated models are then delivered back to the vehicle through over-the-air software updates.
The automotive flash storage backbone
Onboard computing and data processing requires high-performance in-vehicle storage that can handle not only the sheer volume of data, but also its complexity. The data streams generated by L2+ vehicles are inherently multimodal, blending visual information, depth and spatial data, motion telemetry, environmental signals, and outputs from onboard AI models. Each of these data types has different performance, latency, and retrieval characteristics, placing unique demands on the vehicle’s storage system.
Flash memory sits at the centre of this solution. Automotive-grade flash is engineered for the unique demands of the vehicle environment. It delivers the high read and write speeds required for real-time perception and decision-making. It maintains reliability across harsh temperature extremes, constant vibration, humidity, and shock. And it provides the endurance needed to withstand the heavy write cycles created by data logging and AI workloads.
The latest generation of automotive flash boosts performance while increasing capacity and durability. This level of advancement is not incremental. It is a key enabler for autonomous vehicles to become a scalable technology. Faster interfaces and higher densities give onboard computers the bandwidth required to interpret sensor streams, maintain HD maps, and store the data needed for post-incident analysis and regulatory compliance.
Why relying solely on the cloud isn’t viable
Onboard automotive flash storage is essential for one simple reason: driving decisions must happen instantly. Braking, steering, or avoiding an obstacle can’t wait. Sending sensor data to the cloud and waiting for a response would be far too slow. Safety-critical decisions also cannot depend on a network connection. Vehicles routinely drive through tunnels, rural regions, crowded cities, and areas with weak or intermittent connectivity.
Although the cloud will continue to play a major role in training AI models, updating maps, system OS and application software, and learning from fleet-wide data, a vehicle’s computing and data processing need to operate completely on its own. That’s why embedded flash is so critical. It acts as both the vehicle’s real-time working storage and its long-term recorder, capturing and storing events on the road.
The road ahead
As L2+ and higher cars take hold in the market and fully autonomous vehicles and AI-rich driving experiences transition from pilot programmes into mainstream fleets (beginning with robotaxis and commercial trucks, then eventually reaching consumer vehicles) the demand for high-performance automotive storage will surge. Flash memory has evolved from a supporting component into a foundational technology for the automotive supercomputers we used to call cars and trucks.
For more ADAS news, click here.
