Tesla Pushes Simplified FSD to 4 Million HW3 Vehicles, Causing Overheating Failures

Tesla has released a simplified FSD model for older vehicles, compressing an algorithm designed for the HW4 platform and forcing it onto HW3 hardware. The HW3 autonomous driving computer is equipped with only 8GB of memory, has data bandwidth one-eighth that of the new-generation hardware, and suffers from inadequate thermal design. After the update, motherboards run at full load for extended periods, with temperatures reaching as high as 96 degrees Celsius. Vehicles frequently display high-temperature fault warnings, the FSD driver-assistance function is forcibly disabled, and owners are left to pull over and wait for the hardware to cool down.

The new algorithm version contains perception vulnerabilities, failing to accurately identify pedestrians and obstacles behind the vehicle in reversing scenarios. Some vehicles beyond their original factory warranty have had their entire autonomous driving computers burned out by prolonged high temperatures, leaving owners to bear the substantial replacement costs themselves. These failures are not the result of natural vehicle aging. Tesla has previously acknowledged that the ceiling of HW3 onboard computing hardware falls short of the fully unsupervised autonomous driving capability the brand has promoted.

From a legal standpoint, Tesla bears responsibility to the 4 million owners of HW3-equipped vehicles, and should provide free upgrades to more powerful onboard computers or offer financial compensation. However, Tesla has not introduced a unified compensation or replacement programme. Instead, it has pushed FSD updates that substantially increase the load on this hardware, with sustained high-load operation accelerating motherboard overheating and damage, actively amplifying the original hardware defects.

This incident highlights the regulatory differences in over-the-air (OTA) updates for intelligent vehicles between China and the United States. The United States currently has no pre-filing mechanism for autonomous driving software upgrades, allowing automakers to push major version updates autonomously and deal with safety incidents afterwards. China implemented two mandatory national standards for intelligent driving in 2026, adopting a "review first, then push" model that explicitly prohibits pushing heavy-load algorithms to older hardware that fails to meet computing power and thermal dissipation requirements, and requires that upgrades must not weaken basic safety functions such as reversing detection, thereby avoiding similar large-scale risks at the source.