New findings in the Tesla Electronic Parts Catalog suggest that the company is preparing to release, or has already begun using, a new HW 4.5 FSD computer in new cars. While the moniker suggests another minor revision, like the HW4.1 revision that made some minor changes, the technical reality could be a step up in computing.
The Findings
Eagle-eyed observer Todd deRego spotted the new component, explicitly listed in the official Tesla Electronic Parts Catalog.
Part: CAR COMPUTER - LEFT HAND DRIVE - PROVISIONED - HARDWARE 4.5
Part Number: 2261336-S2-A
Price: $2,300.00
The price point is similar to that of other FSD computers on the EPC, suggesting this is likely intended as a replacement or revised part, rather than an entirely new item for retrofit.
The 3-SoC Theory
The most compelling insight comes from longtime Tesla firmware hacker Green. While he hasn’t physically inspected these new units yet, he noted that Tesla’s firmware has contained references to a 3-SoC (System-on-Chip) design for quite some time.
Historically, Tesla’s FSD computers (HW3 and HW4) have utilized a dual-SoC design. This provided the standard level of redundancy: two identical chips processing the world simultaneously. If HW4.5 is really a 3-SoC architecture, it opens the door to capabilities that the current HW4 simply cannot support.
Why 3?
It’s not clear exactly what Tesla would do with the third node, but there are several possibilities.
The most obvious benefit is raw throughput, enabling larger and more complex inference models. As FSD v14 and other upcoming models grow exponentially in size (parameter count), they require more memory and compute to run efficiently. A 3-SoC setup allows Tesla to split the inference load across more silicon, running larger, more intelligent neural networks that would otherwise choke the standard HW4 unit.
However, it can also change how the vehicle thinks and validates its decisions. In a 2-chip system, if the computers disagree, the system faces a dilemma: which one is right? This could result in a safe disengagement and a red-hands takeover warning until the vehicle can resume. In a 3-chip system, Tesla can implement Triple Modular Redundancy (TMR). If one chip hallucinates an obstacle but the other two see a clear road, the system can “vote” to ignore the error and continue driving smoothly. This level of fault tolerance adds another layer of safety to the entire system.
A 3-SoC architecture could allow the car to continue driving using two chips with standard redundancy as it does today, while the third chip runs a newer, experimental version of FSD in the background - Shadow Mode. This would allow Tesla to validate next-gen software in the real world on consumer cars with zero safety risk, gathering data on how the new system would have behaved compared to the active driver or current iterations of FSD.
A Bridge to AI5
With AI5, Tesla’s next-generation FSD chip, slated for production later in 2026 and with volume production in 2027, there is a gap if Tesla needs more computing now. Software progress often outpaces hardware, and this could be a stopgap until AI5 is ready.
Hardware 4.5 could potentially be the bridge to that gap. By introducing a slightly more capable iteration of current technology now, Tesla ensures that millions of cars rolling off the line between now and AI5’s deployment aren’t left behind as neural networks continue to swell in size and complexity.
