Tesla FSD Hardware 4.5 Appears: A 3-Chip Upgrade Before AI5?

Entries in Tesla’s Electronic Parts Catalog point to a new Full Self-Driving computer labeled Hardware 4.5, which appears poised for rollout—or may already be in use. Despite the incremental-sounding name, the change could represent a meaningful increase in computing capability beyond prior HW4 revisions.

What the EPC shows

Spotter Todd deRego found the new listing in Tesla’s official catalog with the following details:

• Part: CAR COMPUTER - LEFT HAND DRIVE - PROVISIONED - HARDWARE 4.5
• Part Number: 2261336-S2-A
• Price: $2,300.00

The pricing aligns with other FSD computers shown in the EPC, suggesting this is a replacement or revised part rather than a standalone retrofit offering.

The 3-SoC theory

Longtime Tesla firmware researcher Green has noted that Tesla’s software has referenced a three‑SoC (System‑on‑Chip) configuration for some time, although he has not yet inspected these units directly. Historically, Tesla’s FSD computers (HW3 and HW4) used two SoCs for redundancy, with each chip processing the driving scene in parallel. If HW4.5 adopts a three‑SoC architecture, it could unlock capabilities the current dual‑SoC HW4 cannot support.

Why three chips?

First, more silicon can deliver higher throughput, enabling larger and more complex inference models. As FSD v14 and other upcoming networks scale in parameter count, they demand additional memory and compute to run efficiently. A three‑SoC design would let Tesla distribute the workload to handle bigger, more capable neural networks without overwhelming standard HW4.

It also introduces new fault‑tolerance options. With only two chips, a disagreement forces a dilemma and may trigger a safe disengagement until the system can continue. With three chips, Tesla could use Triple Modular Redundancy (TMR): if one processor misinterprets an object and the other two agree on a clear path, the system can “vote” to ignore the outlier and proceed smoothly.

Another possibility is parallel experimentation. Two chips could continue to run the production FSD stack with standard redundancy, while the third executes a newer, experimental build in the background—often referred to as Shadow Mode—so Tesla can validate next‑generation software against real‑world driving without adding risk.

A bridge to AI5

Tesla’s next‑generation FSD chip, AI5, is slated to begin production later in 2026 with volume production in 2027. If additional compute is needed sooner, Hardware 4.5 could serve as an interim step, ensuring vehicles built before AI5 arrives aren’t constrained as neural networks grow in size and complexity.