HW3's Legacy: A Financial and Logistical Analysis of Tesla's FSD Obligation

Calculating Tesla's HW3 Autonomous Obligation

Tesla started selling cars with Full Self-Driving (FSD) computers and cameras, known as Hardware 3 (HW3), in 2019. This was the standard for all of their vehicles through 2022 until HW4 (now known as AI4) supplanted them. During the HW3 window of time, Tesla made approximately 3 million vehicles that were sold as fully FSD-ready, "all that's needed is an over-the-air update, and these cars will be able to drive themselves." Now that we're on the cusp of autonomy, are these HW3 cars an albatross around Tesla's neck or an opportunity? 

At a mere 144 TOPs, it's becoming apparent that HW3 likely does not have the compute horsepower or camera clarity necessary to achieve unsupervised FSD. So what's Tesla going to do if they achieve FSD (on AI4 or 5) but HW3 proves to be insufficient? Will they retrofit the HW3 vehicles? If so, how many cars will need to be upgraded, and how much will it cost Tesla (or you)? 

Introduction

The promise of FSD is a monumental technological endeavor that goes far beyond software alone; it is inherently tied to the hardware underpinning millions of Tesla vehicles already on the road. After working with Mobileye (Autopilot 1) and Nvidia (HW2/2.5), Tesla designed their own custom chip for HW3. This marked a critical milestone in their journey. HW3 was built into all Tesla vehicles for nearly 3 years. The HW3 chip was championed as the final hardware piece necessary for full autonomy. Now, with the vast improvements in subsequent hardware generations (HW4, AI5), some owners are wondering if HW3 will be able to cross the unsupervised finish line or if it will be like trying to run Borderlands 4 on a Commodore 64.

Today, we'll examine the size of the HW3 fleet, its FSD adoption rate, the retrofit costs, and how Tesla may deal with upgrades. 

The HW3 Fleet: Size and Timelines

Tesla began installing HW3 in new vehicles starting in April 2019 (replacing the Nvidia-based HW2.5) and continued to use HW3 in primary production until the shift to HW4 in early 2023. This production window generated a large population of vehicles. Based on cumulative delivery data, there are more than 3 million vehicles in the HW3 fleet. This is a substantial number of cars.

The following graph illustrates the estimated size of this fleet over time, factoring in the deliveries until Q4 2022, and then accounting for vehicle attrition with an age-dependent scrappage rate (4.5% annually for the first 12 years).

The cumulative number of HW3 vehicle production peaked at 3,054,357 in Q4 2022. Additionally, some HW2/HW2.5 vehicles from 2017/2018 were upgraded to HW3. That brings our estimated HW3 fleet peak size to ~3.1 million vehicles.

For this exercise, we'll assume:

1. FSD is solved in 2027;
2. Upgrades will be required for these HW3 vehicles; 
3. Upgrades will start in Q4 of that year.

Accounting for scrappage, the active HW3 fleet size at that time is projected to be approximately 2.8 million vehicles. This does not mean that all 2.8 million vehicles will require immediate retrofits; the upgrades will only be required for owners who have purchased FSD. In 2019 through 2022, FSD was not available for purchase in most parts of the world. Tesla officially stated that at the end of 2022, there were over 285,000 customers in the U.S. and Canada who had purchased FSD. That's a 9% FSD adoption rate. Again, using standard scrappage rates, that's still more than 225,000 vehicles requiring retrofit upgrades in 2027. However, retrofits are not the only possible solution (more on this later).

Comparing HW3, AI4, (and AI5) 

Hardware	Compute Power (TOPS)	Comparison to HW3
HW3	144 TOPS	Baseline (1x)
AI4 / HW4	300–500 TOPS	Approximately 2–3.5x more raw capable (With real-world FSD inference gains typically 3–5x)
AI5	2,000–5,000 TOPS (expected)	Approximately 14–35x more capable than HW3 (based on TOPS; production expected in 2026; Elon Musk describes it as 10–40x better than AI4 overall on FSD specific metrics)
Overall Comparison		AI4 has approximately 3 to 5 times the compute capability of HW3. AI5 is projected to vastly exceed this, enabling advanced unsupervised FSD and Robotaxi features, though exact real-world results are yet to be seen.

Can HW3 Do It?

This post generally assumes that an upgrade will be required for HW3 vehicles to achieve full autonomy, but it's important to acknowledge that Tesla is still trying to squeeze as much as possible out of HW3. During the Q3 2025 Tesla Earnings Call, this issue was brought up directly. Here's the response from Tesla management about the future of this massive installed base.

The company's CFO explicitly stated that Tesla is "not abandoning HW3" and offered a clear assurance to concerned customers: "We will definitely take care of you guys." Adding, "My personal daily driver is a HW3 vehicle." Furthermore, Ashok Elluswamy, VP of AI SW, noted that a lighter-weight version of FSD V14, will be coming in 2026 for HW3. This V14 "Lite" will provide owners with many the latest advancements in the supervised FSD, albeit months behind the AI4 deployment.

So, whether or not Tesla finds a way to cram all of unsupervised FSD into HW3, those vehicle owners will have a path to a fully self-driving vehicle. Next, let's look at what those path options may be.

Upgrade Options & Cost

In 2027 (when FSD is solved in this example), let's assume the cost to retrofit a HW3 vehicle with parts (AI5 computer, cameras) and labor is $2,500. For customers who've already paid for FSD, Tesla has an obligation to upgrade them at no cost. For owners who have not paid for FSD, this can be rolled into the purchase cost. However, look at another option.

Tesla has offered FSD transfers (off and on) for some time now (I've even used it). If Tesla were to offer current HW3-FSD owners a $2000 "upgrade incentive" credit towards a new AI5 vehicle with FSD transfer as an early adopter award, many of them might opt for this. They'd get a newer Tesla, they'd be trading in vehicles that are between 5 and 7 years old, and they'd drive off in a native AI5 vehicle (or more likely have the car drive them).

This might mean that Tesla receives a lower margin on these vehicles, but it would stop their service centers from being overrun with retrofit requests, while being cheaper than retrofits.

Above, we estimated that there would be 225,000 HW3-FSD vehicles on the roads in 2027. If Tesla had to upgrade all of these at $2500 each, the total cost would be $562,500,000. However, let's assume, by then, half of the owners will have already upgraded to newer AI4 or AI5 vehicles, and then another 50% of the remaining customers will take advantage of the upgrade incentive. That twindles the "free" upgrade number to just 56 thousand vehicles and reduces Tesla's cost to $140,625,000. This will be easily affordable for Tesla in a future quarter where "FSD is solved" and vehicle and FSD orders are pouring in.

FSD Adoption and Pricing Post-Solution

Once FSD is truly solved, the perceived value of the software will fundamentally change. FSD will transform from an advanced driver-assistance system into a fully validated eye-off, hands-off system. This will allow you to sleep, play games, watch movies, or just look out the window while your car chauffeurs you. It will also be a robotaxi enabler, allowing owners to put their cars into service to generate revenue. This certainty will dramatically increase both the adoption rate and the FSD sale price.

We can conjecture the following changes:

1. Price Increase: The purchase price of FSD (currently $8,000) will increase significantly. It's been as high as $12,000, but this is the "killer app" for cars, and this price is likely to skyrocket. At least $20,000 is a reasonable estimate. It's even possible that in 2028, to have an older Tesla (like a 2018 Model 3) where the ability to transfer FSD to a new car is worth more than the rest of the vehicle. 
2. Adoption Rate Increase: As soon as FSD is solved, the hesitancy surrounding the low (9 to 12%) adoption rate will evaporate. Customers will recognize FSD as a utility or an investment, pushing the take rate on all new vehicles to 50% in 2028 and eventually above 80% in the years that follow. 
3. Retrofit Demand: Every Tesla built since October 2016 has the camera placements to allow retrofits to a new FSD system. The HW2, 2.5, and 3 fleet comprises ~3,542,000 vehicles. This vast population of existing non-FSD-equipped vehicles would become the target of a massive upgrade demand. The potential revenue generated from these post-2027 sales and upgrades would quickly eclipse the cost for the 56 thousand "free" retrofit vehicles.

Crisitunity

Vehicles	Count
Native HW3	3,054,357
HW2/2.5 upgraded to HW3	70,000
HW3 already with FSD	285,000
HW3 vehicles that Tesla will be obligated to retrofit Due to owners not upgrading to AI4+	55,000
HW2/2.5/3 FSD Potential Adoption	2,740,000

A naïve analysis might look at the 3+ million HW3 vehicles and see an obligation to upgrade these to FSD-ready on Tesla's dime. Millions of cars, costing thousands of dollars each, would be a huge liability for any company. However, the opposite is true. Only a small percentage (~9%) purchased FSD. Of those, most have (or will) upgrade to a vehicle with newer native HW and transfer FSD. This leaves the vast majority of the HW3 fleet (97%) with no obligation for free retrofits. And each of these vehicles in the 97% can be converted to a self-driving car if the owner purchases FSD at the higher 2027 price (which more than pays for the retrofit).

Conclusion

Our analysis of the HW3 fleet highlights a moment of both challenge and opportunity for Tesla. The obligation to provide hardware retrofits to tens of thousands of customers who paid for FSD is a substantial financial undertaking, yet one that management has committed to. The explicit promise made during the Q3 2025 earnings call, "We will definitely take care of you guys," codifies the company's intent to fulfill the original FSD promise. The early adopters paid for the development of the technology, and there are multiple ways to reward them for this leap of faith without being financially ruinous. 

The early adopters helped pay the cost for unsupervised FSD, allowing Tesla entry into the vast market that this technology unlocks. This commitment, coupled with the development of the V14 "Lite" software to retain HW3 owners in the ecosystem until a complete solution is developed, ensures that the HW3 fleet remains an active and valuable part of the effort. The hardware retrofit, while costly, will be swiftly offset by the massive increase in FSD adoption rate and the resultant multiplication of the FSD purchase price. This transforms the HW3 legacy fleet from a liability into a highly valuable, revenue-generating asset that paves the way for a more sustainable future of shared autonomous mobility. The proactive communications from the company, especially regarding their promise not to abandon HW3, are essential for maintaining customer trust throughout this transition period.