Tesla (TSLA) has to replace the ‘self-driving’ computer inside about 4 million vehicles or likely compensate the owners of those vehicles.
The liability could be more significant than the largest automotive recall in terms of cost.
In 2016, Tesla claimed that all its vehicles in production going forward have “all the hardware necessary for full self-driving capability.”
Tesla’s use of the term “full self-driving” has changed over the years, but at the time and for years later, CEO Elon Musk claimed that it would mean Tesla owners would eventually receive a software update that would turn their vehicles into “robotaxis” capable of level-4-5 self-driving, which means unsupervised autonomous driving even with no one in the cars.
Advertisement – scroll for more content
Almost 10 years later, this has yet to happen and won’t happen soon in most of the cars Tesla has delivered over the last decade.
Tesla’s claim that its vehicles have “all the hardware necessary for full self-driving capability” quickly proved untrue.
At the time, Tesla was producing its vehicles with cameras, a front-facing radar, ultrasonic sensors, and a “self-driving” computer, called HW2.5.”
Tesla quickly started building new vehicles with a new “HW3 self-driving computer” and admitted that its HW2.5 computer was not powerful enough to achieve self-driving capability.
The automaker started retrofitting existing HW2.5 vehicles for free with new HW3 computers owned by drivers who bought Tesla’s ‘Full Self-Driving’ (FSD) software package.
In 2023-2024, Tesla transitioned to another new and more powerful “self-driving computer”, HW4, in its new vehicles.
Unlike when it transitioned from HW2.5 to HW3, this time, Tesla claimed it would still be able to deliver its robotaxi self-driving capability to HW3 vehicles.
Musk even claimed that FSD will get better on HW3 first, as Tesla’s “focus needs to be on getting FSD on HW3 working super well and provided internationally”. He went as far as claiming that FSD performance on “HW4 will lag at least 6 months behind HW3” because of this.
It took another 6 months, but in January 2025, Musk finally admitted that HW3 computers are not powerful enough to achieve unsupervised self-driving capability.
There are about 4 million Tesla vehicles in the world with HW3 computers:
Hardware Version
Production Timeframe
Estimated Vehicles Produced (Global)
Rollout & Overlap
HW3 (FSD Computer)
Apr 2019 – Late 2023 (phased out)
~4 million (approx.)
Standard in all models from 2019–2022; remained in some cars through 2023. Overlap with HW4 during 2023.
HW4 (FSD Computer)
Jan 2023 – Present (ongoing)
~2.5–3 million (approx.)
Introduced Jan 2023 (S/X first); became standard across all models by early 2024. Overlapped with HW3 in 2023.
When admitting the computer won’t support the promised self-driving capabilities, Musk said that Tesla would retrofit the computers of all HW3 car owners who purchased the FSD package:
I mean, I think the honest answer is that we’re going to have to upgrade people’s Hardware 3 computer for those that have bought full self-driving, and that is the honest answer and that’s going to be painful and difficult but we’ll get it done. Now I’m kind of glad that not that many people bought the FSD package.
Musk says that replacing all the computers will be “painful,” and he is “glad” that “not that many people bought the FSD package.”
Tesla never disclosed the official take rate of its Full Self-Driving (FSD) package, but it did disclose having 400,000 FSD beta testers in North America by the end of 2022.
The take-rate is believed to be much lower globally due to the limited value in other markets where Tesla offers fewer ADAS features under the FSD package.
Globally, it’s safe to assume at least another 100,000 HW3 vehicles with the FSD package, which should bring Tesla’s retrofit requirement to over half a million units.
Musk is right to say that replacing the computers in over 500,000 Tesla vehicles will be “painful.” It will strain its service capacity tremendously, on top of the cost, which will easily surpass $500 million.
But that might just be the beginning.
Tesla promised self-driving hardware in all cars
Musk and Tesla not only made promises to those who bought the FSD package, but they promised anyone buying Tesla vehicles since 2016 had “all the hardware necessary for full self-driving capability.”
As we previously reported, Tesla removed the claim from its website last year and changed the language around the FSD package, which was likely aimed at weakening claims for Tesla HW4 owners, but the case for HW3 owners is more straightforward.
In 2019, Musk claimed “Tesla vehicles are now appreciating assets” because of their future self-driving capabilities. Of course, this proved to be completely wrong.
But there’s one thing that’s true about the value of Tesla vehicles: they would be worth more if they had computers capable of supporting self-driving, which Musk just admitted is not the case. That’s regardless of whether they bought the FSD package or not.
Therefore, there’s a strong argument to be made that Tesla needs to replace computers in all HW3 cars or at the very least, compensate the owners for falsely claiming that the vehicles had “all the hardware necessary for self-driving.”
In fact, there’s already legal precedent for this.
Based on Tesla’s statement that “all cars produced since 2016 have the hardware necessary for full self-driving capability,” the owners of those vehicles need to have all the hardware necessary to have access to these features.
It’s a clear case of false advertising. Tesla says, “Your car has all the hardware necessary for full self-driving,” and when an owner wants to try the features, Tesla tells them, “You have to pay $1,000 for us to upgrade your hardware.” Something doesn’t add up.
Electrek’s Take
I would be surprised if Tesla does as Musk claimed and replaces HW3 computers in any car, let alone over half a million cars, or as it should be, about 4 million vehicles.
It’s too complicated and costly. It would add hundreds of thousands of work hours to Tesla’s already ultra-busy service operations, and it may not even work.
After being wrong about HW2.5 and HW3, the level of confidence in Tesla achieving unsupervised self-driving on HW4 vehicles is not really high, despite HW4 vehicles not only having more powerful computers but also better cameras.
I don’t think it’s realistic to believe that Tesla will enable level 4 or 5 self-driving capabilities in what are, in some cases, almost 10-year-old vehicles through a computer retrofit.
My 2018 Model 3 Performance was originally a HW 2.5 vehicle, and I purchased the FSD package. Tesla upgraded my computer to HW3 in 2019. We are now in 2025, and Musk finally admitted that the computer I bought 6 years ago won’t enable the self-driving capacity I was promised.
My car will never be self-driving, and I don’t believe Tesla will ever offer a free computer upgrade.
I think Tesla will have to compensate every Tesla HW3 owner worldwide. That would mean about 4 million vehicles and a liability of several billion dollars.
At first, instead of the computer retrofit, I think Tesla will use this as an opportunity to encourage people to upgrade, like it did with the “FSD transfer windows.” Maybe it will offer buybacks at a higher rate to compensate owners.
As for those who didn’t buy the FSD package, I don’t think Tesla will offer anything based on Musk’s messaging. It will have to go through the courts.
There are already several lawsuits filed against Tesla over its self-driving claims, and that was before Musk’s admission that HW3 won’t support unsupervised self-driving. I believe that those lawsuits will ramp up this year.
FTC: We use income earning auto affiliate links.More.
A visitor observes a computer bay at the PA10 data center, operated by Equinix Inc., in Paris, France, on Thursday, Feb. 6, 2025.
Bloomberg | Bloomberg | Getty Images
In some advanced economies, electricity infrastructure and cost of utilities are undergoing structural changes because of artificial intelligence-driven demand for data centers.
In the process, U.S consumers could be paying higher utility bills because of the sector shifting costs to consumers, warned a latest paper by the Harvard Electricity Law Initiative.
Meanwhile in the U.K, residents may experience higher wholesale prices in light of a proposed reform to the electricity market that would favor data centers which harness renewable energy.
As pricing concerns emerge, regulation and energy grid reform will take center stage in managing energy prices and meeting changing energy needs.
‘Complex’ special contracts
Special contracts between utilities and data center companies are one of the ways higher costs associated with data centers may transfer onto everyday consumers, identified a report by the Harvard Electricity Law Initiative in March.
Such contracts “allow an individual consumer to take service under conditions and terms not otherwise available to anyone else.” In other words, they can be used to shift costs from data centers to consumers because of the subjectivity and complexity in those contracts’ accounting practices, the report stated.
Moreover, special contracts are approved by the Public Utilities Commission but tend to undergo “opaque regulatory processes” that make it difficult to assess if costs have been shifted from data centers onto the consumer.
To remedy this, the report recommended regulators tighten oversight over special contracts or completely do away with them and opt for existing tariff practices.
“Unlike a one-off special contract that provides each data center with unique terms and conditions, a tariff ensures that all data centers pay under the same terms and that the impact of new customers is addressed by considering the full picture of the utility’s costs and revenue,” according to the report.
Jonathan Koomey, a researcher in energy and information technology, concurs with the need for data centers to pay according to their usage of the energy grid.
“The key point, in my view, is that highly profitable companies who impose costs on the grid with big new loads should pay the costs created by those new loads,” Koomey told CNBC.
Beyond utility companies and regulators, “intervenors in the utility regulatory process also play a critical role,” Koomey said.
Intervenors can include a specific group of constituents or a large commercial or industrial customer who partake in proceedings. They may raise issues pertaining to customer service and affordability and ultimately allow for commissions to hear from a broad group of stakeholders.
“They often can dig deeper than the overburdened regulators into the projections and technical details and reveal key issues that haven’t yet surfaced in regulatory proceedings,” Koomey added.
Overbuilt infrastructure?
Another factor affecting utility prices is the excessive development of energy infrastructure.
Utilities and pipeline companies in the states of Virginia, North Carolina, South Carolina and Georgia are planning a “major buildout of natural gas infrastructure over the next 15 years,” potentially based on an overestimation of data center load forecasts, highlighted a report by the Institute for Energy Economics and Financial Analysis in January.
Proactive decisions on the part of utilities and regulators are needed to prevent ratepayers from being “on the hook” for overbuilt infrastructure, said the IEEFA report.
Policymakers across states have adopted a slew of measures to incentivize, curb and regulate the influx of data center development, from tax breaks to legislative bills, with a focus on ensuring non-data centers consumers do not bear undue costs, according to a report by the Gibson Dunn Data Centers and Digital Infrastructure Practice Group.
Zonal pricing
In the U.K, data centers and consumers face a different pricing challenge amid government plans to transform the country’s electricity market into a decarbonized, cost-effective and secure electricity system.
The zonal pricing scheme that is being explored under the government’s Review of Electricity Markets Arrangements would mark a shift away from uniform pricing to a split electricity market. Under the new framework, consumers in different geographical zones would be subject to different wholesale electricity prices based on the marginal cost of meeting demand at that location.
Modeling from consulting firm Lane Clark and Peacock suggests that Northern Scotland would experience lower wholesale prices owing to their high renewable penetration and relatively low demand.
The rest of the U.K, accounting for 97% of national electricity demand, is poised to see a rise in wholesale prices from the current national pricing model.
The impact on retail prices remains murky as yet.
“It is not clear how this may impact retail prices as wholesale prices are only one part of the overall electricity bill for consumers, and DESNZ still needs to make various decisions,” according to joint comments from Sam Hollister, Head of Energy Economics, Policy, and Investment and Dina Darshini, Head of Commercial and Industrial at Lane Clark Peacock’s energy transition division, LCP Delta.
The DESNZ is the U.K.’s Department for Energy Security and Net Zero.
Will data centers benefit?
While tech firms appear onboard with thelower costs that zonal pricing stands to offer, based on think tank research supported by Amazon, OpenAI and Anthropic, whether data centers do in fact stand to benefit from zonal pricing would depend on their type of operations, according to Hollister and Darshini.
Those potentially well-suited for zonal pricing include data center facilities that handle workloads that can be shifted in time or location, they said.
AI training for deep learning models is one such example. Such workloads can be scheduled during off-peak hours when electricity prices may be lower and synchronized with periods of surplus wind or solar power, which would reduce costs and alleviate grid congestion.
Similarly, data centers that do not need to be close to major urban centers or end users — such as those supporting hyperscale AI training, cloud and large-scale data storage facilities or scientific computing hubs — could also benefit from cheaper electricity when located in regions with high renewable generation and low local demand, Hollister and Darshini said.
However, “not all AI workloads are flexible — real-time inference tasks, such as those used in chatbots, fraud detection, or autonomous vehicles, require immediate processing and would not benefit from time-shifting,” they added.
Latency-sensitive applications such as financial trading and real-time streaming that require close proximity to users would also find zonal pricing “less viable.”
Boosting grid infrastructure
Proponents of zonal pricing point to the benefits of reducing the need to move energy over long distances.
But with the National Energy System Operator’s plans to increase network capability and connect more offshore wind, focusing on grid infrastructure is important, “and zonal pricing won’t eliminate those requirements,” according toHollister and Darshini.
“It’s not just data centers that are going to need this additional capacity on the grid, they’re probably the most high profile ones, but EV charging is going to change the grid. National Grid as an organization have been talking about the change in the demand profile from EVs for a very long time,” David Mytton, a researcher in sustainable computing, told CNBC.
The demands on the energy grid posed by the electrification of vehicles is a challenge shared across the U.S. and U.K.
While the electricity consumption of U.S. data centers is growing at an increasing pace, a report by the Lawrence Berkeley National Laboratory published in December noted that this is playing out against a “much larger electricity demand that is expected to occur over the next few decades from a combination of electric vehicle adoption, onshoring of manufacturing, hydrogen utilization, and the electrification of industry and buildings.”
Given this, the infrastructural and regulatory reforms that emerge out of data center management would be helpful for an imminent era of changing electricity demand, said Mytton and fellow researchers.
A new report claims that President Trump’s tariffs have disrupted Tesla’s plan to source parts for the upcoming Cybercab and Tesla Semi production in China.
The trade war started by President Trump and his constantly changing tariffs has thrown a wrench in the plans of most supply chain managers worldwide.
Tesla is no exception.
For most of its manufacturing programs in the US, the American automaker imports a significant number of parts from China, Mexico, Canada, and Europe.
Advertisement – scroll for more content
This includes its upcoming vehicles: Cybercab and Tesla Semi.
Tesla aims to start production of the vehicles at Gigafactory Texas and a new factory in Nevada later this year and ramp up to volume production in 2026.
Reuters reports that Tesla has suspended plans to source certain parts for the upcoming Cybercab and Tesla Semi from China:
Tesla’s plans to ship components from China for Cybercab and Semi electric trucks in the United States were suspended after President Donald Trump raised tariffs on Chinese goods amid a trade war, said a person with direct knowledge.
According to the report, Tesla was ready to move ahead with the plan when Trump first increased the tariffs on China to 34%, but the automaker is suspending the specific sourcing plans after the most recent increases:
Tesla was ready to absorb the additional costs when Trump imposed the 34% tariff on Chinese goods but could not do so when the tariff went beyond that, leaving shipping plans suspended, said the person, who declined to be named as the matter is private.
Trump raised the tariffs on China to 145% last week, with some expectations announced on Friday — even though Trump later claimed there were no exceptions.
I would take the report with a grain of salt since it is based on a single source, but it certainly makes sense.
The phrase “Trump’s tariffs have disrupted” could be followed by the name of virtually every major manufacturing company globally, and Tesla is no exception.
Due to Tesla’s vertical integration, Tesla shareholders have been claiming that the tariffs would be positive for Tesla, or at least not as bad as they would be for other automakers.
Tesla indeed has impressive vertical integration for the auto industry, but that’s in relative terms. Effectively, Tesla still uses a significant number of parts from other countries, especially Mexico, but also from China.
Mexico would be the most problematic for Tesla, as roughly 25% of the parts of all its vehicle programs built in the US originate from there.
The tariffs on auto parts from Canada and Mexico are currently paused for everything in the USMCA agreement, but Trump signaled that this is only temporary.
As for the tariffs on China, they primarily affect Tesla’s energy business, which relies on cheap Chinese battery cells, but Tesla also imports some Chinese parts for its cars and 145% tariffs will change that.
Tesla, like many other companies, has to start looking for alternatives.
Many of the problems come not only from the excessively high tariffs Trump is imposing on countries, but also from the fact that he keeps changing his mind and making exceptions, making it hard for companies to plan.
In this case, Tesla might have suspended plans with Chinese suppliers only to wait and see if Trump will back off the Chinese tariffs, if Musk can lobby for an exception with the President, whom he helped elect with $250 million in political donations, to shop for suppliers from other countries, or maybe, just maybe, do what Trumps claims his tariffs will do and manufacture those parts in the US.
For some reason, I have doubts about it being the last one, but you never know.
FTC: We use income earning auto affiliate links.More.
It only happens every three years, but it’s spectacular! I’m speaking of course, about bauma – one of the largest trade shows of any kind where heavy equipment manufacturers serving construction, forestry, mining, and more bring out their latest and greatest new job site innovations, and we’ve got a whole bunch of them here, on this special bauma edition of Quick Charge!
With more than two million square feet indoors and twice that outdoors, bauma hosts more than 600,000 guests from 200 countries to see 3,600 exhibitors’ hardware (and, increasingly, software). We’re only going to cover a sliver, but it’s a really cool sliver, you guys – enjoy!
New episodes of Quick Charge are recorded, usually, Monday through Thursday (and sometimes Sunday). We’ll be posting bonus audio content from time to time as well, so be sure to follow and subscribe so you don’t miss a minute of Electrek’s high-voltage daily news.
Advertisement – scroll for more content
Got news? Let us know! Drop us a line at tips@electrek.co. You can also rate us on Apple Podcasts and Spotify, or recommend us in Overcast to help more people discover the show.
FTC: We use income earning auto affiliate links.More.