Musk Backs Down: HW3 Gets Death Sentence, Unsupervised FSD Remains a Distant Dream

A 'Mea Culpa Conference': Even the Iron Man of Silicon Valley Has His Day of Reckoning

On April 22 Eastern Time, Tesla's Q1 2026 earnings call took place as scheduled. What should have been a victory lap showcasing technological prowess unexpectedly turned into a 'mea culpa conference.'

When analysts once again pressed the question that has been asked countless times — 'When will unsupervised FSD actually be available to ordinary vehicle owners?' — the man who once boldly proclaimed 'L5 within two years' finally backed down.

Musk's exact words: 'I'm just guessing here, but probably in the fourth quarter.'

Note the keyword — guessing. This wasn't the supremely confident 'Iron Man of Silicon Valley.' This was a businessman cornered by reality, forced to let the truth slip out.

But the real bombshell came next. He delivered a fatal blow, effectively handing a 'death sentence' to approximately 4 million HW3 (Hardware 3.0) owners worldwide: 'HW3 simply does not have the capability for unsupervised full self-driving. Its memory bandwidth is only one-eighth that of HW4.'

At that moment, Tesla FSD's 11-year history of grand promises reached its most embarrassing finale.

An 11-Year History of Empty Promises: From Bold Claims to Broken Trust

If you're a long-time Tesla owner, this probably feels like swallowing a fly — a live one.

Let's rewind the timeline:

• 2015: Musk confidently told Fortune magazine that 'full autonomous driving will be achieved within 2 years.'
• 2016: Tesla's official website prominently stated that 'all new vehicles have the hardware needed for full self-driving.'
• 2019: At 'Autonomy Investor Day,' Musk solemnly pledged: 'There will be one million Robotaxis on the road next year.'
• 2021: FSD Beta began rolling out to select users, with Musk claiming 'L5 by end of year.'
• 2023: Version V12 was released, with the end-to-end neural network hailed as a breakthrough — 'This time it's really different.'
• 2024: The Robotaxi concept vehicle Cybercab was unveiled, but production timelines were repeatedly pushed back.
• Q1 2026: Musk finally resorted to the word 'guessing.'

Eleven years. Countless rounds of 'next year for sure.' Countless rounds of 'this time it's really different.' Looking back now, this timeline reads more like a carefully scripted rendition of 'The Boy Who Cried Wolf.'

The HW3 Death Sentence: What Happens to 4 Million Owners?

The most devastating piece of information from this earnings call wasn't yet another timeline delay — it was the official verdict on HW3 hardware.

Musk explicitly stated that HW3's memory bandwidth is only one-eighth that of HW4, making it fundamentally incapable of supporting unsupervised full self-driving. What does this mean?

Globally, an estimated 4 million-plus Tesla vehicles are equipped with the HW3 chip. A significant portion of these owners paid an additional $10,000 to $15,000 at the time of purchase for the 'FSD Full Self-Driving Capability' package. Tesla's messaging at the time was clear: the hardware is ready; only software updates are needed to achieve full autonomous driving.

Now, that promise has been torn to shreds by the company's own hand.

What's even more concerning is that Tesla had previously promised free HW4 upgrades for HW3 owners, but there are still no signs of large-scale implementation. Given current HW4 chip production capacity and upgrade costs, performing hardware replacements on 4 million vehicles would be an astronomical undertaking in terms of both engineering effort and financial burden.

This isn't merely a technical issue — it's a ticking time bomb of trust crisis and legal liability. In multiple U.S. states, law firms have already begun soliciting plaintiffs for class-action lawsuits related to FSD marketing claims.

The Technical Perspective: Just How Hard Is Unsupervised FSD?

Setting aside commercial promises, we need to take a rational look at the technical difficulty of unsupervised autonomous driving.

The end-to-end neural network approach adopted in Tesla's FSD V12 and subsequent versions does represent a major shift in autonomous driving technology. Traditional approaches rely on a modular architecture of perception-planning-control, while the end-to-end approach attempts to let neural networks map directly from camera input to driving decisions.

This approach offers an extremely high ceiling, but the challenges are equally apparent:

1. Long-tail scenario handling: In autonomous driving, 99% of scenarios may account for only 1% of the problems. The real challenge lies in those rare but potentially fatal edge cases — temporary hand signals from construction workers, fallen trees blocking the road, wrong-way electric scooters. The end-to-end model's ability to generalize across these scenarios remains questionable.

2. The ceiling of the vision-only approach: Tesla insists on not using LiDAR, relying entirely on cameras. This offers enormous cost advantages, but in adverse weather, strong backlighting, and nighttime conditions, the reliability of a pure vision approach remains an unsolved challenge.

3. Compute bottleneck: Musk's admission that HW3 lacks sufficient computing power underscores that autonomous driving's compute demands far exceed prior expectations. Whether even HW4 or the future HW5 will be sufficient for real-time inference in unsupervised scenarios remains an open question.

4. Validation and regulation: Even if the technology meets the bar, how do you prove system safety to regulators? Data accumulated from Waymo's years of operations in Phoenix shows that even with a 'heavy-duty' approach using LiDAR plus HD maps, unexpected situations still arise. For Tesla's vision-only, map-free approach to clear regulatory approval, the difficulty will only be greater.

Industry Comparison: Waymo Leads, Tesla Trails

Notably, in the same week that Musk 'backed down,' Waymo announced that its driverless taxi service had expanded to its sixth U.S. city, completing over 150,000 paid trips per week.

The strategic divergence between the two companies is becoming increasingly stark:

Waymo's 'heavy-duty route,' while slower to scale and more expensive, has already proven its viability on real roads. Tesla's 'lightweight route,' while theoretically offering greater scalability, has yet to cross the critical threshold from 'supervised' to 'unsupervised.'

How Much Patience Does the Capital Market Have Left?

Autonomous driving and Robotaxi operations carry significant weight in Tesla's current valuation framework. Valuation models from multiple investment banks show that FSD and related businesses contribute 30%–50% of Tesla's market capitalization.

If unsupervised FSD continues to miss deadlines, this valuation premium will face severe scrutiny. In fact, Tesla's after-hours stock price showed notable volatility following the earnings call, with market sentiment shifting from 'unconditional faith' to 'conditional wait-and-see.'

The deeper issue is that Musk's attention is being pulled in multiple directions — xAI's Grok model, SpaceX's Starship, and government efficiency department DOGE affairs. When a CEO simultaneously runs six or seven companies, how much strategic focus can each one receive? This is a question investors must seriously consider.

Outlook: The Dream Isn't Dead, but Trust Is Overdrawn

To be fair, Tesla FSD's technical progress is real and tangible. The V12 version represents a quantum leap in driving experience compared to earlier versions, and the potential of the end-to-end approach has earned broad industry recognition.

But technical progress and commercial promises are two different things.

Musk's biggest problem has never been technical misjudgment — it's his habitual packaging of the most optimistic technical projections as definitive commercial commitments. When 'Elon Time' evolves from an amusing meme into a systematic erosion of credibility, the problem becomes serious.

For 4 million HW3 owners, what they need is not another vague timeline but a clear resolution: either a free hardware upgrade or a refund of FSD fees. Any further delay will only deepen the cracks in trust.

For the autonomous driving industry as a whole, Musk's admission may actually be a positive development — it reminds all practitioners and investors that the road to fully autonomous driving is longer than anyone imagined. Acknowledging this takes more courage than blind optimism.

Unsupervised FSD will arrive eventually, but until that day truly comes, perhaps it's time to stop setting deadlines.