I can't quite remember when it started, but somewhere along the endless scrolling, I fell down the Huawei autonomous driving rabbit hole.

Back then, subsidies were generous β€” national, city, and district subsidies stacked together. An AITO M7 Plus with a sticker price of 249,800 yuan could have 20,000–30,000 yuan knocked off, bringing the on-road price to around 220,000 yuan. The AITO M9 also had trade-in subsidies β€” up to 15,000 yuan for trading in an old car.

Then I started diving into the specs. ADS 3.0 had just launched. Huawei claimed urban mapless NCA covered over 90% of mileage, and highway scenarios were even better. The LiDAR on the M9 was a 192-line unit, with 27 sensors and a theoretical detection range of 250 meters. The Li L9 also used a LiDAR solution in its Max version, priced similarly. XPeng X9 was running XNGP, claiming nationwide coverage across over 300 cities.

I went back and forth, screenshotting spec sheets and bombarding my friends with "look at this LiDAR" until they were thoroughly annoyed.

Then I procrastinated. All the way into early this year.

Subsidies had shrunk significantly. The local add-ons were basically gone, and the national subsidy had been cut. That sweet spot from before had quietly closed.

But I still went for test drives.

Five cars total: the Huawei-linked AITO M7, plus the XPeng X9, Li L9, Zhiji L6, and BYD Han EV with the DiSus system. Figured if I was going to buy, I had to drive them myself to feel confident.

Honestly, after the test drives, I had doubts.

Once on a city street, a scooter was about to cut in from my right. The car reacted maybe half a beat slow. I instinctively grabbed the wheel. That one moment made me realize I still wasn't fully at ease.

Sometimes it was too conservative β€” changing lanes later than I would, occasionally with cars already honking behind before it made the move, crossing the lane line at the last second.

When I'm uncertain, I like to crunch numbers. They help me calm down.

My current car was bought in 2019 for 200,000 yuan. Mid-last year, I casually checked the used price β€” about 30,000 yuan left. Not bad, I thought. This year I checked again: 20,000 yuan. A 10,000 yuan drop in one year, and that rate would likely continue.

I made a spreadsheet and compared two scenarios.

The old car: bought 2019 for 200k, now worth 20k. Lost 180k over six years, averaging 30k per year depreciation. But depreciation slows down β€” last year to this year it only dropped 10k. Going forward it might be just a few thousand per year. I estimated about 10k left in two years, optimistically 12k.

For a new car, budget 200k–300k range. My preferred config would be around 250k on-road. I used 250k as the base and added a 300k option for comparison.

Scenario A: Buy now, sell old car
Sell old car: 20k
New car on-road: 250k
Net spending: 250k - 20k = 230k
That's real cash out of pocket.

Scenario B: Wait two years, buy in 2027
Old car then: estimated 10k
New car then: two possibilities β€” either price wars continue and that 250k model goes for 240k, or specs improve and you end up wanting a pricier version at 280k.

I calculated both:
- Optimistic (new car 240k): 240k - 10k = 230k, same as now
- Conservative (new car 280k): 280k - 10k = 270k, 40k more than now

Looks like the optimistic case is similar? But wait, there's more.

Waiting two years also has hidden costs. The old car still needs to run β€” maintenance and repairs, roughly 10k over two years. At that age, minor issues are more likely.

So the full cost of "waiting two years":
- Optimistic (240k new): 230k + 10k = 240k
- Conservative (280k new): 270k + 10k = 280k

Rough comparison: the optimistic version means spending 10k more to wait two years but getting more mature technology. The conservative version means spending 50k more and still driving an old car for two years.

The math is done. But I still haven't pulled the trigger.

On paper, buying now is probably better value. At worst it's a wash, at best I save tens of thousands. But those test drives still linger in my mind. One accident β€” and that cost can't be quantified.

Two years from now, the tech will be more stable, regulations clearer, liability rules more defined. If something does go wrong, at least we'll know who's responsible.

What do you think? Do the numbers convince you to buy now, or would you rather spend a few extra thousand for a version that lets you sleep at night?