BYD bringing ADAS to everyone

On February 10th, BYD threw down the first of many gauntlets this year when it pushed out autonomous driving for everyone by unveiling DiPilot-100/300/600. While all 3 are interesting, the focus of the night was on DiPilot-100. This was the ADAS package that would become standard for all the BYD models in the 100k+ RMB market. The most amazing part is that BYD added this feature without adding to price at all. The only variants that did not receive DiPilot-100 were the entry models of SeaGull, Qin+ DM-i and Seal 05 DM-i. Many people were stunned when even a 78.8k RMB Seagull came with DiPilot-100. Here are all the offered models

The effect on the market has been huge. Huawei’s Yu Chengdong quickly came out with Weibo post proclaiming that there is a difference between just having ADAS vs having a good ADAS. The entire episode seemed to be very salty to me. Huawei is competing against BYD indirectly in this market. Its ADS SE is been promoted for many BYD competitors in the 150k+ RMB market. So, if BYD can offer ADAS for free, it will have a real adverse effect on sales of competitor models that use ADS SE. If we compare ADS SE to DiPilot-100 head on, you will see that BYD’s solution actually has 1 more frontal camera and 2 more MMW radar. This compares the two:

So even if DiPilot-100 is likely inferior to ADS SE right now, the mass amount of quality data coming into BYD auto cloud in the next few months will allow it to quickly catch up to ADS SE in performance soon enough. Both of them support various advanced parking model and Highway Navigation On autopilot (HNOA). ADS SE supports City ICC+ (Intelligent cruise control) for city autonomous driving while DiPilot-100 will support memorized NOA (MNOA) for city routes by end of this year. MNOA is an interesting concept. It allows drivers to specify 3 or more routes in city to support for Navigation on autopilot. For example, specify the routes between home and work, home and school, home and supermarket. This + HNOA would provide NOA for vast majority of customer driver needs. Beyond NOA and various self driving modes, BYD also added many advanced parking modes, including automatic parking assistance, remote parking assistance and valet parking mode.

While there were fewer details on DiPilot-300 and 600, we know that both of them have support for Lidar and more powerful computation platforms. The following compares them:

DiPilot-300 uses 1 Nvidia Orin-X chip for close to 300 TOPS in the computation platform as well as 1 to 2 Lidar. All the model variants priced at above 200k RMB had at least 1 Lidar. They also all support Nationwide end-to-end soft map City Navigation on Autopilot (CNOA). That is currently the golden standard for L2+ ADAS. DiPilot-600 uses 2 Nvidia Orin-X chips (or maybe Horizon’s J6P chip) for close to 600 TOPS in computation as well as 3 Lidars. Currently, only YangWang models use this configuration. BYD believes this configuration has the hardware necessary for L3 ADAS. This is sensor config for DiPilot-300

For many people, the safety part of autonomous vehicles are the most attractive. All the DiPilot-100 vehicles can already achieve automatic emergency braking (AEB) of up to 100 km/h. They are currently working on raising this to 120 and then 140 km/h. This will allow drivers to avoid many accidents at high speed. There are other active safety features like blind spot detection, lane keeping assistance. Front collision warning, rear collision warning & ELKA. The advanced surrounding sensors + 100 TOPS of compute power offers a lot of active safety protection over human drivers.

I took a look through all the new ADAS versions that came out on Monday. In most cases, the upgrades went further than just adding DiPilot-100. Most of the models also added DiLink-100 cockpit, which uses 6nm cockpit SoC and has better performance than Qualcomm Snapdragon 8155. More CPU computation, faster DRAM, much faster download speed & more advanced process.

This allows for much improved cockpit AI including support for DeepSeek-R1 model. The addition of surrounding cameras also allow the addition of Sentry mode across most BYD models so you can see the surrounding views around the EV, For many people, the addition of Sentry mode and various parking modes are the main draws for buying these upgraded BYD vehicles. In other cases, they also added improved rear suspensions, upgraded interior, improved screens and even a fridge in the case of BYD Dolphin. The new DM models all got upgraded to using DM5.0. The EV models seem to be upgraded to using E3.0. All in all, a lot of different upgrades with only a couple of variants changing prices. So far, orders are up 30% over normal, but that may go up once all the ADAS models actually arrive in the stores.

The most interesting part is just how BYD has been able to keep the cost down for DiPilot-100. This is the version that even Seagull use. How can you make the ADAS upgrade free for a 78.8k RMB unless you can source sufficient chips, sensors and have enough assembly capacity for the computation platform? Only, BYD can do it. Not only that, it has sourced all Lidars and Orin-X chips it needs for putting DiPilot-300 on all the models > 200k RMB

As you can see, DiPilot-100 uses Nvidia & Horizon chips. 300 & 600 use Orin-X. BYD is also developing an 80 TOPS ADAS chips that will further lower its cost. All the algo for DiPilot-100 is self developed whereas higher end use Momenta JV or Huawei. All the computation platforms and domain controller are supplied by BYD Electronics. The 12 camera costs can be lowered to 100 RMB each or 1200 RMB overall. This is supplied internally & externally. The Lidar cost can be lowered to 1500-2500 RMB with Robosense getting 80% of the order & Hesai getting 20%. The 5 MMW Radar cost can be lowered to 500 RMB overall with most coming internally. The 12 Ultrasonic sensors cost 300 RMB overall with self supply accounting for most. based on the above, BYD can hold the hardware cost to under 5000 RMB. That allows it to swallow the cost while keeping sales price the same.

The effect of BYD’s move on the market is huge. BYD had 13% of market from 0 to 100k and 29% of the market from 100-150k RMB. This segment is ripe for disruption. At this point, it would be very hard for competitors to add these features without increasing the sales price. It would also be very hard for competitors to even source the needed sensors from suppliers since this is a dramatic ramp up for suppliers. Maybe they can get enough supplies for a few models in a few months, but BYD is delivering them to store today on all its models. This from 36kr article was very interesting for me recently

BYD can get cost down under 5000 RMB, but competitor cost is 6000 to 7000. How do you make up the 2000 cost differential? More importantly, BYD ordered > 1 million chips from both Nvidia and Horizon. It also ordered Lidars for > 10 models from both Robosense and Hesai. What is left for competitor to do?

BYD says its has the largest auto cloud data set and deploying DiPilot-100 on this many vehicles at the same time will drastically increase the data it gets. That is gold for where it wants to get to. It is also currently building up its own data center, but it has been using local Shenzhen and Guangdong data centers for compute. One can imagine as it is accessing this huge data set, its ADAS will increasing rapidly. We should see very fast improvement through OTAs. The hardware is already better than Huawei’s ADS SE. Therefore, DiPilot-100’s ceiling is pretty high.

And over time, this will also be deployed globally. That’s a real game changer. Who needs Tesla FSD when BYD gives away its ADAS for free?