BYD has a master long term plan in AV

This week, as I posted this tweet on the growth of BYD’s SiC market share from last year to this year, I realized that we have hit a major milestone. In terms of developing and mass producing components that are critical to NEVs, BYD is getting pretty close to having full coverage. It designs and produces its own batteries, all the components in there, the CTB/CTC packs, the electric motor/drive/controller in 8-in-1 or 12-in-1 platforms (aka E3.0 & E3.0 Evo), the power chips, the BMS/TMS, the electric power steering/parking brakes, the suspension systems, the ICE engine, plug-in hybrid systems, the analog chips, rear wheel steering, carbon fiber brakes, LEDs and much more. Fudi Battery, precision, Tech & Motor division are getting pretty close to full coverage of the everything you need for EVs. Once they have this all scaled up, they will not only supply BYD cars, but also plenty of other OEMs. To me, SiC was the last major hurdle in this quest. Now, they’ve managed to develop the fully supply chain for Silicon Carbide and will open the biggest ever SiC fab later this year. Once that happens, they will see significant cost reduction in their EVs.

But there is clearly one part missing and that is BYD’s approach toward the move to AV. A year ago was the height of people dismissing BYD as a company incapable of doing software. It was an extremely stupid thing to say at that time, but it did point to one mistake that BYD did make. BYD was slow in getting into AI and ADAS vs competitors like Huawei and Li Auto. As I since described here, BYD has gotten very serious on ADAS. In fact, Wang Chuanfu spends 20% of his time on ADAS. They have hired a lot of people with the goal of putting at least L2 ADAS on every car. Clearly, even the cheapest can have 1 MMW Radar and 1 main camera for Intelligent cruise control, lane keeping assistance and smart parking options. As is customary with BYD, it has to do everything to keep costs down by eventually internalizing everything.

When Uber partnered up with BYD on that large 100000 vehicle order, it said the following about BYD’s AV ambitions

As a company that has significant AI and software talent, I’m sure Uber would only make this statement after seeing huge progress by BYD here.

BYD basically has several steps when it comes to source components:

1. Get the best it can at reasonable price from any supplier (could be foreign)

2. Bring in additional suppliers (including domestic ones) over time

3. Use its own size to negotiate the best terms

4. Have its own internal team develop competitive alternatives

5. Sub in internally developed parts over time until it can fully get rid of external suppliers

6. Move upstream and internalize as much of the parts in thsoe components as possible. In many cases, replacing foreign chips.

7. Where possible, cultivate domestic suppliers for supply chain security and cost considerations.

8. For critical parts, make strategic investments or do JVs with companies that BYD will rely on (like Robosense, Horizon Robotics, Momenta, Ubtech)

To truly appreciate how far BYD has gone toward achieving these goals, we must actually look at all the major sensors and chips it has invested.

I will start of by look at Millimeter wave radar. MMWRadar is the R when people mention things like 1R1V or 5R1v. It indicates how much radars are used. There are main front view radar and corner radars. The corner radars are lower tech, cheaper and easy to produce. BYD already produced its own corner radar RB1 for a couple of years. As of first 4 months of the year, BYD had a 5% market share here.

BYD is likely not bigger because it may still be sourcing some external ones and its mass market vehicles don’t need as many corner radars.

But BYD has moved into the more advanced front view radar now. Last year, it launched the RF-1 front view radar which uses TI’s AWR2944 platform:

It has now displayed 4D MMW Radar samples, which is the current leading edge technology for front view radars. It’s unclear when these front view radar will start appearing in more models. As can be seen in this list below, BYD’s own models which support NOA was not using them as of Q1 of this year.

But by this point, BYD had shifted from using Bosch to ChengTech radars. I would love to see the full list of front view radar share, but I couldn’t find it.

This does go to the wider point that BYD is such a huge customer. Once it ramps up demand for something, the suppliers need to be ready for that. As in the case of SiC power modules, BYD was not ready to fully supply itself until this year. As long as radar suppliers have an advantage over BYD’s internal product, they will continue to get orders. BYD’s huge demand means that it can source from multiple suppliers.

We see something similar playing out for Camera or Vision. 1V1R means 1 camera and 1 radar. Primary front view camera is extremely important to not only ADAS but also providing rendering of what it sees to driver screen.

As I posted here, BYD is also a major play in front view camera modules. It was already the fifth largest supplier in July. It’s now capable of producing the main 8.3MP CMOS chip as well as the module assembly and the full system for something as high end as primary camera on a car. Now, I assume that it hasn’t been higher on this list, because most of the BYD cars just haven’t gone fully in with front view cameras yet.

The next one to look at is Lidar. A while back, BYD signed a strategic agreement with Robosense and invested them.

As I stated here in a chart, Robosense was the largest Lidar supplier in the first half of the year in China with over 40% market share. Up to that point, BYD is still not a huge user, because only YangWang U8 and some Denza N7 had them. But as with everything else, BYD doesn’t ramp up usage until it’s ready for the cost to come down.

But the part that will drive BYD’s Lidar adoption is not Robosense, but it’s own Lidar production. It had invested in VCSEL producer Vertilite & Raysees in the past couple of years even as it also invested in Robosense. VCSEL is essential for Lidar and other laser applications. In fact, Denza N7 uses laser scanning sensors that are developed internally for preview of road condition. It should not be a surprise that BYD is now displaying its Lidar from BAIC Hyundai.

In the future, BYD anticipates that it can lower the cost of Lidars to 900 RMB, which is much cheaper than the prices that other Lidar suppliers have talked about (~1500). This should not be a surprise since BYD itself is a huge customer, so it can make large investment knowing that it has guaranteed demand for those Lidars. Lowering the cost of Lidars and front view cameras will go a long way in enabling BYD to put advanced ADAS in sub 200,000 RMB vehicles.

At around the same time BYD invested in Robosense, it also established a JV with Momenta to develop ADAS algos for its car. It has 60% controlling stakes in this JV. BYD believes in controlling the core technology in its ADAS. As such, it has to build the central computing platforms and the Operating system (BAS 3.0+ being the latest) for its vehicles

It’s going through multiple different path in employing algos for its ADAS program. Aside from this JV with Momenta, it also recently announced cooperation with Huawei to using their ADS 3.0 for FangchengBao Bao-8 as well as some other mass market brand models. It will likely also seek for low cost and high value solutions with DJI Auto and others. Only the Denza and Yangwang brands will only be using self developed ADAS (likely in JV with Momenta)

But ADAS as a whole goes beyond just the algo. BYD has a huge R&D team for ADAS because it started late and is behind Huawei. As such, it needs to pour in more resources to get better at everything.

Another place where this is obvious is in the AI chip. BYD announced partnership with Horizon Robotics on Journey 5 series chips in 2023. However, it has also been expanding its chip design since 2022 with the goal of developing its own autonomous driving chips and possibly cockpit chips. As BYD showcased in May, it already has self developed 80% of the chips for cars. It’s certainly working on developing the remaining 20%, which also tend to be the hardest to get into.

BYD has also announced using Nvidia Thor for future vehicles. This is likely for Yangwang series vehicles designed to support L4 or L5 ADAS in the future. Since Horizon has developed Journey 6 series of chips and announced BYD as the launch customer, it is likely that the majority of BYD models that support Lidar will be using J6P chips (~600 TOPS) going forward. The Vision only models will likely use either J5 or J6E/M variants (~100 TOPS).

Horizon proudly announced itself as the largest supplier of ADAS chips to BYD and this is likely to expand further in the future.

For now, BYD’s ADAS chip efforts are likely to first replace main AI chips used in cars that need 10 to 30 TOPS computation or for front view cameras that need 5 to 10 TOPS of BEV transformation.

As I discussed here, if Axera can snag orders with an entry level AI chip like M55H, there is no reason BYD cannot design its own lower end AI chips to process Vision data and combine it with radar data. No need to rely on having 2 Journey 3 chips for those low end cars, when you can use 1 of your own chip. The fact that Journey 6 series’ entry level chip J6B did not have BYD as an initial customer tells me that BYD is moving away from Horizon for entry level stuff.

All of this makes a lot of sense from the context of BYD as a large conglomerate. It likes to do everything in house. It is so large that it can scale up production and lower cost with just internal demand. It can have the cheapest Camera module, Lidar, CMOS chips, MMW Radar, Ultrasonic sensors, computation platform and AI chips because its subsidiaries don’t need to make high margins on BYD cars. As long as they contribute to an overall 20-25% gross margin for BYD auto division as a whole, their own margins don’t matter all that much. BYD is also supported by its own electronics subsidiary BYDE that does contracting working as one of world’s larges EMS. As such, they can assemble the computation platforms, the PCBs and all the modules more cheaply than any of BYD’s automotive competitors.

The end result is that BYD will bring L2+ ADAS to the mass market at a lower price point than anyone else. While some have offered Vision only L2+ (like Mona M03) or Lidar supported L2+ ADAS (like Leapmotor C16), it remains unclear whether those can be done in a sustainable manner. If BYD offers a Vision only ADAS to something like Song+ EV or Lidar supported ADAS to something like Seal 06 EV, you know they are doing so at a price point that still generates good margins. That makes these development entirely sustainable.

I feel like BYD is probably 3 years away from deploying ADAS everywhere and probably 5 years away from internalizing all the parts that go into its ADAS platform. But as we’ve seen in its EV development, all of these things will happen. And the large R&D team makes all of this possible.