RAP1/lidar still on track for this year

[Just Passing By]

Everyone seems to be assuming RAP1 + LIDAR is more expensive than the Gen 2.5 solution in the currently shipping R2. Wouldn't it be ironic after all this discussion and angst if RAP1 + LIDAR is actually the lower cost solution.

RJ has said the RAP1 solution is less expensive and that a long range LIDAR sensor was a couple of hundred bucks or low hundreds of dollar. I can easily see LIDAR costs being more than offset by the reduction in processor cost. So that leaves the cost of additional wiring, assuming it isn't part of the standard loom already by default, and the additional assembly cost, which would both seem likely to be low/immaterial.

Sponsored

 

[rfkxyz]

Speculation if Lidar + RAP1 brings early flip to MY28 in winter-spring 2027, or just MY27.5 with MY28 flip coming in R1's typical summer timeframe?

I think ordering ~7/01/2027 is likely since we don't need another car until end-summer next year, but if I need to wait longer for MY28 then that's what Turo is for.

 

[captainjp]

Everyone seems to be assuming RAP1 + LIDAR is more expensive than the Gen 2.5 solution in the currently shipping R2. Wouldn't it be ironic after all this discussion and angst if RAP1 + LIDAR is actually the lower cost solution.

RJ has said the RAP1 solution is less expensive and that a long range LIDAR sensor was a couple of hundred bucks or low hundreds of dollar. I can easily see LIDAR costs being more than offset by the reduction in processor cost. So that leaves the cost of additional wiring, assuming it isn't part of the standard loom already by default, and the additional assembly cost, which would both seem likely to be low/immaterial.

Entirely possible since hardware will be manufactured in-house rather than outsourced.

 

[Jeremy3292]

Everyone seems to be assuming RAP1 + LIDAR is more expensive than the Gen 2.5 solution in the currently shipping R2. Wouldn't it be ironic after all this discussion and angst if RAP1 + LIDAR is actually the lower cost solution.

RJ has said the RAP1 solution is less expensive and that a long range LIDAR sensor was a couple of hundred bucks or low hundreds of dollar. I can easily see LIDAR costs being more than offset by the reduction in processor cost. So that leaves the cost of additional wiring, assuming it isn't part of the standard loom already by default, and the additional assembly cost, which would both seem likely to be low/immaterial.

That just means better margins, not lower price to the consumer.

 

[captainjp]

That just means better margins, not lower price to the consumer.

¯\_(ツ)_/¯

 

[Eric9610]

Entirely possible since hardware will be manufactured in-house rather than outsourced.

Rivian is not manufacturing the processor nor the RAP 1. They designed the architecture of the chip. The whole SOC is being built by TSMC semi. Rivian does not have enough cash to start that kind of business. The problem will be since Rivians volumes are low getting fab slots will be limited. Remember they are fighting with Nvidia, Apple, Qualcomm for production slots.

 

[Mark_AZR1T]

Rivian is not manufacturing the processor nor the RAP 1. They designed the architecture of the chip. The whole SOC is being built by TSMC semi. Rivian does not have enough cash to start that kind of business. The problem will be since Rivians volumes are low getting fab slots will be limited. Remember they are fighting with Nvidia, Apple, Qualcomm for production slots.

Yes, as noted above, Rivian isn't manufacturing the chip. They designed the RAP1 architecture, but TSMC manufactures the silicon, just as they do for Apple, AMD, NVIDIA, and Qualcomm. Rivian controls the ASIC design, but they don't own or operate the fabs or packaging.

I've spent enough years in the semiconductor industry to know that getting first silicon back is only the beginning. RJ has confirmed they have working RAP1 hardware. The much harder part is everything that follows.

The current Rivian platform benefits from over 20 years of NVIDIA's hardware and software ecosystem. Rivian is now building much of that themselves, compiler, runtime, AI stack, and system integration. That's a massive undertaking.

Every first-generation chip has bugs. Every one. The question isn't whether RAP1 has silicon errata; it's whether Rivian can work around those issues in software before customers ever notice.

Bottom line, I think anyone expecting fully mature RAP1-powered R2 customer deliveries this year with LIDAR is setting themselves up for disappointment. I have no doubt Rivian will get there. The question isn't if, it's when.

 

[Marjen]

Everyone seems to be assuming RAP1 + LIDAR is more expensive than the Gen 2.5 solution in the currently shipping R2. Wouldn't it be ironic after all this discussion and angst if RAP1 + LIDAR is actually the lower cost solution.

RJ has said the RAP1 solution is less expensive and that a long range LIDAR sensor was a couple of hundred bucks or low hundreds of dollar. I can easily see LIDAR costs being more than offset by the reduction in processor cost. So that leaves the cost of additional wiring, assuming it isn't part of the standard loom already by default, and the additional assembly cost, which would both seem likely to be low/immaterial.

I agree with this 100% So much talk about the new hardware upper prices. They had this planned from the beginning. They already factored in the new hardware cost to all model pricing they have released. All cars will have the new hardware when it's released, even for people who do not select autonomy+ as all cars need the hardware for people who choose the subscriptions. I believe the launch package deal was just to get more takers before the new hardware was ready as they really need to move R2s.

 

[Just Passing By]

That just means better margins, not lower price to the consumer.

Correct. My lower cost comment was wrt to Rivian's costs not that the vehicle would get cheaper for us.

 

[Mark_AZR1T]

Correct. My lower cost comment was wrt to Rivian's costs not that the vehicle would get cheaper for us.

I don't think it'll be cheaper for Rivian initially, and probably not until RAP2 or later (which is almost certainly already under development). They've replaced much of what came "built in" with the NVIDIA ecosystem, compiler, runtime, optimization tools, validation, and software infrastructure, with their own engineering organization. That's an enormous ongoing expense.

The silicon itself isn't where the savings come from. Rivian doesn't own the fabs or the packaging, they're still paying TSMC and assembly partners. The long-term savings come from owning the architecture and software stack, eliminating NVIDIA's margins, and eventually spreading those development costs across much higher production volumes. Until then, this is a very expensive investment.

Rivian is attempting to create a Tesla-style eco-system from AI/Autonomy to how the vehicle is built, that is what this is all about. Shareholder value.

 

[Just Passing By]

I've spent enough years in the semiconductor industry to know that getting first silicon back is only the beginning. RJ has confirmed they have working RAP1 hardware. The much harder part is everything that follows.

Every first-generation chip has bugs. Every one. The question isn't whether RAP1 has silicon errata; it's whether Rivian can work around those issues in software before customers ever notice.

Bottom line, I think anyone expecting fully mature RAP1-powered R2 customer deliveries this year with LIDAR is setting themselves up for disappointment. I have no doubt Rivian will get there. The question isn't if, it's when.

I don't know what you mean by "fully mature RAP1-powered R2 customer deliveries" but your comments seem to assume that RAP1 silicon will still have significant errata for a launch this year.

However, it seems likely that the reason RAP1 wasn't on the R2 at launch is because they already knew way back in 2025 that they needed more spins of the silicon than they had originally planned for. It had to be this far back in time in order to proceed instead with the Gen 2.5 solution for R2 production in mid-2026.

In other words it seems very likely they have already gone through all the hardware development for RAP1 silicon to the point where the design is good enough to sign off for production. This has to be the case in order to get packaged and tested silicon in time to assemble into boards for late 2026 R2 production. It explains why Rivian and RJ have been confident in consistently stating late 2026 for RAP1, i.e. because because they know the hardware, both RAP1 silicon and the board it goes into, are production worthy.

For context on RAP1 silicon production timing. We used to use a rough 12 week fab manufacturing cycle for production grade silicon plus another 4 weeks for assembly, test, mark and pack, so 16 weeks total. I don't know what TSMC's production cycle is today for whatever process Rivian is using but its possible production RAP1 silicon wafers have already started production in order to meet the end of year availability in an R2.

 

[Just Passing By]

I don't think it'll be cheaper for Rivian initially, and probably not until RAP2 or later (which is almost certainly already under development). They've replaced much of what came "built in" with the NVIDIA ecosystem, compiler, runtime, optimization tools, validation, and software infrastructure, with their own engineering organization. That's an enormous ongoing expense.

The silicon itself isn't where the savings come from. Rivian doesn't own the fabs or the packaging, they're still paying TSMC and assembly partners. The long-term savings come from owning the architecture and software stack, eliminating NVIDIA's margins, and eventually spreading those development costs across much higher production volumes. Until then, this is a very expensive investment.

Rivian is attempting to create a Tesla-style eco-system from AI/Autonomy to how the vehicle is built, that is what this is all about. Shareholder value.

I was talking hardware costs because that is what I believe RJ is referring to. All hardware and software development costs will be amortised over the production volume over time.

BTW I have known about TSMC since the 80's when they started out in the foundry business.

 

[Mark_AZR1T]

I don't know what you mean by "fully mature RAP1-powered R2 customer deliveries" but your comments seem to assume that RAP1 silicon will still have significant errata for a launch this year....

Fully mature as in, it works in a production vehicle that customers drive. That isn't simple. I have no doubt they'll get there. Every first-generation chip has silicon errata. The real challenge isn't the hardware; it's everything that comes after it. Rivian had to build an entirely new software ecosystem around RAP1, compiler, runtime, validation, optimization, and integration. That's the heavy lifting they've never done before.

 

[1oki]

They have been testing the GEN 3 HW for quite some time now. It just wasn't ready for the Q2 launch that RJ wanted so they shipped with HW 2.5. 1st pic was an early prototype I think, 2nd pic is from autonomous day, 3rd pic was a recent spot in California. (Im not too sure about the timelines but making educated guesses on what videos I saw)

 

[technerd]

As a Gen 1 owner who is debating the switch, I'm not getting burned again with Rivian's promises of continuing to push updates out for old hardware. I got my invite today, but think I'm waiting a bit until lidar and Coastal Cloud is out

Sponsored