Rivian‘s founder and CEO RJ Scaringe has reaffirmed his ambition to challenge Tesla in the self-driving space, as the two EV makers pursue increasingly divergent approaches to autonomy.
Speaking with Matthew Berman published Thursday, Scaringe confirmed the R2 mid-size SUV will ship with LiDAR sensors “late this year,” following the model’s initial Launch Edition rollout.
Rivian debuted the R2 Performance on Thursday, priced from $57,990, with deliveries planned for “later this spring” — according to Scaringe.
A more affordable entry-level variant at $45,000 is expected to follow next year.
The company’s autonomy stack in upcoming vehicles will combine 65 megapixels of camera coverage with LiDAR hardware, backed by purpose-built silicon.
“We have a very high level of compute. We built an in-house inference platform, an in-house chip that runs that. It’s an 800 TOPS [Trillion Operations Per Second] chip, so it’s a very, very powerful chip,” Scaringe said.
This hardware-heavy approach, which Rivian calls an end-to-end neural network, puts it at odds with Tesla, which relies exclusively on cameras for its Full-Self Driving (FSD) software.
On the question of scale — a critical factor in training autonomous driving models — Scaringe acknowledged Tesla‘s lead while positioning Rivian as a credible second.
“We have a very high level of compute,” he noted, adding that Rivian will “have the second largest fleet” in terms of data collection, even as Tesla‘s “got a huge fleet.”
Rivaling Tesla
In the past few months, Scaringe has been pointing out that LiDAR has become a “really charged” question.
“10 years ago, LiDAR would cost, depending on which, like if you had a 64-channel LiDAR, it was like a $30,000 sensor,” he said, adding that “now, a high-performance, long-range LiDAR, you can buy for a couple hundred bucks.”
According to Scaringe, the sensors are not the most expensive component of the hardware stack, undermining the cost argument often cited against LiDAR adoption.
“The most expensive part in a self-driving system is the brain. We have to recognize that that’s no longer the reason,” he said.
To Scaringe, a model that’s trained end-to-end can “actually benefit from having multiple sensors.”
“That’s why we don’t have one camera or just two cameras, like two eyes, like a human would have,” the CEO flagged, aiming at Tesla‘s vision-only approach.
Scaringe noted that LiDAR is widely used across the industry for training purposes — including by Tesla — describing the sensors as “part of their ground truth fleet” for model training.
And while Rivian doesn’t have a fleet “that’s nearly as large as Tesla,” they aim to achieve the second largest in terms of data flywheel.
“So, our approach is we have to catch up with and compete with Tesla,” he admitted.
For that, Rivian is betting on more cameras, with “better dynamic range,” supplemented by LiDAR which “provides better safety case for edge case conditions and allows us to train the vision models faster.”
“And the incremental cost to do that is relatively small,” he added, saying that the company was able to offset these costs by bringing inference in-house — as presented by the company last December during its Autonomy & AI Day.
In an “infinite long term,” Scaringe sees a way to “make the case that once the models are very, very robust, you could have less cameras, or you may be able to get away with less radar.”
Training Fleet
FSD is compatible with vehicles equipped with Hardware 3 (HW3) and Hardware 4 (HW4).
HW3 units are, however, unable to support the software’s more recent versions — considering launching a V14 Lite variant to address that limitation.
Since introducing HW3 in 2019 and through the end of 2025, the company has sold approximately 8 million vehicles globally.
Tesla disclosed during its fourth-quarter earnings call in late January that FSD had “nearly 1,100,000 paid customers globally,” as the company works to accelerate adoption of the software.
According to the company’s website, as of Friday, cumulative miles driven using the Full Self-Driving software had surpassed 8.6 billion, with 1 billion of those logged in just the first 50 days of the year.
Rivian, on the other hand, has sold 165,200 vehicles since 2021.
It has guided 2026 deliveries to between 62,000 and 67,000 units.
Rivian only overhauled its approach to autonomy in mid-2024, coinciding with the introduction of the second-generation R1 models.
In the year and a half since — covering the second half of 2024 and full-year 2025 — the EV maker delivered approximately 66,500 vehicles.
That figure suggests Rivian‘s training fleet could represent less than 10% of Tesla‘s, though neither company’s numbers include dedicated training vehicles, reflecting only customer deliveries.
According to Scaringe, the company’s approach is allowing it to introduce levels of autonomy “faster,” as it progresses towards Level 4.
However, “it’s hard to say” when they will become available.
The CEO pointed to “2045, or 2040, or 2038 if, you know, we may be able to optimize sensors,” as the “big question is how well we can cover the corner cases.”
