Rivian Details Uber Robotaxi Economics and Reveals R2 Bill of Materials Is Half That of R1
Evercore ISI's 9th Annual ADAS Conference, September 29, 2026
Rivian executives used Evercore ISI's ADAS conference to lay out the clearest picture yet of how its autonomy bet will translate into revenue, alongside new detail on the cost structure underpinning the R2 ramp. SVP of Autonomy and AI James Philbin and VP of Investor Relations Chip Newcom covered territory ranging from the economics of the Uber partnership to specific bill-of-materials reductions, giving investors a more granular view of a company trying to prove it can scale production while building a credible autonomy stack in-house.
Uber Deal Structure: Vehicle Sales Plus a Mileage-Based Licensing Fee
The most concrete new disclosure centered on how Rivian will actually monetize its robotaxi partnership with Uber. Newcom confirmed that Uber or its fleet partners will purchase the vehicles outright, with Rivian recognizing that revenue through auto sales and COGS. But layered on top is a separate software licensing fee tied to vehicle utilization. "Think of it around utilization as it's driving around on the mileage," Newcom said, declining to specify a per-mile rate but confirming the structure resembles a dollar-per-mile royalty on Rivian's autonomy stack rather than a flat software fee.
On the capital side, Rivian expects to hit the second funding milestone under its agreement with Uber — unlocking $250 million of the total $1.25 billion commitment through 2031 — in the fourth quarter of this year. The initial 10,000-vehicle robotaxi purchase commitment remains contingent on Rivian reaching L4-capable performance, with the service targeted for a late-2028 launch. Newcom was explicit that specific technical and geographic milestones, including expansion to 25 markets by 2031 with at least one in Europe, will govern future tranches, though he declined to detail the exact triggers.
Multimodal, End-to-End Autonomy Stack Positioned Against Both Tesla and Waymo
Philbin spent significant time differentiating Rivian's approach from both ends of the industry spectrum. Unlike Tesla's camera-only strategy, Rivian is committed to multimodal sensing — cameras, radar and LiDAR — arguing it accelerates model performance rather than conflicting with an end-to-end training philosophy. "The model sees every sensor essentially the same way," Philbin explained. "It's a bag of numbers, and it encodes them using its own encoders. So it's able to figure out those cross-correlations and how to best benefit the output using its inputs." He noted that LiDAR costs have fallen from the thousands or tens of thousands of dollars to just a few hundred dollars per unit, which is why Rivian is comfortable absorbing the sensor cost within vehicle COGS rather than treating it as a separate autonomy charge.
Against Waymo's approach, Philbin argued Rivian's advantage is a large, already-shipped customer fleet generating diverse real-world data without needing to expand city by city. He also took a swipe at the heuristic-heavy engineering that underpins traditional L4 development, drawing on his own background at Level 4 companies: narrow-road negotiations that once required "very, very complicated" rule sets "just fall out of the model" under an end-to-end approach, he said, calling it validation that this is "the right scalable approach."
Point-to-Point Autonomy Timeline and the Regulatory Patchwork
Rivian remains on track to begin rolling out point-to-point autonomy to early adopter customers by the end of 2026, with full rollout across the Autonomy Plus fleet targeted for later next year. The initial release will run on existing Gen 2 Phase II and Phase III hardware, not requiring Gen 3 compute. Philbin identified three gating factors to eyes-off driving: raw model performance on long-tail edge cases, offline cloud-based validation methodology, and regulatory approval, which he described as "a complete patchwork" across states, with some treating eyes-off systems as equivalent to L4 and others, like New York, offering no pathway at all.
To source the edge cases that actually move model performance, Rivian has built what Philbin called a "grid system" using vision-language models to scale collection against a roughly 3,000-query set of long-tail scenarios run continuously across the customer fleet, with a machine-learning-driven trigger system that can be updated dynamically without waiting for an OTA cycle. Philbin was unambiguous about the limits of simulation relative to fleet data: "There's no real replacement for real-world data... it's a combination of multiple long-tail events that all sort of congregate into a single scenario," making such situations nearly impossible to construct synthetically.
R2 Bill of Materials Cut in Half, With Specific Line-Item Examples
On the manufacturing side, Newcom reiterated guidance from the second-quarter call that R2's bill of materials will run roughly half that of the R1, with non-BOM costs falling even further. He offered a concrete example: shifting from a fully active air suspension to a semi-active MacPherson strut configuration cuts suspension cost by 70%. Newcom also contrasted Rivian's current supplier leverage with its position during the R1 launch, when the company was "relatively speaking, an unknown operator" negotiating "hat in hand" with suppliers during the pandemic. That dynamic has reversed as volume commitments have grown, giving Rivian more pricing power heading into the Georgia plant buildout.
Commodity cost offsets are not uniformly favorable, however. Newcom flagged that DRAM and lithium prices are both rising, even with Rivian's lithium hedging program in place, partially offsetting the structural BOM reductions from design simplification.
Capacity Plan: Normal at 215,000 Units, Georgia Adding 300,000
Rivian's Normal, Illinois plant carries an annualized run-rate capacity of 215,000 units, with 155,000 to 160,000 of that allocated to R2. The company has broken ground on the Georgia facility, currently in stamping press and foundation work, which is expected to add 300,000 units of annual capacity dedicated to the midsize platform — R2, R3 and the Uber robotaxi variant — bringing combined capacity across both plants to more than 500,000 units.
In-House Silicon and the Gen 3 Margin Benefit
Rivian's forthcoming Gen 3 compute platform, RAP1, will begin rolling out to employee vehicles later this year ahead of a 2027 customer launch, prioritized for R2. Newcom noted a margin benefit beyond raw performance: by moving to in-house silicon, Rivian stops "paying away that margin to our silicon provider." Philbin added that the TPU-like components on the chip are built entirely in-house, with Rivian retaining the IP, while other blocks on the chip are developed in partnership with external providers.
VW Joint Venture as the Template for Future OEM Deals
Asked directly about the prospect of additional OEM partnerships beyond Volkswagen, Newcom declined to confirm any pending deals but pointed to the VW joint venture — which will debut its first vehicle, the ID.1, next year using Rivian's zonal electrical architecture and software — as the proof point the company needs before pursuing further tie-ups. Philbin clarified that the autonomy stack itself sits outside the VW JV, a deliberate choice reflecting Rivian's intent to potentially license that technology separately to other automakers later, once point-to-point is demonstrated in customer vehicles.
On demand, Newcom pushed back on the narrative that industry-wide BEV pullback signals weakening appetite for Rivian's vehicles specifically, arguing that EV adoption has stabilized and that the bigger constraint to date has been a lack of differentiated vehicle design rather than consumer resistance to electrification itself.