Einride brings Nvidia’s Alpamayo into its trucks
Einride’s AI test puts a road date on the calendar: second half of 2027. Einride will test Nvidia’s Alpamayo driving model in its trucks, with a live road evaluation planned for the second half of 2027, FreightWaves reported on 22 October. First the model runs in “shadow mode”: it runs live but does not drive. Our view: this is a research step, not new capacity. Nvidia itself calls it exploring. Einride had more than 5,400 driverless hours across six deployments by 30 June. For forwarders, we think driverless European capacity is a 2028 story at the earliest. Our call: the 2027 road test will run with a human safety backstop, and no Alpamayo-driven Einride truck carries paid freight without one before 2028.
Nvidia’s AI is getting into truck cabs, but Einride’s European driverless lanes are still years away
Einride, the Swedish autonomous trucking company, will test Nvidia’s Alpamayo driving model inside its own self-driving system. A live road evaluation is planned for the second half of 2027. FreightWaves reported the deal on Thursday 22 October. No financial terms were disclosed.
The test starts in what the industry calls shadow mode. The model runs live in the truck, watching the road and deciding what it would do, but it does not steer, brake or accelerate. Einride’s existing safety systems stay in control. The model’s decisions are then compared with Einride’s own driving planner and with how human drivers handled the same situations.
Get The Freight Brief by email
What's going on in freight, in your inbox every morning. Free.
Our view is simple. This is a research project with a road date attached. It is worth knowing about because it shows where the technology behind driverless trucks is heading. It is not a capacity event. No forwarder should change a 2026 or 2027 lane plan because of it.
What Alpamayo is, in plain terms
Alpamayo is a family of AI models Nvidia introduced at the CES technology show in January. Nvidia calls them reasoning-based vision-language-action models. Put simply: the model looks at camera and sensor data (vision), can describe and reason about the scene in words (language), and then outputs a driving decision (action). The idea is that a model that can reason about a strange situation copes better than one that only follows patterns it has seen before.
These are very large models. According to FreightWaves, Alpamayo 1 has 10 billion parameters (the adjustable values inside an AI model; more parameters usually means more capability and more computing power needed). Alpamayo 2 Super has 34 billion. Nvidia says the models are meant to “serve as large-scale teacher models”. That means developers use the big model to train much smaller “student” models for their own driving systems. FreightWaves says Alpamayo 2 Super is built to be distilled into compact models that run on Nvidia’s in-vehicle computer.
PlusAI, which is adapting Alpamayo for trucks, gave real numbers for this. Its teacher model has 10 billion parameters. Its student model, which runs on the truck’s own hardware, has 0.5 billion (PlusAI, 16 March 2026).
FreightWaves points out that Einride’s release does not say which Alpamayo model it will use. That matters. The gap between a 34-billion-parameter teacher in a data centre and a small model running in a moving truck is where most of the hard engineering sits.
Why we read this as early-stage work
Start with the language. Nvidia’s own description is cautious. Rishi Dhall, Nvidia’s vice president of automotive, told FreightWaves: “we’re exploring how reasoning-based AI can help trucks navigate complex driving conditions”. Exploring is not deploying.
Einride’s chief technology officer, Henrik Green, framed the company’s role narrowly too. “Being the first trucking company in this engagement gives us the opportunity to help shape the model for freight,” he said. That is a statement about influence over a product still being built.
Then look at the sequence. The model runs in shadow mode first. The live road evaluation comes in the second half of 2027. Even a successful road evaluation is a test, not a commercial service.
And look at the truck. Einride’s next system is being fitted to DAF’s electric truck. DAF is owned by Paccar (Trucking Dive). According to FreightWaves, Einride, DAF and the Dutch research body TNO are working through 2026 to define and test how the system fits DAF’s platform. Software integration and commissioning are scheduled for 2027. Trucking Dive described the same plan as interface testing in 2026, then integration, validation and functional testing in 2027. So the truck platform and the AI model are both still moving parts in 2027.
What Einride already has
None of this means Einride is starting from zero. Green’s other line to FreightWaves sums up its case: “Our advantage has always been running autonomous freight in the real world, for real customers.”
The numbers back that up, within limits. As of 30 June, Einride reported more than 5,400 driverless hours in contracted customer operations across six autonomous deployments in the US and Europe (FreightWaves). That is a real operating record, but it is six deployments, not a network.
The Alpamayo deal also builds on a bigger one. On 21 September, Einride said it would build the next generation of its autonomous system, the Einride Driver, on Nvidia’s Hyperion platform. Hyperion is Nvidia’s reference set of computer hardware, sensors and software for self-driving vehicles. Trucking Dive reported Green saying at the time: “Building the next generation of the Einride Driver on the NVIDIA Hyperion platform, will enable us to scale”.
The scale plans are large. Einride plans to grow from 250 to 750 trucks by the end of 2027, helped by a recent partnership with Tesla. It estimates it needs 1,500 to 2,000 trucks to reach cash-flow neutral, the point where the business stops burning cash, and projects that for 2028. It says joint business plans with customers represent up to $800 million in potential annual recurring revenue, and that about 80% of that demand could be automated in the medium term (all from Trucking Dive, citing Einride’s August earnings presentation and call).
By Einride’s own estimate, the automated share is a medium-term goal, not today’s fleet.
The race to put chip-maker AI in trucks
Einride is not alone. PlusAI announced in March, at Nvidia’s GTC conference, that it is bringing Alpamayo to its Level 4 trucks. Level 4 means the truck can drive itself with no human needed within a defined area or set of conditions, such as a specific motorway corridor. PlusAI expects driverless commercial operations on factory-built trucks to begin in 2027. Its truck-maker partners include International (part of Traton), Scania, MAN, Iveco and Hyundai, and it lists DSV among its partners (PlusAI).
In PlusAI’s announcement, Dhall said PlusAI is “demonstrating how generative AI can move from the data center to the driver’s seat.”
The pattern matters more than any single deal. A few years ago each autonomous truck developer built nearly all of its own driving software. Now the biggest chip supplier offers a ready-made “brain” that developers tune to their own trucks and routes. Einride says adapting Alpamayo should shorten development cycles and cut the cost of extending its system to new vehicle types and routes (FreightWaves).
Our view is that this is the most important point for forwarders. If it works, the expensive part of opening a new driverless lane shifts. Less money goes into writing driving software from scratch. More goes into validation, safety approval and mapping the specific route. That could make the second and third lane much cheaper than the first.
It also concentrates power. If several truck autonomy firms run on the same Nvidia models and hardware, Nvidia becomes a supplier that shippers and forwarders indirectly depend on. That is not a problem today. It is worth watching when contracts for driverless capacity start to include technology and service-level clauses.
What this means for forwarders and shippers
For a forwarder planning European road capacity, the practical takeaway is timing. The first live road evaluation of Einride on Alpamayo is in the second half of 2027. The DAF truck it runs on is still being integrated in 2027. Einride’s own cash-flow target sits in 2028. In our view, that makes driverless European lanes on this technology a 2028-and-later story at the earliest.
In the US, the picture is a bit nearer. PlusAI is targeting 2027 for driverless commercial runs on factory-built trucks. In our view, US lanes will keep moving first.
So what should a forwarder actually do? Three things, in our view:
- Treat autonomous capacity as a pilot option, not a planning assumption, in any 2026 or 2027 European road tender.
- If you have steady, repeatable lanes between fixed sites, such as a port terminal and a nearby warehouse, those are the routes developers like Einride want first. It costs little to be on their list.
- When a carrier pitches autonomous capacity, ask which model and hardware it runs on, and whether it has passed independent safety assessment. PlusAI, for example, says its system is scheduled for independent assessment against three international standards covering functional safety, cybersecurity and AI safety.
Our prediction
Our call: Einride’s live road evaluation of Alpamayo in the second half of 2027 will run with a safety driver or remote safety operator ready to take control. No Einride truck will haul paid customer freight on a public road under Alpamayo control without a human safety backstop before the start of 2028. We also expect at least one more truck autonomy developer to announce work with Alpamayo by the end of Q2 2027, as Nvidia pushes the model as a shared base for the sector.
Sources