TuSimple Holdings Inc (TSP) 2021 Q4 法說會逐字稿

Good day, and welcome to the TuSimple Fourth Quarter 2021 Earnings Conference Call. (Operator Instructions) Keep in mind that this call is being recorded, and there will be a replay available at ir.tusimple.com following this call.

I would now like to turn the conference over to James Mann, Head of Investor Relations for TuSimple. Mr. Mann please go ahead.

Thank you, Latif. Good afternoon, everyone, and welcome to our fourth quarter 2021 earnings call. With us today are TuSimple's President and Chief Executive Officer, Cheng Lu; and Chief Financial Officer, Pat Dillon. Cheng and Pat will review the operating and financial highlights, and then we will take questions.

As a reminder, TuSimple's shareholder letter and a replay of this call will be available later today on our Investor Relations website. This call is being recorded. If you object in any way, please disconnect now.

Note that TuSimple shareholder letter, press release and this call all contain forward-looking statements that are subject to risks and uncertainties. These forward-looking statements are only predictions and may differ materially from actual future events or results due to a variety of factors. Please refer to the risk factors detailed in our SEC filings.

We will also discuss non-GAAP financial measures, which are not prepared in accordance with generally accepted accounting principles. Please refer to the safe harbor disclaimer and non-GAAP financial measures presented in our shareholder letter for more details, including a reconciliation of the non-GAAP to comparable GAAP measures.

I will now turn the call over to Cheng to begin. Cheng?

Thank you, James. Hello, everyone, and welcome to our fourth quarter earnings call. We have a lot of exciting developments to cover so I'll dive right in.

Since our founding in 2015, we've had a single vision of building the first-to-market, scalable, Level 4 autonomous trucking solution that will transform the freight industry. In December, we took a huge step in realizing our vision when our truck completed the world's first Driver Out semi-truck run on open public roads and 80-mile terminal-to-terminal routes. This is a massive achievement for TuSimple and a huge milestone for the industry. Since then, we have further shown that our Driver Out operations are repeatable.

To date, we have successfully completed 7 runs. The uncut videos of the 7 completed runs are all available online. And to recap, this is on a real freight route covering over 80 miles. The route starts at Union Pacific rail yard in Tucson and ends at a large distribution center in the Phoenix metro area covering both surface streets and highways. There was no human beings in the vehicle, no teleoperation. Our autonomous driving system is in complete control of the vehicle the entire time.

The system was able to smoothly handle challenging driver situations, including emergency lane vehicles, cut-ins, noncompliant drivers and construction zones. The truck is also able to enter into fail-safe operations should anything out of the ordinary happens. In fact, we did have one Driver Out run that triggered a minimum risk condition or fail-safe operations due to equipment failure in the chase vehicle. The minimal risk condition was triggered by design. The vehicle came to a control stop and further improved our safety case. We fixed equipment issue and completed 6 perfect continuous Driver Out runs afterwards.

As part of our initial Drive Out operations, we put 3 safety precautions in place, all of which we will aim to gradually remove to show that Driver Out is scalable. First, we have a chase vehicle that follows and monitors the Driver Out truck. The chase vehicle does not influence normal operations and only has the ability to put the Driver Out truck into a fail-safe mode as an extra layer of safety precaution. As we scale, the chase vehicle will be replaced by our centralized oversight system.

Second, we have a survey vehicle that runs 5 to 6 miles ahead of our Driver Out truck. Given the distance, it does not impact operations or traffic nor communicates with the Driver Out truck in any way. We expect to remove the survey vehicle from our operations later this year. Lastly, there are Arizona state and local law enforcement that observed our Driver Out operations. The police vehicles are primarily unmarked cars that stay approximately half a mile behind the Driver Out vehicle. To ensure they do not interfere with traffic, they'll occasionally exit and reenter the highway. We strongly value our collaboration and ongoing relationship with law enforcement. And going forward, we remain committed to working transparently with them and to write with regulators.

We believe, at TuSimple, it is important to have these early safety precautions as it helps to normalize driver operations and help to build trust with the general public. As I mentioned, we plan to systematically remove them as we scale up.

So what is the significance of Driver Out? First, it is a clear proof of our technology leadership. Everyone in this industry is trying to create a driverless truck. Only TuSimple has been able to accomplish repeatable, Driver Out operations on real freight routes on open public roads. In autonomous driving world, big claims are common. With this achievement, not only are we the clear industry leader, but further show our track record of hitting milestones on time.

Second, we are more confident than ever about our technology. Our previous guidance was that Driver Out program was a 1- to 2-month pilot. We're pleased to share that we expect to continue our Driver Out operations on a permanent basis. Furthermore, our Driver Out operations will start expanding, including hauling freight for Union Pacific later this spring.

Third and most importantly, Driver Out was a zero to moment. Driver Out proves we are feature complete, which means our autonomous driving system has all the capabilities required for true Driver Out operations along commercial routes, and that is able to safely mitigate or contain all the edge cases during operations.

You cannot perform driver -- perform safe continuous Driver Out operations on a commercial route without having all the capabilities developed. And so from here, we believe we have a clear pathway to commercialization in the next 2 years. From Driver Out to commercialization, the keywords are optimization and scaling.

In the freight industry, shippers and carriers care about 2 things: cost and level of service. To commercialize, it requires us to have continuous autonomous freight operations on high-volume routes at a competitive per mile operating cost relative to that of manually driven trucks. Post Driver Out, commercialization is a linear progression that we can make step by step. With our Driver Out runs as a starting point, we plan to scale the ODD and reduce autonomous operating costs, including removing all nonscalable operations. Later this year, we plan to add daytime runs followed by expansion into new routes in Texas. We also expect to remove the survey and chase vehicles.

By the end of 2023, our goal is to have a fleet of Driver Out retrofitted semi-trucks that support continuous commercial freight operations. We expect to be hauling customer freight on multiple routes in the Texas Triangle and along I-10. And we expect to have a clear line of sight on a per mile unit economics being lower than that of human-operated trucks.

To date, no one in this industry has clearly defined what is the endpoint and where are they relative to endpoint. No one can clearly articulate the steps needed to reach commercialization will show evidence that those milestones are met. At TuSimple, we are doing just that.

Over the coming quarters, you should expect to see us hit these visible milestones as we scale to commercialization. In summary for Driver Out, our technology has advanced significantly even from 6 months ago. Driver Out operations on one route is a step that any autonomous trucking technology company will need to take to reach commercialization. What made it even more challenging is the fact that we're working with prototype components and retrofitted truck. Nevertheless, we accomplished this feat, and we believe our technology is ready to start scaling now, not years down the road, but today.

This is a good segue to our hardware partnerships. We continue to have a very strong relationship with TRATON GROUP and its OEM brands, including Navistar and Scania.

Navistar is a critical partner in our Driver Out program. They were with us every step of the way, helping us to develop our retrofitted trucks, providing insights as we develop our safety case and supporting us with suppliers of critical components. We're excited to keep moving forward with Navistar as we scale our Driver Out operations.

For our U.S. production truck program with Navistar, we have made tremendous progress in the supplier selection process as well as setting a clear strategy for components where the supply base requires additional maturity. As a reminder, when we say maturity, we do not mean that hardware components require any type of advanced technology breakthroughs, but rather suppliers have to go through a process to have automotive-grade components that can support scaled production. This is critical because autonomous trucks sold to third-party customers require certification, aftermarket support and warranties.

We will continue to be proactive with the supply base to push forward the component level development schedules. We also continue to work with TRATON and Navistar to continuously pressure test our production process and time line.

In Europe, we have successfully concluded our Phase 1 pilot program with Scania. We have conducted over 40,000 kilometers of autonomous operations along Sweden's main freight corridor. We believe this is the first of its kind in Europe, and together, we have started to work on the next phase of our European partnership.

This January, we announced our most significant efforts to address the supply base and maturity. We and our long-time partner NVIDIA are developing a proprietary autonomous domain controller, or ADC. ADC is a powerful edge computing solution where all the critical processing and decisions of autonomous driving systems are made. ADC is proprietary design that incorporates NVIDIA's next-generation Orin system-on-a-chip hardware and TuSimple's autonomous driving software.

While many in the autonomous driving industry are partnering with NVIDIA, this partnership is truly unique because we're not using an off-the-shelf integrated NVIDIA hardware or software solution. We're working directly with NVIDIA to match our system requirements to the ADC specifications, including how much compute is required, the type of operating system and other onboard software and hardware requirements, the type of sensors that ADC can support and so forth.

No one knows our technology better than us. Being hands on with the ADC development and vertical integration allows us to be -- better control of our time lines and allows us to build a better product. To our knowledge, we're the first and only player to take this bold step of developing an integrated and proprietary compute solution, and we believe it will create tremendous strategic advantage as we proceed to commercialization.

Next, I'll give an update on the CFIUS process. I'm happy to share that we made significant progress with CFIUS. We believe we're in the final stages of the process, and we believe that will come to a satisfactory outcome.

With that, I'll turn the call over to Pat to discuss our financial results and guidance.

Thank you, Cheng. Beginning with our reservation program. We added 100 new truck reservations this quarter coming from 2 large trucking fleets. This brings our reservation total at quarter end to just under 7,000 trucks. After quarter end, we added 2 new reservation partners who reserved a combined 350 trucks. This brings our total reservations to 7,325 trucks as of today. Importantly, our reservation book is composed of a diverse group of major fleets and logistics players who are eager to implement our technology into their networks.

We continue to hold high standards for reservations. We only work with partners capable of implementing our AV technology and those that are ready to make a financial commitment alongside their reservation. The continued additions of high-quality reservation partners such as DHL speaks to the broad-based demand for our technology.

Now shifting gears to our financial results for the fourth quarter. We reported $2.1 million of revenue in the quarter, which increased nearly threefold versus the same period last year and represents a 15% increase quarter-over-quarter. We reported $6.3 million in revenue for full year 2021, which also represents a roughly threefold increase year-over-year. While we received new semi-trucks towards the end of 2021, our revenue growth for the year was primarily driven by increasing asset utilization.

Moving to expenses. We spent $82 million on total R&D in the quarter, including $22 million of SBC. This compares to $32 million in the same period last year. The R&D expense declined by 3% quarter-over-quarter as certain discrete items in Q3 were not repeated in Q4.

We spent $32 million on SG&A during the period, including $10 million of SBC. This compares to $10 million in the same period last year.

Please note that we have changed our financial statement presentation to combine G&A and sales and marketing into a single SG&A line.

Our net loss from operations was $116 million in the fourth quarter of 2021, and our net loss from operations was $411 million for full year 2021. Our adjusted EBITDA in the fourth quarter was negative $81 million, which compares to negative $38 million in the same period last year. For the full year 2021, our adjusted EBITDA loss was negative $279 million. The adjusted EBITDA loss reflects our sustained investment in our technology and commercial development.

We invested $1 million in the purchase of property and equipment during the quarter primarily related to equipment purchases and facility investment. We ended the quarter with a cash balance of over $1.3 billion. We remain very well capitalized to execute on our road map through 2024.

Now to provide our full year 2022 guidance. In 2022, we expect full year revenue of $9 million to $11 million. As Cheng mentioned, we are prioritizing our resources to scale up our Driver Out operations. This means focusing a significant portion of our resources on automating dense routes in the Texas Triangle area and using our own fleet of autonomous trucks. We will be less focused on adding human-operated trucks or partner fleet capacity to our AFN this year. We also do not expect meaningful revenue contribution from our China operations during 2022.

We expect a full year 2022 adjusted EBITDA loss of negative $400 million to $420 million, reflecting continued investment in our technology and commercialization. The guidance includes our recently announced ADC development with NVIDIA as well as costs related to scaling Driver Out operations.

We expect stock-based compensation expense for the year of $155 million to $175 million, reflecting our investment in our team and a philosophy of strong alignment between employees and shareholders. We expect purchases of property and equipment of $30 million to $40 million for the full year. We expect to end the year with approximately $900 million in cash on the balance sheet.

I'll now hand it back to Cheng for a few last remarks.

Thank you, Pat. 2021 was an amazing year for TuSimple. At the beginning of the year, we started our IPO journey. We knew it is going to be critical to earn the trust of our investors, and we're proud to say that we achieved all of our major goals during the first year as a public company, including the start of Driver Out.

As we enter the new year, we're also entering a new phase of our development. Our pathway to commercialization is clear. We expect to continue to scale Driver Out operations and lower autonomous driving operational costs to be the first to market with a commercially viable solution. We're proud of all we've accomplished in 2021, and we look forward to the new year.

With that, we're ready to start the Q&A session.

(Operator Instructions) Our first question comes from the line Ravi Shanker of Morgan Stanley.

Cheng, obviously, 2021 is a big year for you with the IPO and Driver Out, but I'm just wondering kind of how we can track your progress with milestones through 2022. Is there anything -- obviously, you've given us guidance, but guidance is kind of meaningless given where you are in the stage of your development right now. So anything in particular in terms of miles mapped or number of Driver Out runs or specific milestones we can use to track your progress to 2020 would be great.

That's a great question, Ravi. Well, we do expect to give more detailed milestone guidance in the coming quarters, especially during Analyst Day. But I think the way to think about it is, what is the finish line or what is commercialization. And as I mentioned earlier, commercialization means that we can scale Drive Out operations, to have continuous Driver Out freight operations in a -- on dense routes, so Texas Triangle, the I-10. And more importantly, the cost per mile of operations has to be lower or clear line of sight lower than that of a human driver, right? That's how, ultimately, we provide a service or a benefit to the shippers and carrier customers.

And so I also mentioned that from Driver Out as a starting point, when you think about -- I wouldn't think so much about sort of metrics or KPIs, but really the milestones that show us getting closer to that commercialization goal. So we'll show you, for instance, an additional Driver Out routes from nighttime driving to daytime driving. We'll intend to show you the cost of per mile operations and how that's going to scale over time. So of course, today, we do have some safety precautions that we talked about.

As we remove those things, the cost will come down. As we optimize our features, optimize data usage, costs will come down. So those are things that we do intend to lay out in a much more easy to understand way. But at a very high level, today from Driver Out, there really is a clear path. There's things that we can check off one by one to show you. And more importantly, we can show you where we are on that path to commercialization.

So hopefully, that answers your question a little bit better, but more to come, but really think of this as where is it we're trying to go? Because even if we give you numbers, like, for instance, number of miles, right, where disengagements would -- comes only before, what does that really mean? How do you know how far we are to commercialization, right? And that's, I think, the problem the industry has today. And I think what we're saying now and what we're laying down in terms of our path to commercialization is a much better way to frame where we are and where we need to go.

Great. That does make a lot of sense. And maybe as a follow-up, clearly, it sounds like your 7 Driver Out runs were successful and kind of no nasty surprises or anything. But can you just share with us if there were any kind of manageable surprises or any learnings from these 7 runs that you've kind of changed the way you either do the runs or tweak your algorithm or change your hardware, kind of any specific incremental learnings from any of these 7 runs?

Yes. I mean there's a lot of learnings we have from the runs. So a lot of them are built in. I mean I wouldn't call them surprises, but really the next set of developments, optimization improvements that we're going to do. And as I mentioned, we have a clear path to commercialization. So from hereon, there are these clear milestones that we intend to achieve over the coming quarters.

I think the biggest takeaway for us was we felt actually even more confident and really huge credit to our technology team, our operations team, but we feel so confident that the previous guidance we gave was Driver Out would be sort of 1 to 2 months pilot program. And now we intend to scale this and to make this a permanent part of our operations. So that -- I think that's something that really is a very positive development for TuSimple.

Our next question comes from Brian Ossenbeck of JPMorgan.

Cheng, just to follow up on that, I think path to commercialization, basically, so that you're ready to scale now. So I just wanted to make sure we understood what that was in the context of it. It sounds like -- correct me if I'm wrong, it sounds like you're scaling into further Driver Out demonstrations in certain ODDs in certain areas, at least here in the near term.

But maybe putting that aside, is there anything left technically that you feel like needs to be handled and figured out and tested before you were able to really scale? So I just want to really understand the difference between the scaling you're working on now and just kind of the broader end game as it were.

Yes. Yes. They're scaling to commercialization, right, which ultimately is critical. No one's been able to clearly articulate what is commercialization and how we get there. So let's go back to kind of what Ravi asked and we can give you a lot of metrics, but really, how does that tell you where we are, right?

And so we planted a flagpole of what does it mean to commercialize. In the freight industry, you know very well, it's cost and level of service or reliability, uptime, right? So we have to have continuous operations with our autonomous trucks. We have to have them in high-density routes. And we have to be able to show a pathway on a per mile cost that's lower than a human-operated truck. Otherwise, why would shippers or carriers use this technology.

So when we say scaling, and going back to your question about what is left to develop technically, to be able to demonstrate Driver Out operations repeatedly, we do believe that we are feature complete. And what we define future complete is it means that we have all the capabilities of operating a vehicle from terminal to terminal on a commercial route.

If you look at most commercial routes, they're quite the same. And so for instance, we can't do Driver Out operations if our truck isn't able to handle emergency lane vehicles, right? I mean who's going to take over when you run into an emergency lane vehicle. I think in our 550 miles of Driver Out runs, there were 36 emergency lane vehicles. Your truck has to be able to be defensive driving and have -- be able to adhere to noncompliant drivers. And so those are additional features we built in.

So really, today, we are feature complete. We're also able to mitigate. I think what gives comfort is being able to mitigate or contain all the edge cases that you can counter on a route. So from that standpoint, that is already a significant technology development.

And when we say scaling, it's -- we believe that is more of a linear progression, as we mentioned earlier in the prepared remarks. We can show you that systematically. We can show you that from step by step. We don't see that as any sort of technical hurdles, but these are -- it takes capital, it takes manpower and it takes engineering work, but we'll be able to show these things, too.

Okay. As a follow-up, you mentioned the CFIUS process. They didn't give too much detail behind. I don't know if there's anything else you can share in terms of what, I guess, what is an acceptable outcome and maybe something on the timing.

And then I think Pat mentioned that your China operations aren't really going to have too much of a contribution here, in 2022 at least. So maybe just an update on where that stands, if anything has changed on the strategy or how you see that operation going forward.

On CFIUS, we want to say a lot, but we -- unfortunately, we cannot at this time. But just to reiterate, our belief is that we'll come to a satisfactory outcome for TuSimple, and we're near the final stages. So unfortunately, we cannot talk more about what we do. Of course, we'll be able to share and we'll have to file an 8-K around the details.

In terms of China, as Pat mentioned, we do not see China having meaningful revenue contribution this year. That doesn't mean that our China strategy isn't sound. We do believe that our strategy in China is sound. We've always said that China has regulatory risk, just given that today, autonomous trucking for pure Level 4 is still permit by permit, sort of lane by lane.

We have what we believe the best project to commercialize in near term in China, which is the Shanghai deepwater seaport. And at TuSimple, we're always pressure testing and optimizing our resources to make sure we have the highest ROI and value to our shareholders, right?

So today, the U.S. will be first commercialized. But at the same time, we want to ensure that we have upside to our shareholders of when China does take off. I mean after all, it's the second largest framework in the world that is growing very fast.

Our next question comes from Raji Gill of Needham & Company.

Yes. congratulations on the Driver Out program. That's a great validation of the technology. So Pat, I just wanted to talk a little bit about the fiscal year '22 guidance of $9 million to $11 million. You mentioned that you're prioritizing resources for your Driver Out programs, focusing it on those instead of kind of partnership capacity to AFN. Wondering what -- if you could elaborate a little bit further on that decision to prioritize the Driver Out program. And within that context, when we're thinking about going from $6 million to $10 million, what are the revenue growth drivers to get it from $6 million to $10 million?

Great. Thanks, Raji. So when you think about what our revenue operations, what the strategic goals are, it's really around, first, supporting our technology development in real-world freight conditions and then also to develop and maintain our long-term commercial partners with partnerships with customers who are going to implement our technology over the long term.

So for us, the -- as you said, there are different components of it. Where we see the highest ROI is really around the AV operations, both scaling the Driver Out operations as well as continuing to scale AV operations with a safety driver behind the wheel. This is actually what our customers have told us as well. This is where they want us to focus. They want to be able to participate in the technology and have it all-free for them using AV technology. And that is a bigger priority versus more manual freight.

We do retain a lot of strategic flexibility to add either more of our own human-operated trucks or more partner freight. And we will -- whenever the time is right or whenever it's strategically advantageous, we can pull some of those levers to do so. But trying to be as capital efficient and focus on what's most important to drive value. We think it's really around the AV operations today. And our customers have validated that, too, that, that's where they want us to be focusing our time and making sure that we're getting them more AV miles as opposed to more manual miles hauling freight for them.

The guide from kind of $6 million to $10 million at the midpoint, it's really a function of the number of trucks that we have in our commercial operations and the utilization of those trucks. So we did add more trucks at the end of the year here. You'll see in our letter, we have approximately 100 trucks globally. About 75 of those are in the U.S., and roughly half of those trucks are dedicated to commercial operation.

So we're adding more trucks to the count, and we're increasing the utilization of those trucks, notwithstanding some of the challenging market conditions around hiring drivers and dealing with some of the pandemic-related outages where we've had to have drivers who have taken time off due to COVID restrictions. So hopefully, that gives you a little bit of the building blocks around how we go from $6 million to $10 million.

Yes. That's really helpful. And for my follow-up, when you're thinking about over the next 2 to 3 years and kind of getting to that kind of $1 billion number when you're kind of fully commercializing these trucks, these autonomous trucks, I know, Cheng, you had mentioned that the Driver Out programs that you're very dedicated in expanding in Texas and elsewhere are going to be major kind of proof points to validate that you guys can scale and commercialize Driver Out technology.

Wondering what you've completed in December, what you intend to complete this year and your focus on that, how that will help, either potentially accelerate the transition to your -- to that 2-year kind of commercialization. Will it reinforce it? I'm curious to see how you're thinking about the Driver Out programs now and when you're thinking about it during the IPO process or prior to that and kind of bridging the gap there. Hopefully, that makes sense.

Yes. No, good question. For us, after Driver Out completion, again, we did this not because we want to hit a time line. We did this because we felt it was safe and we're ready. We truly believe that more confidence that we can actually pull forward commercialization despite not having serious production trucks. We're having integrated trucks on the market yet.

And again, when you define commercialization, there's kind of main definitions of it, right? Is it -- are you selling it to third party? Are you -- how many are you selling -- how many trucks you need to be commercialized?

Well, for us, commercialization ultimately is the ability to provide a reliable autonomous freight service to shippers. And from commercialization to series production is something that we'll continue to do with our partners TRATON, Navistar, Scania. And that's something that requires production-ready parts, warranty, services, certification. That's kind of -- that's going to continue. And that's how you get even more trucks onto our network, right, tens of thousands.

So at a very high level, really, it hasn't changed at all. I think if anything, we pulled forward. We believe we'll commercialize and be able to show you, and now we're actually giving you more detail about the road map to commercialize over the next 2 years. And we'll continue to give you more detail over the coming months. So please stay tuned.

Got it. If I could just squeeze in one more question. You mentioned that you're developing a proprietary ADC custom chip with NVIDIA and kind of working hand-in-hand with that with NVIDIA. Can you discuss the rationale of working closely with NVIDIA, making sure that your software can run on their Orin SoC processor? What was the rationale for that? And how do you feel that differentiates your technology, say, versus other self-driving technologies or other approaches to the market?

Sure. Maybe to give some examples or situations that we're facing. So today, we look at the markets, there's not a readily available or even a clear direction -- a clear path of a product that has, ADC product, that has the amount of computational power that our system requires. And so that's one of the benefits of being vertically integrated and working with NVIDIA. We can have a design that has the right number of Orin that allows us to have the enough computational power to operate our virtual driver software application layer.

We have to ensure that the ADC can take in all the sensor inputs that support our autonomous driving system. The ADC is not just our software and their hardware, but also other software on chips on ADC. And so what are those, what are the requirements, does that fit with our system or not? Those are designs that we now have ability to make rather than if you have this being built by, for instance, a Tier 1 supplier. Of course, their goal is to have scale, right? They're taking requirements from everyone. One small change will take a long time.

And so that delays when ADC comes to market and also creates a lot of uncertainty whether or not this component or this ADC can meet the requirements of our autonomous driving software. So for those reasons, this ADC is really geared towards ensuring that we have serious production as soon as possible. And I think it was a step that we felt that we needed to take.

Our next question comes from Dan Levy of Credit Suisse.

First, I want to ask about just the capital requirements over the next couple of years. Pat, I think you made a comment that the IPO proceeds are enough to carry you to commercialization in 2024. Maybe you can just unpack that. And has the Driver Out pilot changed your view on the cost or the capital outlay required to reach commercialization? Maybe you could just give us a flavor for as you continue to work up, just how much more capital-intense this gets if this is any different from what you previously expected?

No. Good question, Dan. I think there's no material deviation from what we've been saying our capital deployment plan for the next 3 years. So we feel very fortunate to have the kind of balance sheet that we have where we are fully capitalized on our base plan. And we feel very confident in our capital efficiency and our ability to get through to 2024 commercialization with the capital that we have on the balance sheet today.

Okay. So capital is not a constraint for you in your path to commercialization?

No. No. I mean we're very focused on being capital-efficient, but I think you should interpret from our remarks that we are proceeding full speed ahead on the development of the technology and commercialization and very confident that the balance sheet we have today will get us through to commercialization in 2024.

Got it. And then I'd like to follow up on some of the prior questions on Driver Out, if we could just maybe focus on this one Tucson-Phoenix route and give us a flavor -- I know you said you're going to give us more details in the future, but give us a flavor of what we can extrapolate to future commercialization. And more specifically, between where you are today to what you would ultimately need for commercialization showing that it's continuous, that it's lower cost, what would be the remaining items that you would need to complete on this route to show that, hey, like a route like this, which you said technologically is already feature complete that that's ready for commercialization. I know the vehicle plays into it, but maybe you could unpack what we could extrapolate from that to commercialization.

Yes. Great question. So commercialization is it's uptime level of service for the shippers and cost. So level of service, right, we can't only be hauling freight at nighttime, although nighttime is where freight gets hauled and moved quite the most, but -- so daytime. So for instance, that's something that we can demonstrate or demonstrate next, right?

We also need to have more continuous operations, and that's a function of sort of actually hardened hardware, so better hardware, as well as optimization of the operations so that the turnaround time is fast. I mean we demonstrated the ability to do back-to-back operations for Driver Out runs on the same night. And of course, we have to increase the number of those.

In terms of cost, we are technically feature complete. There are these safety precautions that we put in place. So for instance, a survey vehicle. I mean we can turn that survey vehicle. If you have 30 trucks on one route, those 30 trucks all become survey vehicles of each other, right? So that's something that naturally will go away, and we can show that to you over the next coming quarters. We had a chase vehicle. We have to remove that as part of our centralized oversight system. So that's in the works now, but that's not something that, again, technically we cannot achieve.

So if you think along the framework of scaling means expanding the operational design domain, so daytime, nighttime, new routes, uptime. And then also, cost, right? Because everything comes down to cost. So if you can get from point A point B on a per mile basis and have a clear line of sight that is lower than human-driven truck, then that becomes a real service, a real product. So -- and that cost is these operational items, these nonscalable operational items and kind of anything associated with that. So -- that really comes to optimizing the technology, the features more and then doing away with some of these nonscalable items.

But I mean, today, if you think about it, we have optimized one route, optimized -- automated one route. And this is truly a commercial route. And this route alone can generate significant revenue. And actually, in a way, for trucking, density is our friend. Actually, the more dense routes because we put more trucks on them, they'll update the maps faster. That becomes a fleet of survey vehicles for each other, but that becomes our friend. And so that allows us to really commercialize.

Our next question comes from Alex Potter of Piper Sandler.

Great. I have 2 hardware-related questions. So the first one, I think I can understand the strategic rationale and the benefits regarding, I guess, pace, speed to market with regard to vertical integration and the partnership with NVIDIA.

Are there any other areas, hardware specifically, where you're seeing bottlenecks in the supply chain or things that you just can't buy off the shelf that you feel like you need to take, I guess, more -- devote more time and energy to in-house designs? Or is that just on the controller? And other than that, you can more or less farm everything else out.

Yes. Good question. It's primarily controller. For other sensors and components, we do see either a clear line of sight or actually many, many suppliers working towards that problem like, for instance, LiDAR. I mean, there's plenty of well-capitalized LiDAR companies now that are working towards -- between start-ups and Tier 1s that are working towards building production-ready LiDAR solutions, right? So we feel good about that even though today, it's not something that's readily available in terms of production solution.

Cameras and radars and GPS units, those are pretty easy. And in the braking and steering components, the large Tier 1s are working on that. There's a clear road map on when the next-generation components are coming out, and we feel good about that, right? It's not something that hasn't been developed before. It's just an iteration of a generation of new hardware. So we feel generally pretty good about those things.

So yes, no, just to summarize, I mean, I think the compute is something that's unique to Level 4 driving because only we need something that's powerful and also has to be quite tailor-made to TuSimple solution just because this application layer sits in. So really, that's the main one that we feel vertical integration is most important.

Okay. Great. Very helpful. And then the last question is just on, I guess, the timing of finalization for the bill of materials. I know that this compute platform and a lot of these other components that we're talking about, next-gen versions coming available, all of that will play into the ultimate timing. But how long do you think it will be before you can set that design in stone and send it over to Navistar?

Well, we -- I mean we're continuing to evaluate the supply base. What we gave guidance was that we'll look to set the bill of materials in spring of 2022, and so that's part of our production process. I think we'll definitely have more details in the coming months.

And I think 2 things to take away. One is, right now, working with non-production-ready components actually puts more onerous on the software, right? So we're actually, obviously, doing more work than more development than we might need to otherwise. So I think the path actually gets easier from here, not harder. I think that's point number one.

And then point number two is that we're proactively taking steps to address the supply chain maturity where we're not making -- we're not using this as excuse, but we're really taking steps to address it, of course, the ADC being one of the proof points.

Our next question comes from Joseph Spak of RBC Capital Markets.

I guess one near term on the guidance and one longer term. The '22 revenue guidance, I mean, it's basically at the run rate you're currently at, but you mentioned you plan on adding trucks. I think you said you have 75 in the U.S., but it sounds like maybe only half of those are running. So maybe you could sort of help us with that a little bit and what you expect the trucks running to be at by the end of the year.

And then on the EBITDA loss, it's about 25%, 26% worse than the current run rate. And I heard you talk about some of the investment with NVIDIA, and I know you're hiring engineers, but maybe you could help us a little bit breaking down that additional cost for next year.

Yes, sure. So we will continue to see the revenue run rate on a quarterly basis increase throughout the course of 2022. So the trucks that we mentioned that are coming in, some of which are being upfitted for AV including the commercial operations trucks, those will be placed into service and start to generate more revenue every quarter. So we continue to see a steady ramp in the revenue base throughout 2022.

With respect to the EBITDA loss, there's a couple of things at play here. One is just full year impact of some of the hiring that we made throughout the course of 2021. 2021 was a year where we did add significantly to our team. And as a reminder, over half of our cash costs is really around compensation for our team. So you're starting to see a lapping effect of the full year impact of some of those hires.

We'll continue to add new hires to the team this year. I think '21 was probably -- when you think about the rate of growth, and the absolute number of hires for the year was probably more significant than what the hiring will be for 2022 because we have hit a critical mass. We'll continue to add and -- selectively where we need to add to the team. And then the items that you mentioned, like the ADC project as well as just concentrating our efforts around expanding the Driver Out operations so that we can scale those to commercialization through the year.

Okay. In the letter, you made mention of some of the mapping team enhancements and the technology and getting maps to update quicker and be more robust as you get ready for commercialization. I mean without, I guess, getting too technical, like can you help us understand what that means? Are you automating that process so that you have the ability to scale and map routes quicker than prior? Or how does that sort of really play into the story here?

Yes. I think when you think about where the mapping technology is going, it is -- to some extent about mapping new routes faster, although that we do that very quickly already. Really where a lot of our technology is going is the updating of the maps so that every time one of our trucks travels a particular route, if there's a new element that we're able to almost instantaneously update the map and send the updates to the map to any truck that's following on that same route behind.

So as you mentioned, or as Cheng mentioned earlier, we're going to be removing the survey vehicle from our Driver-Out operations this year and having the quicker map updates so that if there is a pop-up construction zone or some other element of the road which was previously mapped has changed, that we can efficiently push that update to any truck that comes behind it for the next route.

So it's really critical to making the technology and our Driver Out operations more scalable. And that's where you've heard us talk for the several quarters now about some of our investments in mapping, and it's really starting to pay off in terms of making this scalable and a critical piece of getting to commercialization.

Our next question comes from Colin Rusch of Oppenheimer.

Can you talk about your progress in terms of integrating or potentially replacing some of the existing logistics and workflow solutions that are out there?

Could you be a little bit more specific or...

Yes. I mean I'm just looking at -- with these fleets that already have existing solutions around managing logistics and workflow and as you move towards commercialization, it's not just around the hardware and the functionality of the vehicles, but really having your system integrate with the existing programs that you're going to be kind of integrating into with some of these clients. I just want to get a sense of how far along you guys are in that process.

That's a good question. I think at TuSimple, we're the first, and we've always said that ultimately, what we're developing is a solution that has integrated -- a holistic solution that integrates into our partners' supply chain, right, in order to make it more efficient for them. And so they can provide a better end-to-end logistics solution because ultimately, that's where everybody wants to go is providing the best end-to-end logistics solution for their customers.

And the integration of our system into their TMS is a process. That is why we have reservation partners today because really, it's not just collecting truck count, but starting that integration process.

If you reflect on 2021, we have over 160,000 miles, actually now more than that, of autonomous miles driven with UPS. So again, we'll be getting deep into their supply chain and integrating with their service. We're starting to haul freight for New Pacific. How do we integrate with the railroad and help improve or expand their service? We have great partners on traditional truckload carriers. We have partners, of course, parcel with UPS and DHL.

So to really answer your question, that's absolutely what we're doing today. And that's where we see the highest ROI in terms of our resources is ensuring that we can have more of our trucks working with these strategic partners to go deeper with them to integrate with them. And I think that also creates barriers to entry for latecomers.

That's very helpful. And then the second question is really around the significance of the Driver Out testing for your learning cycles and the cadence of those learning cycles. Obviously it's still early days, but -- and you got your plan laid out for us. But I'm just wondering in these early runs, if you're starting to get a sense of just how accelerated the learning cycle is for you and some of the things that you're starting to see that you're going to have to solve, but also bringing this to market and the pace of that for the organization.

Yes. And I think you hit the word, the significance of the Driver Out. And I think what's important for us to keep saying is that Driver Out is not a demo. It's not a shiny object that happen once. And Driver Out is significant because we can now show you from Driver Out what is the linear road map to getting to commercialization. And that's something that no other autonomous trucking -- driving company can do today. Where is the end and where are we relative to that end? Right?

So that beats any sort of metrics that we can share with you, right, number of miles driven and number of trucks. I mean all these things, again, are nice metrics, but how do you know where the endpoint -- how do you know where you are relative to the endpoint. And that's something that I think, today, we have a marker of continuous Driver Out operations on one route, and this is where we need to go in terms of commercialization.

So I think in terms of significance, too, I mean, it took us 6 years as a company to get to Driver Out operations on one route, so if we scale now, of course, the time -- the development cycle gets shorter and shorter, right? I mean, ultimately, what is it we're trying to do? We're trying to automate more lanes and deliver freight -- enable freight capacity to our shippers, right? So as we show you that we can automate more lanes faster than 6 years, it definitely will be. As we show you that we can expand the operational design domain, as we show you that the cost on a per mile basis will come down, I mean all those things should give confidence that development cycle is being shorter and that we're closer to commercialization at a reasonable scale.

Our next question comes from George Gianarikas of Baird.

I guess you mentioned on the call and in your prepared remarks that the development road map is linear, and you hired a 10% sequential increase in R&D personnel. Can you just sort of explain to us what the over 1,000 people in R&D are working on incrementally when you have the NVIDIA co-development? But what are the incremental problems that your R&D team is working on over the next couple of years?

Well, I think this goes back to -- I mean there's still -- this is a very long answer, but it still goes back to -- there's still a lot of engineering work. If you take -- start from Driver Out to optimize the features which we have completed to do daytime operations to do a new route in Texas Triangle, to do more than one route in Texas Triangle. There's engineering work to remove -- to update -- to have mapping as a more live update so that every truck, can they become a survey vehicle for each other, right? And as you have more trucks on the same routes that's hauling freight, then actually that makes it safer for the whole route that we're automating.

So there's still a ton of work. And of course, there's work on the hardware side of production vehicles with TRATON, Navistar. There's work that we're doing with different suppliers in terms of partnerships or working with them and requirements and testing those components. So really, the list can go on. But I think when we say linear, it's that -- again, going back to Driver Out's significant because from here, we can show you what are the steps required to get to commercialization. And as we check each one off, that should give you more confidence that we're closer and closer to commercialization, right?

And again, today, we look at the whole industry, how do you get confidence that people are moving closer to commercialization? How do you get confidence of where is commercialization, where people are relative to that goal? And I think that's really the significance of this Driver Out operations.

Got it. And one more on your patents. You guys have done a good job of every quarter outlining your total patents issued and the sequential growth in those. How core are those to your long-term competitive advantage relative to the other companies in your industry?

Yes, it's going to be critical. And I mean this is not something that's new to autonomous driving. It's -- if you look across the mobile sort of telephone, right, cell phones, telecom, electric vehicles, I mean, you get to a certain point of maturity where you do have patent infringements, you have patent -- I mean you have companies more [rollout] being sold off for the patents that they own, right? So today, it's not something that shows up in terms of our -- the competitive advantage. But you fast forward 3 years, 5 years, it will definitely be a huge advantage for TuSimple.

Our next question comes from Todd Fowler of KeyBanc Capital.

Great. so I wanted to ask on the reservations, and I know that these can be a bit lumpy, but it seems like that there was a nice couple of orders that came in early this year. Cheng, do you think that, that's a reflection of the success of the Driver Out runs that you've had? And can you maybe speak to the interest level that you've seen post Driver Out at this point?

Yes. I mean absolutely, because the Driver Out is significant, it's a real proof point of where we are to commercialization. It's also interesting because now you have -- now if you start thinking about what does autonomous trucks can do for the industry, right? You have players that historically are in trucking now get into trucking. Potentially, you have players that can leverage autonomy to expand their business. So now I think it also creates a little bit of a competitive dynamic among different types of carriers and shippers. All those, and of course, lastly, is just the supply chain. I mean the driver shortage is getting significantly worse, right? You're seeing labor costs going higher. You're seeing fuel costs rising really fast. So I think those confluence of factors are leading to interest. I mean the interest is always there, but certainly we're seeing even more interest now.

Yes. No, that certainly makes sense. And I guess just for my follow-up, Pat, I'm not sure if you want to share any color around the potential cadence of EBITDA throughout the year throughout '22. And I guess what I'm really kind of curious about as we exit '22, was your expectation that the EBITDA losses would start to narrow as we exit the year? Or is it pretty consistent with an even run rate with your guidance?

Yes. I think you'll see some modest increase in the EBITDA losses as we progress through the quarters in 2022. That's because we will continue to hire folks, and that will increase the labor base that drives a meaningful percentage of the EBITDA loss. But I do think that what you'll see is consistent with 2021 being a big hiring year and 2022 being more selective to fill out specific spots on the team. You'll see a deceleration in the losses. They'll continue to grow. But the second derivative will be, I think, indicative of the fact that we are hitting a critical mass on this and pushing forward towards these final stages of development over the next few years as we get towards commercialization.

Our next question comes from Ken Hoexter of Bank of America.

Congrats on the Driver Out ramp. Can you talk about the scaling, though? I mean you mentioned the number of trucks. You've got 100 trucks in the fleet globally. But last quarter, you talked about Navistar having delays. Can you talk about where that stands now, what we should expect to come online? And it sounds like are you putting them all towards the commercial development? Is there still -- are you still scaling up the driver-filled seats? I just want to understand the size and scale we should look for.

I think you'll -- we'll probably have roughly still this 50-50 split in the U.S. between trucks that are dedicated entirely to testing and technology development and the other half hauling freight commercially. I would say the penetration of autonomy within our commercial fleet is continuing to increase, which is important. That's what our customers are really demanding is freight capacity, but increasingly, they want more autonomous freight capacity so that they can participate in the technology development.

So the penetration is continuing to increase, including being able to do Driver Out operations commercially, which is what we announced around our Union Pacific partnership, which will start the spring hauling freight for them on a Driver Out basis. So you'll see more trucks coming online over the course of the first part of this year, and that will show up in the cadence of the revenue ramp throughout the year.

And is there a relation to the delay last year in terms of getting the physical equipment? Can you update on timing that the trucks have arrived?

Yes. The trucks have arrived, roughly 20-plus new trucks that have arrived. Some of those are going through the upfit process right now to get fitted for AV equipment. Some are being placed into service for both commercial and testing operations. So we're starting to see that relief.

Hiring drivers is still challenging. I think, as you know, from the ecosystem, this continues to be extremely challenging. But we're able to -- I think we've been able to be successful in bringing in drivers and making sure that all of our trucks are ceded. But it is certainly a tight labor market, which obviously makes it difficult in the near term for us, but just further underscores the need for our technology over the long term.

Our next question comes from Scott Group of Wolfe Research.

I apologize if I missed this. Is there any Driver Out plan for a new route or a different route this year?

Scott, we haven't shared the exact details, but that is something that we expect to give more guidance in the coming months.

Okay. And then just bigger picture, if I think about 2025 or so, how have your assumptions evolved from a year ago when you think about number of trucks, miles per truck, rate per mile? Where are you getting more optimistic about 2025, where are you getting less optimistic about 2025? And if you have a different year you want to talk about, that's fine.

Yes. I mean I think, directionally, the rate from -- it looks like it's going up. And I can't predict the future, but I think trends are going higher. I think in terms of number of trucks -- that is our series production. As we said, we have to have integrated trucks with our OEM partners in order to get to thousands and tens of thousands.

One of the risks that we have identified more and more of over the last year is the supply base maturity. Of course, we're taking active steps to address that. We feel confident about our strategy. So that's something that we continue to pressure test the time line. And we'll have more updates on that as our production program with Navistar naturally runs its course.

Well, no, I mean, I think if anything, again, as we mentioned in the earlier call, we are pulling forward commercialization I think end of 2023 in terms of actually be able to have a continuous autonomous freight service for customers on dense routes. That's something that we actually feel more even more confident of today compared to a year ago.

Our next question comes from Jeff Osborne of Cowen and Company.

Great. I just had one question. I was curious if you could touch on, Cheng, some of the things that you can't control. You've done a great job on the call talking about the things you can control, but in particular, insurance. Any update on the Liberty Mutual partnership and any new developments on the regulatory side that we should be aware of would be great to hear.

No. On the regulatory side, there hasn't been material changes, nothing, I think, to signal a change in regulation. I think actually the continued rhetoric messaging from the federal and state governments have been positive for autonomous trucking, especially given how supply chain is today and the driver shortage getting, I think, worse and worse. I think, obviously, the supply chain is something that we need to be more in control.

No. I mean, I think, generally speaking, we feel like given the team that we have, given the capital that we have, the partners that we have and what we are in technology development, really, the fact that we could do Driver Out continuously means that things are in our control, right? And this is not a -- I said a year ago, this is not about tech stack, right? This is not a zero to one, hey, maybe it works, maybe doesn't work. This is -- today, it works, and now we're scaling to commercialization through expanding ODD and optimizing the cost.

This does conclude today's conference call. Thank you for participating. You may now disconnect.

Thank you, everyone.