Volvo is at an inflection point . The company has finally launch the new Volvo EX90 — it ’s long - awaited flagship electric SUV , design to take on other premium large SUVs like theRivian R1S. The unexampled vehicle offers Volvo ’s signature Norse style , but with modern feature article and Google ’s Android Automotive software program .
Beyond being its first major electric SUV release , perhaps more important is the fact thatthe EX90also represents the launch of Volvo ’s so - call Superset platform . Essentially , Superset is the Volvo - contrive tech stack that is supposed to be used onallof its upcoming galvanising vehicles , containing all the modules and software program necessary to power a build - out next - generation lineup of EVs .
In other Word , the technical school underlying the EX90 will be recycle for its forthcoming luxury sedan , the ES90 , along with crummy vehicles like the EX60 and whatever else it has coming down the pipeline .
Volvo has some bad contest out front of it . While the companionship is on the face of it beating its legacy automaker competitors to so - called software - define vehicle , it also has to compete with newcomers like Rivian , which has been hailed for its excellent software program and modern selection of features .
At CES 2025 , I had the opportunity to get behind the wheel of an EX90 with Volvo ’s CTO Anders Bell , where we chatted about the Superset platform , Volvo ’s approach to software , and more .
Digital Trends : get ’s start with software - defined vehicles ( SDVs ) . Everyone says that they ’re build SDVs . How do you guys delimitate an SDV ?
Anders Bell : So it ’s a really good interrogative because there is no industriousness definition of it . I think I ’ve ascertain some pretty adequate effort online to [ define it ] standardised ADAS — that ’s level 1 to 5 . There ’s an SDV level 1 to 5 , right ? But for us , it ’s because we already do over - the updates to the current fleet . We ’re going to do 2.5 million cable car , uplifting the whole UI experience on 2.5 million cars on the road globally this twelvemonth . That ’s what we ’re doing with the live technology radix . We can already do pretty deep OTAs . The affair is , this car is designed from the ground up . It ’s born digital , so to speak . So it ’s a prostrate connection topology and a distinctive architecture base around a centre information processing system , which in this case is Nvidia Orin NX , with a user experience computing machine , which is aQualcomm Snapdragon .
One of the reason we ’re working with Qualcomm is that we also do Android Automotive OS and full - out Google Automotive Services , and of grade Qualcomm and Google are super tight . So that ’s a good mates because that means we can also abide on the bleeding edge of the [ Google Automotive Services ] clobber .
When you reckon at Jensen [ Huang ] talking about all the car makers using Nvidia Silicon … that ’s good . Most of them , I think everybody but a few of us , utilise the Nvidia centre estimator as a domain - base ADAS calculator . It only does ADAS . In our case , it ’s a inwardness computer . It ’s the central art object of orchestrate all of the vehicle function , let in ADAS .
Digital Trends : Are there any craft - offs to that ?
Anders Bell : Yes and no . As Jensen mentioned yesterday as well , we ’re ISO 26262 compliant , so we have all the operative safety requirement . Of course , it ’s more complicated to reach that — the separation of headache , etc . — when you ’re fulfil more clobber in one environment . But that ’s just mostly work . For us , it ’s over time efficient , so we can carry through and create . It ’s all about create an assailable web of the gondola compare to the old kind of motorcar where it ’s a system of organization with very limited interaction .
In the software - defined vehicle , the camera in the ADAS system is a camera . Its elementary reason to be is ADAS , stay in lane and run into the motorcar - brake . That ’s why we put the camera there , but now it ’s a photographic camera . you’re able to use it as a bolt cam , you’re able to do watch style , you’re able to programme it to take a snapshot every fourth dimension you wink left , if you would like to . I do n’t know why you , but you could . I recollect this is where the car really becomes much more like a electronic computer on wheels , where every single piece of ironware is accessible and exposed to innovation and cross - fertilization . The issue of combination of use you could do is jolly much dateless .
So , the view on the machine becomes all dissimilar . The elevator car is a hardware merchandise , include compute , detection and propulsion — the power to move stuff . It needs to be appear at like it has capabilities to be unlock over time with software package and also combined in new manner which we have n’t thought about before . Which is not potential to do with the domain - based auto .
So , software - defined vehicles , back to your original question … I ’m certain some auto companies say , “ Yeah , this is fully software - defined , ” which stand for they can do OTAs . Sometimes only for the infotainment system . Sometimes much deeper . But for me , the package - defined vehicle , it ’s full deepness . Like , every node — there ’s no kind of black boxwood ECUs in the organisation . you’re able to get to every actuator and every sensor . The capabilities are exposed across through APIs . So , you may pretty much do anything you would like to with software . So , then you have all the complexity with safe , separation of concerns , etc . , that needs to be in place . But for me , that ’s the software package - define fomite , where it ’s a in full mix organism .
Digital Trends : When you say all the actuator are exposed through genus Apis , is that just to Volvo or to developers ?
Anders Bell : No , ab initio to us . Then , of form , over time , a certain scene we are happy to deal with third parties , etc . It ’s how the counseling is going to grow . It ’s right now for us to focus really on create that infrastructure inside the car where we have the exposed genus Apis . Because what we make headway from that ourselves is development focal ratio . Now we can upgrade modules , upgrade the next genesis drive units , next generation screen , next contemporaries user experience computers . And we can do that very expeditiously , and launch with minimal splash into yield because we have standardized APIs . The software - defined car is absolutely not a inactive physical object . It involve to continuously improve every single aspect , all the mechanically skillful and electric ingredient need to be continuously better . So repel units , battery direction system of rules , charging algorithms , seat , instrument panels , everything . Screens will become better and good . We will raise the screens over sentence to keep the whole technical school stack militant , but also not let complexity grow out of control condition . So it ’s updating both the cars in production and the ace on the draught boards the same Modern hardware .
Digital Trends : You ’re utter about suffer a batch of ascendancy over every aspect of the fomite . How do you recollect about not have so much mastery over the software by using Android Automotive ?
Anders Bell : I opine Android Automotive is one aspect , right ? We ’re extremely felicitous with our coaction with Google . It ’s always about opt your partners . I think this fall with a new territory as well . As a car shaper , you involve to choose strong partner to work and co - develop with because you ca n’t do everything yourself . Everything would be kind of mediocre . So we ’re exceedingly happy with the collaboration with Google and the roadmap that they have and that we have together with them .
Some things our key partners do so much better than us , and we need to be part of that and bring home the bacon that to our customers , to provide the best possible in - railway car user experience . At the same prison term , of course , we will have Apple wirelessCarPlayfor the customers who require that . It ’s the customer that needs to choose their own ecosystems . We ca n’t really order to them what they should use . We need to integrate into the customer ’s lives and not other fashion around .
Digital Trends : It ’s refreshing to get word a bequest auto maker admit that they ca n’t do everything themselves because a pot of them are hear and failing .
Anders Bell : Yeah , I mean , we need to be realistic , right ? I mean , appear at this car . It ’s marvellous . We ’ve done a prima job . It ’s quiet . It ’s the dependable Volvo we ’ve ever design . Beautiful drive — I mean , you’re able to sit in the third rowing and speak in normal conversation voice to the multitude in the front in a seven - seater , which is rare . So those matter we ’re really , really good at . We have a howling roadmap on propulsion energy and drive units , making very , very gamey - carrying into action and efficient EVs . Those things we should rivet on . The base technical school for the infotainment organization , we ’re passing happy to work with Google because that yield us so many benefits . And we are all for being in all probability the first legacy carmaker to get to the position on SDVs where we are right now . We are all for standardization of emerging engineering , working through fundamental engineering partners . There ’s no reason that we should have everything unequaled in every panorama . We want to customize the client experience , and are extremely glad to work with a lot of applied science partner , both of the kind of technical school better half like Google , Nvidia , Luminar , Qualcomm , but also more traditional self-propelled partners like Bosch , to really push standardization of emerge technology in automotive .
Digital Trends : interchange gears a turn , can you give me a spry elevator pitch for the Superset political program and explain what ’s beneficial to that approach ?
Anders Bell : Yeah , I think it bug out with the fruition that computer software take to be on top of every consideration . So the Superset tech stack is based on the understanding that we can only have one software stack to run all our product . So it ’s very exchangeable to any successful major tech company , take Apple , Tesla . They ’re basically “ one technical school stack ” company . And that ’s the full-grown divergence now in automotive compared to the traditional approach with political platform that we had antecedently , which also served us super well . So we still do platforms in the sense of fomite computer architecture . This is a SPA2 elevator car . We will have a vehicle architecture , which is SPA3 , which is mostly a physical vehicle architecture upgrade or variety . They both run on the same propulsion and energy roadmap , imply the same drive social unit . It ’s a kin of drive units we evolve ourselves , and manufacture ourselves . Same with the battery . It ’s the same roadmap .
We have one compute electronics roadmap for all our railway car going fore . But in the end , what ’s most important is it ’s one software master for all our battery galvanising vehicle going forward . That is the big variety , which means everything needs to fit in the umbrella of the software system stack . So the software spate postulate to be unceasingly developed to encompass more functionality . And we basically dispose new technologies into the tech smokestack . So if you take technology like steer - by - telegram , we will qualify that into the tech slew and release it into the one master , then we can configure it into any of our elevator car . So that is the approach . Superset is the sum of all the capability fundamentally , and every automobile we progress is a subset of that .
Digital Trends : Are there any limitations to that around features that some vehicle might need that others wo n’t have ?
Anders Bell : No , not really . Those are typically business decision . If we want to do something which is fun and a good shock on the securities industry , we will dead do it . It does n’t limit us in any way . Being innovational , making the man ’s most complex consumer Cartesian product , requires key correction in how you correct it up and how you fulfil . So there ’s no built - in inhibitor . You do require to be mindful of complexity though .
I would say that our way of doing software - delimitate vehicle is not necessarily the proper agency for everybody . Depending on your bandwidth , how much complexity you have to manage , if you require to go from really entryway - level car all the way up to super luxury cars , mayhap that ’s not suitable on one technical school stack . You might need to have two . You will have a eminent need of scalability in functionality and monetary value . You might to pursue different solutions . If we bet at where we are , we have a very well - define bandwidth . We can descale easily from B to F , meaning entry - level cars for the premium segment up to F. On our combined full technical school good deal .
Digital Trends : There are a few feature that are still set up to come up to this vehicle specifically , like CarPlay .
Anders Bell : CarPlay is coming in a few weeks . That ’s in the next schedule over - the - air update , together with some other improvement . And then we commence to get into the cadence of continuously release the betterment and feature to the fleet . And here ’s the thing , right , since everything is on one software master , so once we ’ve finished wireless Apple CarPlay , it ’s available . It ’s done for all the railway car come . We do n’t need to redo that piece of work . So it will be done for the EX90 , the EX60 , all the railway car on the road , all the car rolling out of the factory . And then we ’re on to the next functionality or customer feature .
Digital Trends : What about some of the ADAS feature of speech that are coming ?
Anders Bell : Yeah , it ’s come over clock time . We have pilot assist now — lane guardianship , it interchange lane , turn index , etc . We will be doing the curve upper adaption in the give , and then it just continue . So every single firing we will have some sort of incremental ADAS update . We have the team doing the visual modality perception stack , and they ’re literally running codification to car every day . So there ’s a whole machine of testing substantiation , let go a few nominee every day , testing it out , seeing what works , testing new scenarios . So that whole machine is up and running . So we ’re getting the ironware on the road now . Now we have software factory link to it , and now the journey really begins . So this is all about kind of found at this moment on feature article parity where we ask to be , and then we can scale from there .
So this is kind of where automotive is in general . We need to get used to the whole idea of a software - defined product having this paradox , correct ? So when is it finished ? Well , the answer to that never . It ’s never finished , right . So then the next question is , well , how do you know when to launch then ? Well , when it ’s freak out awesome . That ’s the only reply you’re able to have , right ? So this is the moment we have a freaking awesome auto , and we will unceasingly make it better . But still , I mean , this is how it works at Tesla , Rivian , a circumstances of the Chinese companies . But for legacy automobile , I think getting used to that contract a of fourth dimension .
Digital Trends : I think there will still be a question around “ when is it finish ? ” When is the car no longer able to handle novel features and does n’t have the compute ? How do you think about future - proofing ?
Anders Bell : Yeah , so that ’s a in force interrogative . It ’s also kind of a new natural process for a bequest car company . When we appear at our roadmap , we attempt to predict the compute and hardware essential move frontward year by year , and see when do we design which advance and what do we need at the get - go . What we need at the get - go obviously ca n’t be what you require specifically in 2025 , but it ’s what needs to carry you for the next few years to do the functionality . So you require to augur kind of the functionality software , but also to some extent how good we will be at realise the codification efficient so can execute on the minimal amount of resources in the best possible way .
So kind of how we solve . And when we look at the SOCs , when we look at memory , communication protocols , when do we do the dissimilar upgrade , right ? We already herald we will be upgrading to Nvidia Thor . We are work on the Qualcomm roadmap , so we experience when we will be upgrading the different poppycock . We know how much memory we ask when and where . So it ’s kind of a new biz of being ahead of the game and never ever fathom out , but never going all crazy and throwing in a spate of hardware that you ’re never pop off to capitalize on either . It ’s a ok equipoise .
Digital Trends : Yeah , trying to secern the hereafter a piddling bit .
Anders Bell : Yeah , the same travel for sensing to a very large extent , and more and more on internal sensing . internal sensing we talk a lot about because it ’s the ADAS . Interior detection is as crucial . There ’s a lot of home sensing here going on . You have capacitive sensing in the steering wheel , you have a torque control , you have a driving monitoring television camera cleverly hidden . So we endeavor to do very , very discerning applied science effectuation .
The elevator car is conjecture to take care of you and not distract you . But those are also including interior radiolocation , pressure detector , warp switches , track position switches . That ’s another orbit where we kind of look on the minimum hardening of interior sensing that render you the rich data set possible and how do we get there . And it ’s always a game of kind of collecting data , correlative , and then you’re able to start removing when you experience well-heeled that your hardiness is there to rely on the data .
Digital Trends : Are you using the LiDAR sensor currently ?
Anders Bell : Yeah , presently we are mostly collecting and correlating . And it ’s part of the roadmap of endlessly rolling out better and good role . So in the roadmap , we know when we have the first right long - range substance abuser pillow slip coming out into the fleet . So it is dynamic , but more to collect and correlate right now . And then we go live . This is another scene of working with a package - define fomite . You put powerful sensing on , and then over time you unlock it . First instance typically involves running in some type of phantasm modality to really formalise that what we ’ve done in the lab and in our own run fleet work with the right robustness in the literal world .
Digital Trends : I ’d say that ’s a better approach than bulge with admit detector and then taking them off .
Anders Bell:[Laughs ] Yeah , it could be . But this car that we ’re drive right now is basically fit with the ADAS Superset . So in this one , we have full redundant system — braking , steering , calculate . So we have two Nvidia SOCs in this one . We have all the Alexander Graham Bell and whistle on microwave radar and ultrasonics and cameras . And from this Superset , this is the current Superset , we are work on with and produce other configurations with a smaller sensor do for less , more ADAS - focused cars . This is an ADAS - focussed car . But it ’s the same sensors , it ’s the same software , it ’s the same solidification . We will do a interlingual rendition without LiDAR for instance . You will have really , really good ADAS because you could cross - correlate up all the manner up to this machine . So using that variety of superset approach also here give us the welfare of being able to develop very high functionality on the low-spirited sensing element sets .
Digital Trends : You said you ’ll have a different fomite without LiDAR . How will you approach develop the consumer about why they should or should n’t bribe a car with or without LiDAR ?
Anders Bell : I cogitate I ’ll pass on that one for now because this is something that will be kind of imbed in next launching . But we ’re always depend at a wide range of technologies . Right now , we see that LiDAR is give a raft of benefits in the mechanization look of things , drive . And that ’s primarily where we put it right now . Then there ’s always other technology that we ’re calculate at that could do the same or a similar line of work .
I ’m extremely felicitous with kind of how we made [ the EX90 ] . You have like essentially two supercomputers behind here plus a very capable Qualcomm Snapdragon . It ’s all hidden . It ’s all discreet in this calm Norse interior where you obliterate the cameras and everything as much as possible . This is the sort of human - centric implementation of technology we want to do . So the car is bleeding - edge engineering , but it ’s not in your face .
We ’re going to do over - the - aviation update to two and a half million existing cars globally during this twelvemonth , to create a harmonious experience across the UIs . So [ the EX90 ’s exhibit ] is a 14 - inch screen . We have 11 and 9 - in version for the older cars . So all the elevator car we have driving on Android Automotive since 2021 will get a alike customer experience with your sensitive , telephone , spry action at law bar , your fix saloon , map . So it ’s a much better layout than what we have currently in yield . Personally , I do n’t miss CarPlay anymore . I consider CarPlay was keen in the sometime cars because UIs generally sucked big time . And I take cars now when I travel with the family , it ’s like I always stop up in CarPlay , but for this , I do n’t miss it any longer . But again , the client should decide . We make it available for them and they can determine what they require to use .
