Best of LinkedIn: Software-Defined & AI-Defined Vehicles CW 39/ 40

00:00:00: Provided by Thomas Allgaier and Frennis, based on the most relevant LinkedIn posts about software-defined and AI defined vehicles in calendar weeks thirty nine and forty.

00:00:10: Frennis supports tier one automotive suppliers with early stage market validation for their R&D efforts By combining secondary research direct OEM expert interviews And facilitated customer meetings.

00:00:21: you can find more info In The Description.

00:00:24: So diving right into that We are really unpacking some of the absolute top trends emerging across LinkedIn right now when it comes to well software defined and AI-defined vehicles.

00:00:35: Yeah, I mean if you're a mobility professional listening right?

00:00:38: Now.

00:00:39: You already know how incredibly fast this whole conversation is shifting.

00:00:43: Oh absolutely moving at light speed.

00:00:45: Right because usually you know When we talk about a vehicle getting a software update or mostly thinking About the mechanics up like How do we securely push this new code over the air to the car?

00:00:55: Yeah, the basic OTA updates.

00:00:57: Exactly!

00:00:58: But honestly that's already old news...the focus has completely shifted.

00:01:01: now The battle today isn't just about updating car.

00:01:05: It is about who controls actual intelligence That lets vehicle learn on its own.

00:01:09: It really a fundamental rewiring of industry brain.

00:01:14: We're seeing this distinct evolution away from software defined vehicle and moving toward the AI-defined vehicle.

00:01:21: Which

00:01:21: is a huge jump!

00:01:22: It is, to really understand what that actually means for automakers and suppliers we have to look at the underlying logic that drives these systems

00:01:29: Right.

00:01:30: let's ground this right away because I mean these acronyms get thrown around constantly in meetings.

00:01:34: So true...I actually saw a really sharp post from Raghunath Kallimuthu who crystallized this difference perfectly.

00:01:41: He points out that SDVs they're built on deterministic rule based logic

00:01:46: Like a strict command structure.

00:01:48: Yeah, an explicit.

00:01:49: if this then that kind of setup.

00:01:51: Human engineers write the logic and the car executes it exactly as written.

00:01:55: no deviation right.

00:01:57: but The AI defined vehicle on the other hand is Cognitive its adaptive.

00:02:02: It relies on these centralized foundation models to make real-time decisions basically by ingesting massive multimodal sensor streams all at once

00:02:11: Which is just a profound difference in engineering philosophy, right?

00:02:15: Steph and Fuchs made really fascinating observation about this transition on LinkedIn.

00:02:19: Oh yeah what did you say?

00:02:21: He noted that well an SDV lets the car change after you buy it which was The big revolution of last five years.

00:02:28: Right waking up to new feature your dashboard

00:02:30: Exactly.

00:02:30: but an AID V means the car actually learns After You Buy It.

00:02:35: That creates This massive organizational challenge.

00:02:39: How so

00:02:40: Well?

00:02:40: The question he raises is pretty critical.

00:02:42: Can automakers actually make corporate decisions as fast as their own vehicles are learning out there in the real world?

00:02:49: Wow, okay let's unpack this with a practical example just so we're all on the same page.

00:02:52: yeah

00:02:53: Let's do it.

00:02:53: think of an SDV like your smartphone getting an overnight operating system update you know.

00:02:58: So your existing apps Just run a bit faster or maybe You get a new menu layout

00:03:03: right.

00:03:03: predictable coded improvements

00:03:05: exactly.

00:03:06: but AIDV is like having a smart assistant that realizes, hey you always avoid the highway when it's raining.

00:03:14: So it proactively adjusts car suspension for slick city roads and reroutes without ever saying single word.

00:03:20: It's anticipating what you need not just executing command.

00:03:24: Exactly.

00:03:25: That's really great way to visualize but getting proactive learning assistant actually work inside two ton machine moving at seventy miles per hour involves some serious technical friction.

00:03:37: Oh, I can only imagine.

00:03:39: Yeah Dr Patrick Bartsch highlighted this incredible contradiction happening right now in the public eclipse S-Core specification.

00:03:46: Wait for anyone who isn't like deep in the software architecture weeds What exactly is the s core specification?

00:03:53: So it's essentially an open source reference architecture.

00:03:56: Think of it Like a blueprint For how vehicle software systems should be built and standardized across The entire industry.

00:04:01: Okay

00:04:01: got It

00:04:02: And Barts found two completely conflicting requirements sitting Right there the exact same document.

00:04:08: Wait, really in the same spec?

00:04:10: Yes one section of the spec demands deterministic build time inference to guarantee functional safety

00:04:17: which basically means rock solid predictable code where you know exactly what will happen every single time.

00:04:22: right because you want your anti-lock brakes to work the exact Same way Every single time You have a pedal.

00:04:28: yeah I definitely Want my breaks To be predictable.

00:04:30: Exactly.

00:04:31: but then Another section of that same document demands streaming generative AI agents for context-aware interactions.

00:04:38: So fluid probabilistic AI, that kind of figures things out on the fly?

00:04:43: Right and as Barge points out it reads like two different engineering teams wrote The Requirements without ever sitting in the same room.

00:04:49: Oh man

00:04:50: It really highlights the core tension of the ADV.

00:04:53: you know how do you blend non deterministic creative AI with strict rigid automotive safety regulations.

00:05:00: Which naturally brings up a huge hardware constraint too.

00:05:03: I saw Sean say hi, Wade in on this exact problem.

00:05:06: Oh yeah the power and compute issue.

00:05:08: Right

00:05:09: He pointed out that putting it five hundred watt liquid cooled server In the trunk of car just to run a conversational AI agent makes absolutely zero commercial sense.

00:05:17: No you'd completely kill the EV battery range.

00:05:20: Exactly!

00:05:21: The range and thermal budget of a vehicle would just be ruined, so... ...the real transition to AI-defined vehicles isn't going to be powered by massive chat GPT style models running locally in the cabin.

00:05:33: No it can't be.

00:05:34: It's gonna have to rely on compact highly efficient edge models that operate strictly within the vehicle's tight power constraints

00:05:41: Because the vehicle is essentially the ultimate restricted edge node.

00:05:44: You've got limited power, limited cooling and very

00:05:47: limited space.

00:05:49: And David Kelly brought up another critical boundary that limits how much we can rely on the cloud to handle this heavy lifting connectivity.

00:05:56: What happens when network

00:05:57: drops?

00:05:58: Right like driving into a tunnel or an underground concrete parking garage.

00:06:02: Precisely AI models in cars cannot rely solely onto five-G connection to some Cloud server Safety.

00:06:09: Critical functions absolutely must have an on vehicle fallback

00:06:13: Because your steering and braking systems can't just wait for a network ping to decide what to do next.

00:06:18: Exactly, it's a literal matter of life-and-death.

00:06:21: So to run those advanced AI models natively on the edge without completely draining in battery or crashing system when cell service drops The actual physical and digital plumbing has radically changed.

00:06:33: No absolutely you cant slap neural networks like its built by now.

00:06:38: no

00:06:38: thats a recipe for disaster.

00:06:40: We really have to talk about the physical reality of manufacturing these machines.

00:06:44: Samuel Kaye argued this point, really well.

00:06:46: What did he say?

00:06:48: He looked at the agonizing sixty-month vehicle development cycle that's pretty standard in Europe and pointed out That there isn't a labor efficiency problem.

00:06:56: So it is not just engineers working slowly.

00:06:58: No

00:06:59: Not at all.

00:07:00: It is an architectural complexity problem

00:07:03: Which means The way they are building software layers Is fundamentally broken from start Right.

00:07:08: To compress that development cycle down to twenty-four or thirty months, OEMs have to stop buying software the way they buy hardware parts.

00:07:16: Like where they just order a finished black box from a supplier?

00:07:19: Exactly!

00:07:20: They need to dictate machine readable APIs instead shift heavily to high fidelity virtual validation and really embed compliance directly into their continuous integration and continuous deployment pipelines...or CICD

00:07:34: Right.

00:07:35: And if you're a software engineer listening, You know that CICD is standard everywhere in Silicon Valley?

00:07:40: You write code it gets tested automatically and it gets deployed.

00:07:43: It's smooth and constant.

00:07:44: Yeah but an automotive waiting for physical prototype just to test your code Is what causes those massive delays.

00:07:50: Simeal basically saying they have to automate the testing virtually.

00:07:53: There's also a massive physical roadblock to this too.

00:07:56: Anterchat Singh B shared mind-blowing stat about What this architectural complexity looks like In real world.

00:08:02: Oh right, the wire harness.

00:08:03: Yes!

00:08:03: The wire harness

00:08:04: It's wild.

00:08:06: Modern wire harnesses and premium cars now run up to five kilometers long.

00:08:10: Just think about that for a second.

00:08:12: Just picture threading five kilometers of copper wiring through metal chassis.

00:08:17: They are so massive And incredibly heavy That human line workers can't even lift or install them anymore.

00:08:22: they literally require Massive robots To position them.

00:08:26: it is crazy.

00:08:27: its totally unsustainable.

00:08:30: but Anterjaud noted that moving to a zonal architecture is finally starting to solve this.

00:08:37: Okay, how so?

00:08:38: Well he pointed out in recent BMW models adopting the Zonal setup cut six hundred meters of wiring and reduced the harness weight by thirty percent.

00:08:50: Wow, okay.

00:08:51: I have to stop here for anyone not living in the electrical engineering department What exactly is Zonal architecture?

00:08:58: because if it cuts six hundred meters of heavy copper Why isn't every single car on the road built this way today?

00:09:04: It's a great question.

00:09:06: So think of the traditional way of building a car like central brain with a dedicated completely separate nerve running to every single muscle.

00:09:15: Okay so you have a wire running all the way from the central computer in the front to the left taillight, another to right taillite and other to rear sensor.

00:09:23: Exactly!

00:09:25: Zonal architecture changes that it groups car into geographical hubs or zones.

00:09:31: Hence name

00:09:32: Right.

00:09:33: So you have one high speed data cable going from main computer to a rear zone hub in trunk.

00:09:39: That hub then distributes signals locally to tail lights and sensors.

00:09:43: Oh, so it drastically reduces the amount of wire traversing... ...the whole length of a vehicle.

00:09:47: Exactly!

00:09:48: That makes perfect sense.

00:09:49: So back to my question If physics and engineering make this such an obvious win what is stopping the entire industry from doing this overnight?

00:09:58: Well according that same discussion on LinkedIn The real roadblock isn't technology at all It's organizational structure

00:10:06: Really?

00:10:07: Corporate politics

00:10:09: Basically?

00:10:09: yeah.

00:10:10: In traditional automaker you have highly siloed departments To implement a zonal architecture, the department that builds the central computers might have to take on massive budget increase and build these advanced zone hubs.

00:10:22: Ah... Also the Department that buys wiring can save money?

00:10:26: Exactly!

00:10:27: And nobody wants be executive whose budget just skyrocketed even if it saves the company money overall

00:10:32: Because NOBODY owns the whole vehicle view.

00:10:35: to force trade off.

00:10:36: Precisely The corporate silos are literally weighing cars down.

00:10:42: But you know..the market is forcing their hand regardless.

00:10:44: Yeah, Matt Damascino highlighted some McKinsey data showing that domain and zonal architectures are projected to absolutely surge anyway.

00:10:53: The numbers

00:10:53: a huge

00:10:54: they are.

00:10:54: we're looking at going from twenty nine percent of global light vehicle production in twenty twenty five To seventy seven percent by twenty thirty-five

00:11:02: because it's some point.

00:11:03: the old way just collapses under its own weight.

00:11:05: It's too complex

00:11:06: right.

00:11:07: Konstantin Shoroshorsky had a really candid take on this whole transition.

00:11:11: What did

00:11:11: he say?

00:11:12: He argued that the automotive industry doesn't actually have a software problem, it has customization problems.

00:11:17: Oh!

00:11:18: That's an interesting way to put it

00:11:19: Right...he says reinventing unique electrical architectures and basic infrastructure for every single car model is just pure waste.

00:11:26: It's like every smartphone manufacturer trying to invent their own version of Bluetooth from scratch For every new phone they release.

00:11:33: Exactly Yolkham Langenwalter echoed that exact sentiment.

00:11:37: He pointed out that OEMs are wasting massive amounts of engineering resources duplicating middleware

00:11:42: Things like AutoSAR, right?

00:11:44: Yes

00:11:44: And just for context autoSAR is essentially standardized automotive software architecture.

00:11:50: It's the invisible plumbing That lets different computers in the car talk to each other.

00:11:55: Right

00:11:55: and Langenwalter point Is that automakers should really be standardizing that plumbing across-the-board

00:12:01: because customers Do not buy a car because it has a uniquely coded operating system kernel.

00:12:06: No, they don't even know what that is

00:12:07: exactly by for the features They can actually see and feel like advanced driver assistance systems ADAS or autonomous driving capabilities.

00:12:15: That's where OEMs should be putting their differentiation.

00:12:18: And By The Way if you want to keep track of these rapidly shifting architectures?

00:12:22: You Know What It Means For Your Specific Corner Of The Mobility Industry Make Sure To Subscribe So You Catch Our Future Deep Dives.

00:12:28: Yeah We Are Covering This Evolution Constantly so Definitely hit subscribe.

00:12:32: Absolutely,

00:12:33: but you know that realization that building the foundational plumbing from scratch is just unsustainably expensive?

00:12:40: It's driving some massive industry moves right

00:12:42: now.

00:12:43: it really is

00:12:44: because ripping out five kilometers of copper wiring Requires an entirely new software brain and automakers suddenly find themselves staring at a development bill They just cannot pay alone

00:12:54: which brings us to The great build versus partner dilemma.

00:12:57: yes

00:12:58: this strategic divergence happening Right Now Is wild.

00:13:02: Dan Zalinsky highlighted a really fascinating story about this, you probably heard that Honda Nissan recently ended their merger talks?

00:13:09: Yeah

00:13:09: That made a lot of headlines.

00:13:10: But what's incredibly telling is that instead of just walking away completely they agreed to jointly develop and standardize the core ECUs And middleware for their next generation STVs.

00:13:21: Wow!

00:13:22: That Is A massive signal To The Market.

00:13:24: Huge

00:13:24: You Have These Fierce Historic Competitors recognizing that the foundational software layer is simply too expensive and too complex to duplicate.

00:13:33: They have share of the burden.

00:13:34: Yeah, And Honda's pushing this strategy even further.

00:13:38: Quasar Unis noted that HONDA is also partnering with Applied to use their vehicle OS.

00:13:43: So they are clearly making a choice to partner for the foundation so it can focus its internal capital elsewhere.

00:13:49: Exactly And we are seeing similar shockwaves in Europe, too.

00:13:53: Oh for sure!

00:13:53: According to Inebea UFO Volkswagen reportedly bypassed a full-stack automated driving offer from NVIDIA and instead chose the UK based AI startup called WAVE for its autonomous driving tech.

00:14:05: WAVE yeah?

00:14:06: Yeah so the plan is integrate WAVES end-to-end AI right into VW's existing hardware & sensor suite.

00:14:13: It s very pragmatic approach.

00:14:15: I mean you take best in class start up AI just drop it onto your scaled manufacturing base.

00:14:20: Okay, I have to push back here though.

00:14:22: Just a bit

00:14:22: all right Let's hear it.

00:14:23: if i'm a legacy automaker like vw and I just slap A third party startups brain into my chassis What am I actually selling?

00:14:30: That's

00:14:30: the big question.

00:14:31: Right?

00:14:31: don't they risk becoming just hollow metal boxes where all The value is in the software They don't even own?

00:14:36: that Is the exact fear driving the opposite strategic approach In the market right now.

00:14:41: who's doing the opposite?

00:14:43: look at Volvo in their software subsidiary Zensie act.

00:14:46: okay Dr.

00:14:47: Juergen Dickman shared some great insights on how they are deliberately taking the harder road.

00:14:51: Interesting!

00:14:52: Yeah, Zensi Act is building its own in-house based software platform.

00:14:56: It's a highly painful incredibly slow and expensive process but it grants them absolute control.

00:15:03: So they own end to ends?

00:15:04: Yes They own data, dictate update cycles And control entire product strategy.

00:15:09: So they aren't handing the keys to The Kingdom, some tech giant.

00:15:12: Right and technically it allows them do something really crucial for safety regulators...the guardrail system.

00:15:18: What's that?

00:15:19: By building of themselves They can keep a deterministic rule-based safety guard rail completely independent from AI neural network.

00:15:28: Oh, I see.

00:15:29: So if the AI suggests a lane change?

00:15:32: The rule-based system checks the physics and the sensors And If there's a car right there in the blind spot it overrides the AI and says no absolutely not

00:15:40: exactly.

00:15:41: It's a hard stop.

00:15:42: That makes a lot of sense.

00:15:44: You really see the financial value of owning that core technology when you look at the Chinese OEMs.

00:15:49: Rahul Aruga talked about this.

00:15:50: Right?

00:15:50: Yeah

00:15:51: He pointed out how companies they are actively monetizing their in house IP.

00:15:56: It's crazy what they're doing.

00:15:58: companies like Lee Otto and Neo are spinning off their internal chip divisions to secure external customers.

00:16:03: Yeah, an X-Ping is literally licensing its electronic architecture back to Volkswagen.

00:16:09: That is the ultimate flip right there

00:16:10: it really is.

00:16:11: They took their massive R&D costs And just turn them into revenue streams by selling to the legacy players.

00:16:16: But ultimately you can have all this strategy You want?

00:16:20: The real proof Is in the execution.

00:16:22: definitely.

00:16:23: Maristela Colonostacio brought up the recent Gartner Digital Automaker Index, and it is incredibly revealing.

00:16:29: Oh I saw this!

00:16:30: Hyundai and Kia rocketed to tenth place in those global rankings

00:16:34: Which surprised a lot of people.

00:16:36: Right.

00:16:36: But why did they jump?

00:16:37: Because...they aren't just doing PowerPoint presentations about these software-defined futures.

00:16:43: No They're actually building

00:16:44: them.

00:16:44: They are actively manufacturing & deploying these zonal architectures And connected operating systems across their high volume fleets today.

00:16:52: They are treating software architecture as a hardcore manufacturing discipline

00:16:57: and this leads to a critical realization for anyone in the mobility space right now,

00:17:01: which is

00:17:01: whether an OEM decides to build their AI in-house like Volvo or partner up like Honda.

00:17:07: deploying This Tech at scale completely rewrites The financial model of the vehicle for its entire lifespan.

00:17:13: The unit economics just fundamentally change because

00:17:15: the car doesn't stop costing the automaker money once it rolls off the lot

00:17:19: exactly.

00:17:20: Federico Magno co-authored a white paper with Bosch that hits this right on the head.

00:17:25: AI architecture is now a continuous business model decision.

00:17:28: Yeah, it's not one and done sale anymore.

00:17:30: Right when you move to an AI defined vehicle You aren't just processing simple commands locally any more?

00:17:35: You have complex agentic AI interactions happening constantly in the background.

00:17:40: The

00:17:40: car was constantly calling APIs sending data querying models

00:17:44: Exactly, and their analysis showed that because of this continuous computing the cost per active user could skyrocket by up to two hundred seventy five percent over time.

00:17:53: Wait even if the costs of AI compute goes down?

00:17:56: Yes!

00:17:57: Even if raw price of AI computing tokens drops The sheer volume of continuous intelligence running in background drives overall cost massively upward.

00:18:06: Wow so automakers have to figure out who pays for on-going server bill five years into vehicles life

00:18:12: Which means they will inevitably have to charge the customer subscription fees or find ways to heavily monetize their vehicle's data.

00:18:19: And that software explosion is creating a massive secondary market too?

00:18:23: Oh, for sure!

00:18:24: Frank Turlup shared a stark warning of the aftermarket.

00:18:27: The automotive software market has projected nearly double reaching over eighty three billion dollars by twenty thirty-three.

00:18:33: That

00:18:33: is huge.

00:18:35: Vehicles truly are becoming smartphones on wheels

00:18:37: Which completely upends the traditional repair and maintenance industry.

00:18:41: Just think about what this means for a regular mechanic or collision repair shop.

00:18:45: Yeah, it's totally different ballgame.

00:18:47: Fixing car after offender bender isn't just about bending metal back into shape or swapping a bumper anymore.

00:18:52: Repairs are now going to have verify exactly which software version of the car is running.

00:18:56: Yep They'll check if an over-the air update changed how sensors behave right before crash And they have to validate that the vehicle's cybersecurity is still fully intact before hand-the keys back.

00:19:07: It basically transforms a mechanical fix into a rigorous IT audit.

00:19:12: Exactly!

00:19:13: The scale of it just staggers.

00:19:15: Vivek Aviya pointed out, that modern EV has over one hundred million lines of code

00:19:20: That is honestly hard to wrap your head around Right.

00:19:23: Because of this immense complexity Cybersecurity can no longer be treated as final IT compliance check box right after the car goes production.

00:19:31: No, it has to be a fundamental life cycle engineering discipline because the attack surface isn't just that infotainment screen on dashboard anymore.

00:19:39: It includes public charging networks car plugs into mobile apps they own or uses to unlock doors and even third party API's.

00:19:47: car use is navigate

00:19:48: completely interconnected web everything from core api contracts at OEM level down local collision repair shop.

00:19:56: tied together by this shift static software to adaptive AI

00:20:00: The entire ecosystem has to level up at the exact same time.

00:20:03: They do, and as we navigate this massive transition it leaves us with a really critical almost philosophical question to consider.

00:20:11: Okay what's that?

00:20:12: Well if an AI-defined vehicle is constantly adapting and learning from its human driver's behavior over time who is legally and ethically responsible?

00:20:21: If the vehicle learns about habit That contributes to an accident Maybe years after it left the factory?

00:20:27: Oh wow That is, that has a lot to think about.

00:20:30: It really does.

00:20:31: If you enjoyed this episode new episodes drop every two weeks.

00:20:35: Also check out our other editions on charging and battery tech commercial fleets an autonomous and mobility markets.

00:20:41: Thanks for tuning into the discussion today.

00:20:43: Thank You so much For joining us On This deep dive.

00:20:45: don't forget To hit subscribe So you never miss An update as this incredible industry Continues to evolve.