Best of LinkedIn: Software-Defined & AI-Defined Vehicles CW 43/ 44

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

00:00:08: Freenes 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.

00:00:19: You can find more info In the description.

00:00:21: Yeah!

00:00:22: It is definitely an interesting couple of weeks in data.

00:00:25: Oh For sure.

00:00:26: So look Imagine you're driving sixty miles per hour right?

00:00:29: In the blink of an eye, I mean just a hundred milliseconds your car moves nearly nine feet.

00:00:34: Right Which is basically the width of a crosswalk.

00:00:38: and today we're doing it deep.

00:00:39: dive into why.

00:00:40: putting artificial intelligence Into that kind of fast-moving really high stakes physical environment Is completely breaking how the auto industry builds cars?

00:00:48: Yeah It's breaking everything.

00:00:50: exactly like if you're a mobility professional listening right now.

00:00:53: You've heard the phrase software define a vehicle for years.

00:00:57: but um Looking at the stack of intelligence from the past two weeks.

00:01:01: The theoretical debates they are over.

00:01:03: we are seeing actual messy real-world execution

00:01:07: yet, but transition From you know vision to actual industrialization is finally happening.

00:01:12: and For anyone building product roadmaps right now?

00:01:15: The main takeaway Is that this isn't just about like pasting an iPad onto a dashboard or adding A slick voice assistant.

00:01:23: Right We Are observing This Fundamental Honestly, painful reorganization of the entire engineering and supply chain ecosystem.

00:01:32: Yeah, painful is definitely the word.

00:01:34: so to make sense Of that reorganization for you we are going to cut through The noise today and focus on the absolute top trends shaping this space across all our sources.

00:01:44: We'll start with the unsexy but critical foundation of architecture And compute.

00:01:49: then we're gonna look at This massive philosophical split happening right now between AI defined vehicles and physical AI.

00:01:56: That was huge.

00:01:57: It really is.

00:01:58: and finally we'll get into the friction points, right?

00:02:00: The messy realities of partnerships connectivity life cycles And how geo political regulations are literally halting assembly lines.

00:02:08: So diving in to that first theme architecture and compute the core issue Is that you just cannot run advanced AI on legacy hardware.

00:02:16: yeah

00:02:17: automakers Are realizing before they can deploy any of this flashy artificial intelligence, they have to completely tear out and like rewire the vehicle's entire nervous system.

00:02:27: The analogy that always comes to mind for me is it's trying to run a modern smart home with four-case security cameras an automated climate control but you're running it on nineteen fifties.

00:02:38: noven tube wiring.

00:02:39: That's exactly what

00:02:40: It Is But its actually worse than that isn't because your talking about gigabytes of video and sensor data every single second.

00:02:48: Yeah,

00:02:48: the bandwidth requirements alone are staggering which is why your nineteen fifties wiring analogy hits home so well.

00:02:55: Jay & John laid out this excellent blueprint in his post Of The four architectural foundations that are required before AI can scale across any vehicle platform.

00:03:03: Okay?

00:03:03: What's the first one?

00:03:04: So the first and arguably, the heaviest lift for OEMs is moving to a zonal vehicle architecture.

00:03:10: Right getting rid of the mess?

00:03:11: Exactly!

00:03:11: Historically automakers just kept adding separate electronic control units or ECDUs every single time they added a feature.

00:03:18: so like one box-for-the-wiper's One for the brakes

00:03:20: For seats exactly you'd end up with hundred separate computers And literally miles of heavy copper wire.

00:03:26: Wow

00:03:27: Zonal design Just throws that out.

00:03:29: it consolidates computing power into these centralized hubs based purely on their physical location in the car.

00:03:35: So, you know front rear left right it massively reduces wiring weight and simplifies how processing power is actually allocated.

00:03:42: Okay so you clean up the physical wiring You centralize the brain.

00:03:47: What are the other three foundations he mentions?

00:03:50: Well The second Is the abstraction of software from hardware.

00:03:54: Right

00:03:54: Because historically code was hard-coded to a specific chip.

00:03:58: Abstraction creates this middle layer meaning you could swap out an obsolete chip for a newer one without having to rewrite millions of lines of code.

00:04:05: Which is huge, with the life cycle of car?

00:04:07: Exactly!

00:04:08: Then third foundation cloud enablement.

00:04:11: and this isn't just about beaming down software update.

00:04:13: it's creating continuous engineering feedback loop where the car uploads its edge case data back into the Cloud To retrain models

00:04:21: Right.

00:04:22: And fourth which has been baked in every level from silicon up Is security & Data Protection.

00:04:29: Okay, so looking at those four pillars I really have to ask is the industry actually ready To swallow its pride here?

00:04:35: because for decades automakers operated on this belief that they needed to you know own and build And customize every single line of code.

00:04:43: Every piece of silicon just to differentiate their brands.

00:04:47: Are they?

00:04:47: Actually realizing They need to stop building these bespoke foundations For every single vehicle model?

00:04:53: yeah The sheer complexity of these systems Is basically forcing Their hand And we are seeing major capitulations on that old build at all mentality.

00:05:01: Really?

00:05:01: Yeah, Raphael Sotomayor and Nikhil Reddy Kottapalli shared a massive update highlighting this exact shift.

00:05:08: so Volkswagen Group literally one of the largest automakers on the planet has officially selected NXP as thirty two k five microcontrollers to serve is their common standardized building block.

00:05:20: Wow

00:05:20: yeah for their futures own architecture across All they're various brands

00:05:24: Standardizing microcontrollers across an entire global portfolio, I mean that is a massive departure from the old way of doing things.

00:05:31: It's huge?

00:05:32: Buying a common pre-integrated block instead of engineering it from scratch... That changes the daily workload on OEM engineering team completely!

00:05:39: It really does.

00:05:40: And Mark Gershik actually made a compelling argument about this during a panel at NXP Tech Days.

00:05:45: He told the audience that OEMs need to move aggressively from build To buy for these deep foundational layers.

00:05:53: Yeah, that makes sense.

00:05:54: Think about the microprocesses running in a car like start-up sequence or The cryptographic handshake that securely processes an over-the-air update.

00:06:02: I can safely say no customer In the history of the world has ever bought a car because it had A superior cryptographic hand.

00:06:08: shake.

00:06:09: right

00:06:09: and That is Gertrack's entire point.

00:06:12: consumers only care About the features they actually interact with

00:06:14: exactly.

00:06:16: automakers have been burning thousands Of engineering hours treating those foundational elements as brand differentiators.

00:06:22: If an OEM wants to get a software-defined vehicle production faster, lower their safety risk manage the budget they have to buy preintegrated software foundations.

00:06:32: Yeah!

00:06:33: They need redirected most expensive engineering talent toward application layer, autonomous driving and UI stuff that actually sells car

00:06:40: Exactly.

00:06:41: And when you consolidate architecture and optimize it The financial left side on back end is staggering

00:06:47: Right because of cloud costs.

00:06:50: Thomas Lavlozansich noted a recent move by BMW regarding their connected vehicle platform Orbit which handles data for over twenty four point five million Connected vehicles globally.

00:07:01: That's some

00:07:01: massive amount of data.

00:07:03: Oh, it's wild.

00:07:04: so BMW shifted this massive platform onto AWS Graviton processors.

00:07:10: For context Gravitone uses an arm-based architecture Which is vastly more efficient for specific microservices compared to like traditional legacy server chips.

00:07:19: Okay,

00:07:20: so what was the result?

00:07:22: By making that switch BMW cut their CPU consumption by fifty-seven percent.

00:07:26: Fifty

00:07:27: seven percent!

00:07:27: That's

00:07:28: insane.

00:07:29: and they realized about six hundred thousand dollars in annualized infrastructure savings.

00:07:33: Wow...that proves that optimizing the back end compute architecture isn't just some you know abstract engineering exercise.

00:07:39: it is a highly effective operational cost lever.

00:07:42: Absolutely When you're dealing with millions of connected endpoints constantly pinging a server, a fifty-seven percent drop in processing overhead translates to massive long term financial endurance.

00:07:51: Okay so what the computing foundation laid right?

00:07:54: The nervous system rewired and the cloud backend optimized.

00:07:59: we really need to look at how all that computing power is actually being utilized on the front end.

00:08:04: yeah because the intelligence strategy Augustine Friedl mapped this out in the brilliant two-by-two matrix and his post showing a clear split between AI defined vehicles, AIDV ,and physical AI.

00:08:17: Yeah!

00:08:17: And the distinction between those two paths is honestly going to define the next decade of capital allocation.

00:08:22: How so?

00:08:23: Well...AI

00:08:24: Defined Vehicles are inward looking.

00:08:27: The intelligence shapes cars interior experience The driving dynamics Predictive maintenance.

00:08:32: But physical AI is a much more aggressive strategy where the intelligence extends beyond the car into the broader physical world.

00:08:39: So humanoid robotics urban air mobility

00:08:43: and we are seeing players like Tesla an x-pang pushing incredibly hard Into that physical AI space.

00:08:48: Yeah, very hard

00:08:49: Xiaoping.

00:08:50: he even presented X Peng's physical AI ecosystem at The Paris Motor Show.

00:08:54: They were launching their new electric SUV right alongside like bipedal humanoid robots in flying cars.

00:09:00: It's crazy

00:09:01: But frankly, I need you to defend this strategy for me because anyone working on the factory floor.

00:09:06: This aggressive push from physical AI seems like a dangerous ego driven distraction?

00:09:11: Is

00:09:11: it?!

00:09:12: We have OEMs that still struggle get a smartphone to sync via Bluetooth reliably and now they want to build humanoids using the same A.I foundation.

00:09:21: Look!

00:09:21: The skepticism is completely warranted but the economic theory behind physical AI actually really sounds.

00:09:28: Okay, walk me through it.

00:09:29: The single most expensive part of artificial intelligence is training the foundation models.

00:09:34: If you spend a billion dollars training in neural network to understand physical space physics object permanence just for a car Why limit that multi-billion dollar brain to just a car?

00:09:45: Oh I see yeah

00:09:47: if you could reuse that same spatial intelligence model and cross a humanoid robot factory arm and autonomous drone, you theoretically amortize those massive R&D costs across multiple trillion dollar industries.

00:10:00: Amortizing the cost makes sense on a spreadsheet for sure but deploying it in The Physical World is where the theory just hits wall of physics.

00:10:07: Yeah does.

00:10:08: Lavanya Suresh Kanan broke down.

00:10:10: why deploying AI On-the-Edge like inside moving vehicle Is an entirely different nightmare compared to running model In temperature controlled data center.

00:10:18: It all comes into the brutal constraints of edge physics.

00:10:22: Latency is the most obvious one.

00:10:24: Returning to your opening thought, at sixty miles per hour The car travels two point seven meters roughly nine feet in just hundred milliseconds

00:10:32: Right.

00:10:33: So if the AI model takes too long to you know fuse the camera data Recognize a pedestrian calculate the trajectory and command-the-breaks.

00:10:41: an accurate answer that arrives A fraction of a second late Is legally and physically a fatal error.

00:10:48: And you're trying to do all that calculation inside a metal box sitting in the Arizona Sun, right?

00:10:52: Exactly.

00:10:54: Kanan points out the issue of thermal throttling because in a data center You just pump in more liquid cooling but in a car...you have a strictly fixed cooling budget.

00:11:03: Right can't just add fans

00:11:04: exactly.

00:11:05: and if the processor heats up it slows down To protect itself.

00:11:09: when a chip slows down It might drop the frame rate of the AI vision system and dropped frames equal blind spots at highway speeds

00:11:16: which leads directly into this terrifying safety gap highlighted by Roberto Hiroshi Uneda.

00:11:22: The transition to AI creates a completely new category of vehicle failure.

00:11:26: Yeah, this part is wild.

00:11:28: Historically the auto industry relied on deterministic engineering.

00:11:33: if wire shorted sense or broke are mechanical linkage snapped car through diagnostic trouble code that check engine.

00:11:40: like him on there was physical trail evidence.

00:11:43: But Yoneda asks, what happens when an AI model makes an unsafe driving decision?

00:11:48: Yet every conventional diagnostic box in the vehicle reads green.

00:11:52: Right The camera works perfectly!

00:11:53: The internal network has zero latency...the thermal management is fine..The code executes exactly as it was compiled but the neural networks simply misinterprets the visual scene and make a fatal steering decision.

00:12:06: This system functions flawlessly to execute a catastrophic error.

00:12:09: There are no fault codes because nothing broke in the traditional sense, right?

00:12:14: And the industry's current validation processes like checking for software bugs or testing edge cases on a track are struggling to prove behavioral safety of neural network that is constantly learning and changing.

00:12:26: Which

00:12:27: is terrifying!

00:12:29: They found that sixty-eight percent of automotive organizations are currently using or piloting Agentec AI in product development.

00:12:43: Sixty eight percent?

00:12:44: Yeah,

00:12:44: and just to clarify for everyone agentec ai means the system doesn't Just analyze data and give you a prompt.

00:12:50: it actually makes decisions And executes actions on its own right.

00:12:54: so sixty eight percent Are testing these autonomous agents but a staggering five percent Actually have mature AI governance in place.

00:13:02: It is a classic case of capability outpacing organizational competence.

00:13:08: When you move agentic AI closer to safety critical physical decisions, the lack of governance isn't just an IT problem anymore it becomes a massive liability hazard.

00:13:17: oh definitely

00:13:19: because how do you sign off on the safety of a system?

00:13:22: when?

00:13:22: You can't deterministically prove.

00:13:24: however

00:13:24: will react every single novel situation.

00:13:28: And because no single company can solve the thermal throttling, the latency and the neural network safety validation on their own.

00:13:35: The sources show this massive surge in complex partnerships.

00:13:38: Yeah everybody is teaming up

00:13:40: But bringing three or four massive tech-and-auto companies together into a single vehicle architecture creates brand new legal and life cycle frictions.

00:13:49: yeah the era of the lone wolf automaker Is absolutely over that technical requirements are just too vast.

00:13:55: Alexander Schaub mapped out a perfect example of this friction by looking at Nissan's ProPilot program.

00:14:01: Right.

00:14:01: Nissan is using Wave to provide the AI driving intelligence, and Valeo is acting as the system integrator tying the sensors and compute hardware together.

00:14:11: so you have Nissan brand Waves Brain and Valeos Nervous System in three-way partnership.

00:14:18: like when that learned driving model makes mistake who actually owns safety case on liability?

00:14:25: Well, this concept of architectural authority is becoming the most hotly contested issue in automotive boardrooms right now.

00:14:31: If Nissan delegates the intelligence to Wave but Wave's model requires Vallejo specific LiDAR parameters to see clearly who gets a final say on whether total system is safe enough for public roads?

00:14:43: Right.

00:14:44: Procurement contracts written ten years ago simply aren't equipped to handle shared liability on a learned AI

00:14:50: model.

00:14:50: Yet despite those liability questions, everyone is still teaming up.

00:14:54: Kasar Yunus and Jennifer W posted about applied intuition partnering with both Honda and Nissan.

00:15:03: It's a land grab for foundational partners right now.

00:15:05: It

00:15:06: really is, oh by the way if you want to keep tracking how these massive OEM partnerships shake out and who ultimately wins that battle for architectural authority make sure you subscribe so you catch our future deep dives.

00:15:18: Yep You won't wanna miss it.

00:15:20: And as these partnerships lock in we really have to look at the life cycle friction they expose.

00:15:26: Yeah Joe Mussani and Marie Hagelin brought up the glaring mismatch in the connected car ecosystem.

00:15:33: Physical hardware ages significantly faster than digital

00:15:36: networks.".

00:15:36: This is such a huge point, right?

00:15:38: Automakers are building these complex safety cases around the assumption of constant high bandwidth connectivity using five T or satellite networks to monitor robotaxies and stream updates.

00:15:49: but The mechanical vehicle has built last fifteen to twenty years cellular network sunset much faster.

00:15:56: Like, we saw millions of early connected cars lose their emergency SOS features when the three G networks were turned off.

00:16:03: We are going to see it again as four g eventually faces out.

00:16:06: yeah The mechanics of this mismatch our brutal.

00:16:10: if a vehicles AI system relies on cloud processing to supplement its edge computing and that vehicle loses Its network connection because the telecom industry upgraded its towers That fifteen year old car is effectively pricked.

00:16:23: It's like buying a brilliant cloud-connected brain but putting it in body where the Wi-Fi antenna is permanently glued inside and goes obsolete for five years.

00:16:31: That

00:16:31: exactly yet!

00:16:32: They can no longer glue a Wi-fi antenna into dashboard, forget about that... they have to design physical connectivity architecture to be physically modular AND upgradable from day one.

00:16:43: Speaking of upgradeability we need to talk about the double edge sword over the air or OTA updates.

00:16:49: When OTA works, it feels like magic.

00:16:52: Javier Varela pointed out how Rivian pushed this fun Halloween update that turned their vehicles into a ghostbusters universe.

00:16:59: It customized the external lighting changed the interior audio to spooky themes Altered the UI.

00:17:05: That is delightful You know?

00:17:07: That's the promise of the software-defined vehicle A car that gets more fun and capable over time.

00:17:12: Yeah!

00:17:12: It creates incredible brand loyalty because The vehicle feels alive and responsive.

00:17:17: But Armin Horan brought up the dark reality of treating an eight thousand pound electric truck like a rolling smartphone.

00:17:24: This

00:17:24: was scary!

00:17:25: It was, he reported on a Rivian owner who experienced a cascading software glitch while driving at highway speeds because overwrites core system files.

00:17:37: If that validation process isn't flawless, a bad piece of code doesn't just crash your Spotify app.

00:17:43: it can knock out your instrument cluster you're power steering assist or regenerative braking.

00:17:47: while moving at seventy miles per hour The same pipeline delivers.

00:17:51: the fun Ghostbusters theme can deliver life-threatening system failure.

00:17:56: Securing OTA Pipeline is not an engineering and safety problem anymore either.

00:18:01: It has become a geopolitical weapon.

00:18:03: Yeah, this is crazy.

00:18:05: Kelly

00:18:05: Barnard highlighted the stark reality of the new US connected vehicle rule which heavily scrutinizes vehicles that have a quote unquote sufficient nexus to China or Russia specifically regarding data security and connectivity hardware.

00:18:19: The example here's wild by the way.

00:18:20: the Volvo EX-NINDI is authorized to be sold in the United States but the Polestar III is blocked under.

00:18:28: they roll off the exact same assembly line in South Carolina.

00:18:31: They share the same physical platform, they use the same software stack

00:18:35: but The Commerce Department isn't looking at where the car is screwed together or the quality of the code.

00:18:40: They are evaluating corporate holding structures right?

00:18:44: Both Volvo and Polestar are backed by China's Geely Holding But due to subtle differences in their corporate governance ownership percentages And how data access could theoretically be demanded by a foreign government the regulatory bodies viewed their risk profiles differently.

00:19:00: It is a brutal lesson in modern automotive manufacturing your engineering can be flawless, Your software abstraction can be perfect...your AI models could totally safe but if you're corporate holding structure triggers data security red flag and Washington your vehicle simply cannot go to market.

00:19:18: So looking at this entire landscape I mean, the thermal throttling at the edge.

00:19:23: The terrifying silent AI failures...the liability nightmares of three-way partnerships.... ...the threat of highway speed OTA glitches and geopolitical assembly line bans.

00:19:34: You have to ask why are automakers still pouring billions into this?

00:19:38: Well Pedro Pacheco ran numbers on the Gartner Digital Automaker Index And the financial bottomline is absolute.

00:19:44: The data proves that despite all these headaches, the automakers that achieve high digital and tech maturity are seeing a direct strong correlation with operating profit growth.

00:19:54: Exactly!

00:19:54: Mastering software-defined architecture directly equals financial endurance.

00:20:00: The value creation of getting this right lowering your backend AWS costs standardizing your microcontrollers successfully deploying safe AI it is simply too massive to ignore.

00:20:12: It really is a complete painful reinvention of what a car actually.

00:20:15: Is

00:20:16: it?

00:20:18: And to wrap up our analysis of these sources I want to leave you with scenario to ponder that combines all These friction points we've talked

00:20:25: about.

00:20:26: okay, let's hear

00:20:28: if the industry has already struggling today With cellular network sunsets software glitches defining architectural authority and fresh partnerships.

00:20:38: Think ahead to the year twenty forty

00:20:41: Oh boy.

00:20:42: Imagine a fifteen-year old AI driving model operating in third hand physically degraded vehicle trying to communicate its edge cases over a six G network that didn't even exist when the car was manufactured with That vehicle inevitably misinterprets a scene and makes a fatal mistake.

00:21:00: Who is going to be legally?

00:21:02: And morally responsible for validating his behavior?

00:21:05: Yeah, that is a massive complex question.

00:21:08: That every procurement engineering and legal team Is going to have to answer very soon.

00:21:12: if you enjoyed this episode new episodes drop Every two weeks.

00:21:16: also check out our other editions on charging in battery tech commercial fleets an autonomous and mobility markets.

00:21:22: Thank You so much for joining us on the steep dive.

00:21:24: Don't forget to hit subscribe So you never miss an addition?

00:21:27: And we will see you next time.