Best of LinkedIn: Next-Gen Vehicle Intelligence CW 15/ 16

00:00:00: Brought to you by Thomas Ulgeyer and Frennus, this edition highlights key LinkedIn posts on next-gen vehicle intelligence in weeks fifteen and sixteen.

00:00:08: Frenness 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.

00:00:20: You can find more info the description.

00:00:22: So welcome, everyone to the deep dive.

00:00:24: We are constantly tracking all this shifting currents in the mobility industry.

00:00:28: and well today we're looking at a top vehicle intelligence trends surfacing across our curated sources from calendar weeks fifteen and sixteen right?

00:00:37: And Our goal Today is really just cut through The noise and figure out what it's actually changing under the hood.

00:00:42: for you

00:00:43: the listener Yeah Right now I mean automakers Are essentially promising us vehicles that think and communicate like A genius level AI.

00:00:52: But when you sort of peel back the marketing, they're trying to run that hyper-advanced software on what is basically the digital equivalent of a nineteen nineties fax machine.

00:01:01: The entire industry has obsessed with moving past the whole Software Defined Vehicle, the SDV straight into the AI defined vehicle or ADV.

00:01:11: but the sources were reviewing show this massive collision between theoretical software capabilities and brutal hardware realities.

00:01:19: It is a phenomenal bottleneck, I mean everyone wants the vehicle to be this intelligent agent.

00:01:23: but let's look at the vision first because it is undeniably compelling.

00:01:28: Indrajeet Khadke laid out the core mechanism of what an AV actually an app for everything.

00:01:38: Right, because right now it's all menus!

00:01:40: Exactly you aren't going to dig through three layers of touchscreen menus to find the Ruity Froster or like adjusts the ambient lighting.

00:01:47: instead the architecture shifts into intent-based requests.

00:01:50: So wait meaning that just tell your car what you want.

00:01:52: experience rather than pulling individual levers yourself?

00:01:55: Precisely You just tell the vehicle, adapt the cabin for comfort.

00:01:59: And then the AI has to orchestrate the climate control adjusts the seat massagers maybe dim the lighting and alter the suspension dampening all simultaneously.

00:02:09: Wow!

00:02:10: It acts as an orchestrator you know reasoning out the steps to achieve your intent.

00:02:14: Okay I see the appeal of that.

00:02:16: i really do and automakers are leaning into this heavily right now.

00:02:20: I mean Jean-Philippe Sanso and Benoit Cullen.

00:02:22: both highlighted Nissan's new long-term roadmap.

00:02:26: Oh, I saw... Yeah and Nissan is aiming to put this exact kind of AI driven intelligence at the core of ninety percent of their vehicle lineup by twenty thirty.

00:02:35: They aren't treating AI as a premium add on feature anymore.

00:02:38: they are positioning it as The foundation of the car itself.

00:02:42: yeah i saw that timeline And i think we need To take A serious step back here.

00:02:45: you Think It's too ambitious.

00:02:47: I mean, twenty-thirty is basically one single product life cycle away in automotive years.

00:02:52: When we look at what's actually rolling off the assembly lines today a lot of this ADV terminology just incredibly premature.

00:02:58: Yeah that's fair point.

00:02:59: Roller Adondo was pretty blunt about it.

00:03:02: He suggests that right now slapping AI defined on a press release is largely a marketing ploy.

00:03:07: Right, just to get funding?

00:03:08: Exactly!

00:03:09: It's a way to attract venture capital and pump up valuation.

00:03:11: Well I have to say...I agree with Eridondo there Hold on.

00:03:14: If we look at the actual physical architecture of modern car How can anyone claim it's AI-defined?

00:03:21: They really can't.

00:03:22: Thomas M points out this absolute absurdity in his post.

00:03:26: he notes that the vast majority of vehicles currently in production are still running on roughly a hundred and fifty monolithic legacy electronic control units or ECUs.

00:03:36: right

00:03:37: we're talking about cars dragging around like five thousand meters of copper wiring just to get these little isolated ships to talk to each other over legacy CAN buses,

00:03:47: And That is exactly why The twenty thirty timeline Is so suspicious.

00:03:51: Thomas notes that you simply cannot have an AI-defined vehicle without first having hyperconverged compute.

00:03:55: Wait, let's break down for the listener really quickly.

00:03:58: What exactly is Hyper Converge Compute in the context of a

00:04:01: car?

00:04:02: So think if traditional cars are having one hundred different tiny brains.

00:04:06: Okay One brain controls the windshield wipers another controls the brakes Another handles radio.

00:04:12: They're all separate low power chips.

00:04:14: Right

00:04:15: those were ECUs.

00:04:16: Exactly Hyperconverged compute means ripping all of those tiny brains out and replacing them with one or two massive data-centered grade supercomputers right in the middle of Chassis.

00:04:27: You centralize processing power because until you do that, you cannot run advanced AI.

00:04:34: It requires a massive unified pool memory and processing speed That hundred fifty scattered ECUs simply can not provide.

00:04:41: So essentially legacy OEMs are taking hardware defined car bolting a software-updateable infotainment screen into the dashboard and just calling it an SDV.

00:04:51: Basically, yeah Well meanwhile Thomas points out that Tesla is pulling in one point three billion dollars in full self driving revenue alone.

00:04:58: And they can monetize that software at scale purely because their physical architecture was actually built from the ground up with centralized compute.

00:05:05: It's all about the foundation.

00:05:06: I mean you cannot build a skyscraper on or wooden foundations.

00:05:08: That's a great way to put And trying to transition from those hundred and fifty scattered chips, to a centralized brain is causing what Hitchhom Samoody accurately describes as integration hell for European legacy automakers.

00:05:23: Integration

00:05:23: Hell.

00:05:24: let's dig into the mechanics of that because if you are an OEM You don't build the whole car yourself!

00:05:29: You buy the seats From one supplier The brakes form another The cameras from Another

00:05:33: Correct...and historically bolted those physical pieces together.

00:05:37: But in a software-defined world, performance is emergent.

00:05:40: the software has to connect everything.

00:05:42: Samoody points out that legacy OEMs are trying to build unified platforms using fifty or more different tier one suppliers.

00:05:50: and here's The Fatal Flaw.

00:05:52: Those suppliers consider their software code to be proprietary intellectual property

00:05:56: So they refuse to share it?

00:05:58: That is wild if we think about like smartphone.

00:06:01: Imagine buying a phone where the camera lens was programmed by Sony, Yeah.

00:06:05: The battery by Panasonic and screened by Samsung.

00:06:08: If they all put up proprietary firewalls And refused to let their code interact?

00:06:12: The phone would be total brick.

00:06:14: You couldn't even take photo Because the Camera app didn't tell the battery supply power.

00:06:18: That is exactly the dynamic playing out in automotive right now.

00:06:21: if your AI wants to orchestrate the cabin but the seat suppliers code won't talk to the HVAC suppliers' code because they use different operating languages.

00:06:30: And, and they refuse to open their APIs?

00:06:31: Yes!

00:06:33: Then AI is totally useless But you know... To be fair with legacy players Imran Khan brought up a really vital counter perspective.

00:06:42: What did he say?

00:06:43: He argues that European Auto isn't collapsing it's just actively restructuring.

00:06:47: The future market leader isn't necessarily going to be the company with the best stamping plant.

00:06:52: It's gonna be a company that masters this exact integration

00:06:55: capability.".

00:06:56: Well, they still know how build incredibly complex durable machines at scale and getting this integration right requires massive engineering breakthroughs on the physical side which actually brings up HManguino's post...which completely reframed how I view his entire shift.

00:07:13: Yeah!

00:07:13: That was great.

00:07:14: With all this obsession over code and algorithms, the hardware engineers are actually the unsung heroes of this entire era.

00:07:21: Absolutely because we tend to forget that software has a physical footprint.

00:07:26: Compute generates heat!

00:07:28: Right but why is it suddenly such a crisis in a car?

00:07:31: I mean engines get hot.

00:07:33: cars are used to dealing with heat.

00:07:35: Sure

00:07:35: engines get Hot But engine Heat Is Managed Mechanically.

00:07:39: We know notes.

00:07:40: When you centralize AI compute modules, when you run billions of operations per second on a single system-on ship... The thermal density is astronomical.

00:07:50: Okay, if you put a data center grade processor into a traditional plastic ECU housing in the dashboard it would literally melt itself into slag within minutes.

00:08:00: Oh my

00:08:00: god.

00:08:00: So hardware teams are being forced to solve the hardest physical thermodynamics problems in generation

00:08:05: Right.

00:08:05: so they're having to engineer vapor chambers aerospace grade printed circuit boards and advanced thermal interface materials just To keep these AI brains from frying exact.

00:08:15: Manguino also mentions the networking cables.

00:08:17: You can't run AI data over old copper wire, you need ten plus gigabit per second Ethernet links.

00:08:24: Good luck with that on a bumpy road Seriously!

00:08:26: Try keeping a delicate high-speed data connection perfectly stable across a vehicle that is vibrating violently On a pothole filled highway for fifteen years.

00:08:37: The physical constraints are staggering But we're starting to see this execution actually happen in the real world.

00:08:44: Sean Sehe shared a fantastic analogy about the Ferrari Enzo.

00:08:48: Oh yeah, I remember that!

00:08:49: When The Enzo debuted in two thousand and two... ...the public thought it looked bizarre.

00:08:52: It was aggressive..it didn't have classic elegant curves people expected from Ferrari.

00:08:57: But Sehe points out that it was designed architecture first.

00:09:00: Function and aerodynamics dictated the form.

00:09:02: So breakthrough architectures always look strange And make old guard uncomfortable before they become new industry benchmark

00:09:09: Exactly, and we are seeing that leap happen today.

00:09:13: Magnus Osberg highlighted Mercedes successfully rolling out their MBOS on the all-electric C class.

00:09:20: They're proving a scalable centralized software platform can actually work in mass production.

00:09:27: Yeah And Ralph Brandstetter showcased Volkswagen's aggressive push.

00:09:30: with China electronic architecture The CEA they are integrating onboard agentic AI right now In massive rollout.

00:09:37: It is happening.

00:09:38: But here's where the plot thickens.

00:09:40: Executing those massive centralized computing architectures requires a level of silicon and software capability that traditional automakers simply do not possess in-house.

00:09:49: No, they don't.

00:09:50: They have to partner with massive tech giants to pull this off.

00:09:53: That creates a brutal battle over who actually controls the vehicle.

00:09:57: Jack Dunkley posed the critical question regarding this ecosystem shift.

00:10:00: He asked Do OEMs want to be software companies or are they resigning themselves into just becoming integrators?

00:10:06: Which is tough pill

00:10:07: Really tough.

00:10:09: Because if the core operating system is external, If the compute platforms are consolidated around a few chip giants and middleware driven by an open ecosystem What does the automaker actually own besides the bet metal?

00:10:22: It's terrible dilemma.

00:10:23: I mean...if OEM demands complete control And tries to build the silicon and the operating systems entirely themselves it takes billions of dollars.

00:10:33: But they lean on dependencies.

00:10:35: say just buying an off-the-shelf brain from a tech giant, it's fast but the risk becoming totally commoditized.

00:10:41: And that dependency introduces a massive strategic blind spot that almost no one is talking about.

00:10:46: Seth Cronin brought this up and if you are an executive or engineer in tier I supplier listening to this right now.

00:10:52: This why your partnership agreements taking months to finalize.

00:10:55: Only IP stuff?

00:10:56: Yes

00:10:57: Car makers and chipmakers, let's use GM and NVIDIA as a hypothetical example here.

00:11:01: When they co-develop these new AI hardware architectures who actually owns the resulting intellectual property...

00:11:07: Hold on I haven't thought about that!

00:11:09: If it is a joint development agreement wouldn't they just share the IP?

00:11:12: Well in theory yes but Cronin emphasizes that these agreements are often signed quickly.

00:11:17: in the hype of big press cycle The actual IP provisions are left super vague.

00:11:22: That sounds dangerous.

00:11:24: Extremely!

00:11:25: If you co-invent a new way for an AI chip to process radar data, and don't clearly resolve who has the right to license that tech... You're setting a trap yourself.

00:11:36: What happens three years later when OEM wants update vehicles code?

00:11:40: But the chipmaker claims it updates fringes on their underlying hardware patent.

00:11:44: Oh wow… so literally end up legally paralyzed unable to update your own fleet software without getting sued by you development partner.

00:11:52: Exactly, it becomes a catastrophic licensing nightmare!

00:11:55: You have secure those rights before single line of code is written.

00:11:58: That's massive hidden risk.

00:12:01: but despite the IP minefields ecosystem expanding rapidly and moving way beyond just controlling infotainment screen.

00:12:08: Oh yeah

00:12:09: much further.

00:12:09: Brian Carlson noted that LG Energy Solution joined the SDverse consortium.

00:12:17: We are moving from AI managing your Spotify playlist to AI Managing the chemical degradation of your battery cells in real time To extend their lifespan.

00:12:25: and look at this sheer capital flowing into this expansion.

00:12:29: Spencer Collins highlighted that chip Giants arm AMD.

00:12:33: Qualcomm just poured one point two billion dollars Into wave

00:12:36: right?

00:12:37: That is a colossal bet on accelerating what they call embodied AI.

00:12:41: Okay, wait What exactly differentiates embodied AI from the AI we already know?

00:12:46: So most AI right now is disembodied.

00:12:47: It lives in a server somewhere, it processes text or images and embodied AI means giving that intelligence of physical body —in this case—a two-ton vehicle... ...and letting it interact with the physical world!

00:12:58: Oh I see….

00:12:58: …it has to actually understand mass momentum physics And human unpredictability In real space.

00:13:04: Which brings us To The Most Chilling Part Of This Entire Deep Dive.

00:13:07: We're talking about Two Ton Vehicles With Super Computer Brains Running Billions of Lines of Code constantly learning from their environments and interacting with third-party agents.

00:13:18: The security implications of that are terrifying.

00:13:21: A male kid, Hiri & Christoph Horn both pointed out a fundamental paradigm shift happening right now?

00:13:27: In the past automotive cybersecurity was about securing static systems.

00:13:31: Right you build a firewall You look for known vulnerabilities in code And catch them

00:13:36: Exactly.

00:13:37: But within AI defined vehicle you are no longer securing static code.

00:13:41: You're attempting to secure unpredictable behavior

00:13:44: because the system is reasoning, it's probabilistic not deterministic.

00:13:48: right

00:13:49: once an AI systems starts learning and deciding on the fly testing every possible outcome becomes mathematically impossible.

00:13:55: your attack service gets incredibly weird.

00:13:57: Horn and Kateri highlight threats like prompt injection attacks on voice assistants.

00:14:02: Wait,

00:14:02: how does a prompt-injection attack work in the physical car?

00:14:05: I mean... I know you can trick a chatbot on our website but what is that look like on the highway?

00:14:09: Okay imagine your driving And your vehicle Is constantly listening for your intent based commands.

00:14:16: A bad actor could theoretically embed a malicious Inaudible audio frequency into song playing on radio or encode a hidden prompt into a digital billboard.

00:14:26: you drive past.

00:14:27: You're kidding?

00:14:28: No, seriously!

00:14:29: The vehicle's cameras and microphones pick up that data... ...and the AI interprets it as command with administrative privileges tricking in to unlocking doors disabling breaks shutting down battery management system.

00:14:42: That

00:14:42: is horrifying.

00:14:43: It translates a digital logic-trick into real world physical danger.

00:14:47: Wow

00:14:48: And if behavior is that unpredictable how do you even regulate.

00:14:51: it sought to brought a product and brought up this massive crisis brewing in certification.

00:14:56: Yes, agentic AI acts proactively.

00:14:59: it can predict a hazard like a pedestrian acting erratically on the sidewalk And adjusts the steering before the human driver even registers The threat

00:15:07: which is an incredible leap forward for safety.

00:15:09: It is.

00:15:10: but Pradhan asks the ultimate question How do you certify?

00:15:14: A safety system that never stops learning?

00:15:17: well under our current regulatory frameworks You simply can't.

00:15:20: The global functional safety standard for the automotive industry is ISO, two six to six-two.

00:15:26: But that standard is built entirely on deterministic behavior.

00:15:29: meaning if X happens... ...the machine must always do

00:15:34: Y Exactly.

00:15:35: If the radar detects an object at exactly five meters, the brakes must apply with exactly this amount of force that is testable.

00:15:42: you can verify it right but if The car has a reasoning AI It might decide to break on Tuesday But on Wednesday it might decide that swerving slightly into the next lane as a mathematically safer option based On wet road conditions it just learned about.

00:15:54: oh wow ISO two six two six twos simply wasn't written for machines That think and adapt.

00:15:59: we are building probabilistic Machines And our regulatory agencies Are still using determinists.

00:16:04: We are entirely flying blind from a regulatory standpoint, and the security issues go beyond just how software behaves in traffic.

00:16:12: It extends to foundational identity of vehicle itself.

00:16:17: George Ripley noted that as vehicles become these massive lucrative digital nodes modern auto crime is shifting rapidly.

00:16:24: He pointed out huge spikes in VIN cloning and fraudulent title activity.

00:16:28: Because if you can spoof a vehicle's digital identity, You can access its data streams Its payment gateways or just sell a stolen car more easily.

00:16:37: Right And Ripley argues that Vehicle Identity needs A complete overhaul.

00:16:41: he suggests using blockchain and zero knowledge proofs To secure the identity of The Car on the Network.

00:16:48: Interesting But hold On.

00:16:49: I know Blockchain from the Crypto World but How does a zero-knowledge proof actually stop someone from cloning a VIN number?

00:16:54: Okay, so the problem currently is that when a car authenticates itself to a network say To download an update or process a toll payment it essentially broadcasts as its underlying identity data which

00:17:06: a hacker could intercept

00:17:08: Exactly.

00:17:09: A zero-knowledge proof is a cryptographic method that allows the vehicle to mathematically prove, I am an authentic registered Mercedes C class without actually revealing the underlying secret key or explicit identity data itself.

00:17:23: Because of

00:17:24: the data it's never broadcast?

00:17:26: It cannot be intercepted and cloned by a thief.

00:17:29: That makes total sense.

00:17:30: You verify truth without exposing the proof.

00:17:34: But even if we solve individual identity Merrick VanZura offered a very stark warning that scales this threat up to the geopolitical level.

00:17:42: Yeah, this part was intense!

00:17:44: Right he argues that autonomous driving and AI vehicle intelligence are not just commercial features.

00:17:49: they're national security assets.

00:17:51: This is such critical point.

00:17:52: look at the sensor suite on an ADV.

00:17:54: you have millions of vehicles equipped with high-definition cameras led R and radar.

00:17:59: yeah They are mapping our cities in real time down to the centimeter.

00:18:03: They know traffic patterns, infrastructure vulnerabilities and movement of

00:18:07: citizens.".

00:18:07: These vehicles were basically roaming surveillance machines.

00:18:10: Exactly!

00:18:11: And Vensura points out that Europe is currently failing to protect this asset by struggling with integration hell European OEMs increasingly relying on foreign technology stacks

00:18:22: so they're losing control

00:18:23: Right.

00:18:24: They are acting mostly as hardware wrappers for external intelligence, and Vensura warns that this leaves them incredibly vulnerable.

00:18:32: if the underlying data pipelines in AI brains mapping your cities or controlled by entities outside your borders you have a massive geopolitical blind spot driving down your own streets.

00:18:44: it is The Ultimate Trojan Horse but its parked millions of driveways.

00:18:47: Well, we have covered an immense amount of ground today.

00:18:50: We started with the utopian hype at the aid ID looked to unsung hardware heroes fighting physical thermal limits unraveled IP minefields deep tech partnerships and finally stared down existential safety national security threats building reasoning machines.

00:19:06: it is a complex web challenges but if we synthesize all these insights leaves us rather profound question.

00:19:13: consider as look for future As the mobility industry successfully pushes toward these true AI-defined vehicles, as our cars become these hyperintelligent reasoning nodes in a vast digital ecosystem that never stops learning predicting and optimizing.

00:19:29: At what point does the human driver sitting behind the wheel became the most unpredictable risky an outdated component in the entire vehicle architecture?

00:19:39: Wow!

00:19:40: We literally become the legacy hardware.

00:19:42: The Human is the Outdated ECU.

00:19:45: That is definitely something for everyone to think about.

00:19:47: If you enjoy this episode, new episodes drop every two weeks.

00:19:50: also check out our other editions on electrification and battery technology future mobility in market evolution And commercial fleet insights.

00:19:57: Thank You so much for joining us For This Deep Dive Into The True State Of Vehicle Intelligence.

00:20:02: We hope that gave a clearer picture of what's actually happening under the hood.

00:20:05: Don't forget to subscribe To The Feed So you don't miss next one.

00:20:08: Catch ya Next Time.