Best of LinkedIn: Next-Gen Vehicle Intelligence CW 41/ 42

00:00:00: Brought to you by Thomas Allgaier and Frennus, this edition highlights key LinkedIn posts on NextGen vehicle intelligence in weeks forty-one and forty-two.

00:00:08: Frennus supports automotive enterprises and consultancies with market and competitive intelligence, decoding disruptive technologies, customer needs, regulatory change, and competitive moves.

00:00:19: so product teams and strategy leaders don't just react but shape the future of mobility.

00:00:25: Welcome back to the deep dive.

00:00:27: Today, we're diving into next-gen vehicle intelligence, pulling out the key insights from what we've seen discussed online over the

00:00:33: past couple of weeks.

00:00:34: Yeah, it feels like the conversation is really sharpened, doesn't it?

00:00:38: Definitely less talk about just throwing features out there and much more focus on, you know, getting the foundations right.

00:00:44: Scalable platforms for AI software, the whole cockpit experience.

00:00:47: Right.

00:00:47: So our mission here is to kind of filter through all that noise and give you the core strategic takeaways.

00:00:52: We noticed the big battles kept coming up like build versus buy and getting real about data.

00:00:58: And it's critical to get this right now.

00:01:00: The scale here is huge.

00:01:02: I saw a post from Swaraj post sale framing this, the automotive software market.

00:01:07: billion dollars last year.

00:01:09: Wow.

00:01:09: Yeah.

00:01:10: And expected to hit something like thirty two point three billion dollars by twenty thirty.

00:01:14: So if you make the wrong platform bets now, well, fixing it later is going to be incredibly expensive.

00:01:19: That really does put a fine point on it.

00:01:21: OK, let's kick off with maybe the trickiest area for a lot of companies, SDV architecture.

00:01:27: That whole build versus buy thing just won't go away.

00:01:29: Software defined vehicles.

00:01:30: Yeah.

00:01:31: SDVs basically moving intelligence into centralized code from fixed hardware.

00:01:36: But actually doing it.

00:01:38: That seems less about the tech itself and more about, well, internal politics and organization.

00:01:43: And

00:01:43: there's some pretty stark numbers floating around about this.

00:01:45: Maximilian Wenner shared some findings from Andreas Heinecki, I think it was from at ULB, twenty twenty five, about where the actual effort is going.

00:01:52: Oh yeah, that was eye-opening.

00:01:53: Get this, seventy to eighty percent, eighty percent of SDV development capacity is apparently spent on the foundational layers.

00:02:00: Right, the plumbing.

00:02:01: Connectivity, security, middleware, OTA updates.

00:02:06: basic analytics.

00:02:08: All necessary stuff.

00:02:09: But it delivers absolutely zero visible value to the customer.

00:02:13: Exactly.

00:02:13: It's not what sells the car.

00:02:14: So only what, twenty or thirty percent of all that engineering time and money goes into the actual features people see and might pay for?

00:02:22: Which begs the question.

00:02:24: If hyperscalers like AWS or Azure offer robust scalable infrastructure already, why are OEMs pouring resources into building those non differentiating bits themselves?

00:02:34: It seems massively inefficient.

00:02:36: Well, that inefficiency is forcing the rethink.

00:02:39: If you could just buy reliable infrastructure, why reinvent it?

00:02:42: Why sink billions there?

00:02:43: Okay,

00:02:43: I see the logic, but we keep hearing the tools and methods are actually ready.

00:02:47: Robert Fay noted this.

00:02:49: If the tech solution is there, why is it taking so long to change?

00:02:52: Yeah, because it's not really a tech problem.

00:02:54: It's political, deeply political.

00:02:55: Peter Wendorf had a great take on this.

00:02:57: He talks about non-decision making.

00:02:58: Non-decision making.

00:02:59: Yeah, essentially the defenders of the status quo, maybe folks rooted in traditional mechanical engineering, using bureaucracy, committees, inertia, basically anything to slow down radical change and protect the old ways.

00:03:12: Organizational drag, that makes sense.

00:03:14: Especially when speed is so crucial now.

00:03:17: So the way around that internal friction is working together.

00:03:21: Standardization.

00:03:22: Exactly.

00:03:22: Standardization is like the WD-IV for the organization here.

00:03:25: Look at the Eclipse S-Core project, Andre's Calusa Detail.

00:03:29: Right,

00:03:29: that involves BMW, Mercedes-Benz, Electro-Bit.

00:03:33: some big names.

00:03:34: Yeah, building a modular safety-focused open source foundation.

00:03:37: The idea is let's share the stuff that doesn't differentiate us, that's seventy-eighty percent, and free up our engineers to focus purely on what makes our cars unique.

00:03:45: Okay, but shifting heavily to open source, that opens up a huge can of worms around liability, doesn't it?

00:03:50: Venetians to do design brought this up.

00:03:52: If maybe seventy percent of the code is OSS and those licenses typically say no warranty, what happens when something goes wrong?

00:03:58: Who's on the hook?

00:03:59: the car maker, the tier one supplier, the open source community.

00:04:03: That's the million dollar question.

00:04:05: And the consensus Dine mentioned seems to be heading towards a hybrid model, kind of like Enterprise Linux, you know.

00:04:11: How so?

00:04:12: Instead of using raw community code directly in critical systems, OEMs would use commercially supported distributions.

00:04:19: These vendors provide a validated stack, long-term support, security updates, and crucially, contractual liability.

00:04:27: Ah,

00:04:27: so you buy the support and the liability wrapper along with the code?

00:04:31: Pretty much.

00:04:31: Yeah.

00:04:32: You design liability into the sourcing process rather than scrambling after an incident.

00:04:36: Seems like a sensible approach.

00:04:37: Okay, that architectural stuff is clearly a massive ongoing headache.

00:04:40: Let's switch perspective now to what the driver actually sees and feels.

00:04:45: Theme two, AI, HMI, and the conversational cockpit.

00:04:49: Yeah, the big shift here, as Said Farozraza put it, is thinking of AI as the new OS for the customer experience.

00:04:57: Moving away from digging through menus.

00:04:59: Right.

00:04:59: towards interactions that are more proactive, based on understanding intent.

00:05:03: The car should sort of anticipate what you need.

00:05:06: And we're seeing major OEMs jumping in with sophisticated AI pretty quickly.

00:05:10: David Richardson mentioned GM expanding its AI efforts, linking future eyes off driving goals with conversational AI using Google Gemini specifically.

00:05:19: And Mercedes just launched something significant too.

00:05:22: Steve Bosra and Magnus Osberg were talking about the new MBUX virtual assistant in the CLA.

00:05:29: This isn't just basic voice commands, right?

00:05:31: Not at all.

00:05:32: It integrates large language models, Gemini, again directly with things like Google Maps, so it can handle really complex contextual stuff.

00:05:40: Like

00:05:40: what?

00:05:40: Give me an example.

00:05:41: Like asking, find a vegan place for eight people, not more than ten minutes away from here.

00:05:46: That kind of multi-layered request.

00:05:48: And apparently they got this complex system market ready in just ten months.

00:05:52: Ten months?

00:05:53: That speed is becoming a weapon itself.

00:05:55: It

00:05:55: really is.

00:05:56: But there's another layer of complexity.

00:05:58: The ecosystem is fragmenting globally.

00:06:01: Magnus Osberg also pointed out that somewhere like China, it's a completely different digital world.

00:06:06: Right, you can't just rely on Google or WhatsApp there.

00:06:09: Exactly.

00:06:09: So Mercedes has to completely localize the cockpit stack.

00:06:13: They're using platforms like AMP and local AI like Byte Dances, Dubau, just to match how users behave there.

00:06:20: It's not just translation.

00:06:22: It's adapting the entire digital interaction.

00:06:24: A huge localization effort.

00:06:26: Absolutely.

00:06:26: And looking even further ahead, Jytho Pai mentioned agentic AI, using things like the strands agents SDK.

00:06:33: Yeah.

00:06:34: This is where the car becomes less of a single assistant and more like a team of special

00:06:39: multi-agent system.

00:06:40: Yeah, you talk to the car naturally and different agents handle safety, comfort, navigation, entertainment, all working together proactively.

00:06:47: That sounds powerful.

00:06:49: But weaving all this together, the AI, the cloud, the localization that puts immense pressure on the underlying architecture to be fast and flexible, which brings us neatly to theme three, the architecture of speed.

00:07:00: Right.

00:07:01: Two big problems to solve here.

00:07:02: First, where do you actually put the processing power?

00:07:05: And second, how do you stop software development being bottlenecked by hardware availability?

00:07:10: OK, workload placement first.

00:07:12: Shanxi made a good point.

00:07:13: Architecture first.

00:07:14: Then decide on cloud or edge, not the other way around.

00:07:18: Makes sense.

00:07:19: Because different tasks have different needs, right?

00:07:21: Real-time stuff like sensor fusion.

00:07:23: That needs instant local processing.

00:07:25: We're talking maybe twelve milliseconds latency.

00:07:27: Can't wait

00:07:27: for the cloud for that.

00:07:28: No way.

00:07:29: But.

00:07:30: Big AI model retraining, large scale analytics, that can happen in the cloud, maybe forty seven milliseconds away.

00:07:36: A competitive architecture has to nail both local immediacy and global learning.

00:07:41: OK, and the second problem, breaking the hardware dependency.

00:07:45: Virtualization.

00:07:46: Robert Fay called it a time machine, which I quite like.

00:07:49: It basically kills that dead time developers spend waiting for physical prototypes.

00:07:54: So how does that work if the hardware doesn't even exist yet?

00:07:57: You simulate it, you use virtual ECUs.

00:07:59: Sudakar Melody talked about projects integrating platforms like Siemens PAV E-Sixia with detailed hardware models like Armzina.

00:08:07: This enables what they call shift-left development.

00:08:09: Shift-left

00:08:10: testing much earlier in the cycle.

00:08:11: Exactly.

00:08:12: You can get software for complex things like lane keep assist.

00:08:15: Pretty mature before the first physical ECU is even available.

00:08:18: Teams can work in parallel.

00:08:20: It just massively cuts down time to market.

00:08:22: Got it.

00:08:23: But running all that simulation and virtualization efficiently needs a serially optimized cloud back-end, right?

00:08:28: Well,

00:08:29: absolutely.

00:08:30: Kamaslav Lazanchic shared a really impressive case study from BMW.

00:08:34: Their platform manages something like three hundred seventy-five Amazon EKS clusters, running over thirteen hundred microservices.

00:08:42: It's huge.

00:08:43: That sounds incredibly complex to manage efficiently.

00:08:45: It

00:08:45: was, but they adopted Carpenter, which is basically an intelligent autoscaler for their cloud compute resources.

00:08:53: It watches the workloads and automatically adjust the infrastructure needed.

00:08:56: Ends payoff.

00:08:57: Pretty significant.

00:08:57: Around twelve percent better CPU utilization and they saved over a million dollars a year on infrastructure costs.

00:09:04: So that back-end optimization directly saves money.

00:09:07: and boost efficiency.

00:09:08: And this kind of centralized power is being enabled by faster networking in the car too, right?

00:09:12: Stefano Tony highlighted Broadcom's thirty GBit automotive Ethernet switch.

00:09:16: That's a world away from old can buses.

00:09:19: Definitely.

00:09:20: But it's still tough going.

00:09:21: Martin Schleicher pointed out that while server zone architectures promise simplification and savings, many OEMs are still really struggling with the actual implementation.

00:09:31: It's a big shift.

00:09:32: Which brings us back down to Earth a bit.

00:09:34: All this amazing tech needs to be grounded in reality, especially safety.

00:09:39: Let's move to theme four, the safety reality check.

00:09:42: We need to talk failures, regulations, and that balance between code and machine.

00:09:47: Yeah, this is crucial.

00:09:48: Yeah.

00:09:48: Michael Ennergomez had a really strong counterpoint to the software uber Alice mindset.

00:09:53: What was his take?

00:09:55: He argued pretty forcefully that a car maker's core identity is, or should be, building great machines.

00:10:01: Software supports that, enhances it, but shouldn't try to replace fundamental mechanics entirely.

00:10:06: And trying to make everything software leads to problems.

00:10:09: His view is yes, buggy features, reliability nightmares, and it distracts companies from their core strength.

00:10:15: building safe, reliable vehicles.

00:10:17: And we're seeing real-world events and regulatory reactions that seem to back this up.

00:10:21: Yingnan Yao mentioned fatal accidents in China like that Xiaomi Su-Seven incident with the door handles.

00:10:26: Right, where they reportedly didn't release electrically.

00:10:28: Yeah.

00:10:29: And that's shifting consumer focus sharply back towards basic hardware safety.

00:10:34: It's leading to new rules mandating mechanical door releases, for example.

00:10:37: And Robert Wigmo made the point that even seemingly simple quality failures, like getting locked in or out of of a very expensive car.

00:10:45: Yeah.

00:10:46: Often that traces back to really basic process gaps.

00:10:50: Disconnected teams, requirements not properly verified.

00:10:53: It's not always some futuristic AI glitch.

00:10:55: Meanwhile, though, ADES reliability is becoming a huge competitive factor.

00:10:59: Javit Khan announced Aptis Gen-AIT radars specifically for hands-free driving in tricky situations.

00:11:05: But the real differentiator seems to be the data used to train these systems.

00:11:09: Bruno Fernandez-Ruiz introduced Nexars Baptist.

00:11:11: one point oh.

00:11:12: They achieved better incident prediction.

00:11:14: How?

00:11:15: By training on an absolutely massive data set, ten billion miles driven, sixty million unique driving events.

00:11:21: Ten billion miles.

00:11:22: That just dwarfs what you can get from simulation alone.

00:11:25: It really proves the power of large-scale real-world fleet learning for ADAS safety.

00:11:30: And it's a good reminder that even the absolute basics are critical.

00:11:34: Frank Turlop emphasized camera calibration.

00:11:36: Seems simple, right?

00:11:38: You'd think so.

00:11:39: But a misaligned camera, even a smudged lens, it can completely mess up ADS functions, false warnings, or systems just shutting down.

00:11:48: It doesn't matter how smart the AI is if the input sensor data is bad.

00:11:52: Exactly.

00:11:52: The physical machine and the human process around it, they still have to be absolutely solid.

00:11:57: So

00:11:58: let's try and pull this all together.

00:11:59: What's the big picture?

00:12:00: It feels like the industry is definitely pushing towards standard virtualized platforms.

00:12:05: They're going big on AI for personalization.

00:12:07: Yeah.

00:12:08: But at the same time, they're really wrestling with these deep questions about liability for shared code and even their core identity.

00:12:16: Are we primarily software wizards now or still master machine builders?

00:12:20: It's a real tension.

00:12:21: And this software focus is even hitting the factory floor.

00:12:24: Right.

00:12:24: Eulina JT Wozniowska.

00:12:26: talked about the shift from software-defined vehicles to software-defined manufacturing, using AI analytics to optimize production itself.

00:12:33: Okay, so here's a thought to leave everyone with.

00:12:36: If Andreas Heinke is right, and maybe seventy-eighty percent of internal SDV effort is currently spent on infrastructure that OEMs should be buying or sharing, and if the final customer experience increasingly relies on these big, collaborative, open ecosystems.

00:12:52: Then the provocative question is, when most of the underlying code is shared or bought, where does the unique, sustainable, competitive advantage for an OEM actually come from in the future?

00:13:03: What really makes you different?

00:13:05: something to think about.

00:13:06: If you enjoyed this episode, new episodes drop every two weeks.

00:13:09: Also check out our other editions on electrification and batter technology, future mobility and market evolution, and commercial fleet insights.

00:13:16: Thank you for joining us for this deep dive into next-gen vehicle intelligence.

00:13:20: Hit subscribe to ensure you don't miss our next edition.