Best of LinkedIn: Smart Manufacturing & Industrial AI CW 33/ 34

00:00:00: Provided by Thomas Allgeier and Frennus, based on the most relevant LinkedIn posts about smart manufacturing in industrial AI in calendar weeks thirty-three and thirty four.

00:00:09: Frenness is a B to B market research company that supports smart manufacturing providers with building feature by future competitive intelligence That shows exactly how their product stacks up against the competition.

00:00:21: you can find more info in description

00:00:23: Right, so to sort of set the stage for this deep dive today we're exploring The top smart manufacturing and industrial AI trends that are just dominating the conversation across LinkedIn right now.

00:00:36: Yeah And our mission today is to strictly cut through the conceptual hype and look at the physical realities of what is actually scaling on the factory floor.

00:00:46: Yeah, exactly!

00:00:47: And to do that we need to start with this just massive paradox that's kind of staring the whole industry in the face right now.

00:00:53: because imagine spending like three million dollars on this state-of-the art predictive AI system only to find out it's completely useless.

00:01:01: because uh The Lead Engineer just prefers to run his shift using an Excel spreadsheet he built in like twenty eighteen Right.

00:01:08: We have this incredibly high ambition right now, but severely.

00:01:12: I mean severely stalled execution and Bo and Zhu actually share this really striking statistic from the SAP Manufacturing Summit over in Shanghai.

00:01:23: that perfectly illustrates this.

00:01:24: The

00:01:24: seventy-four percent one?

00:01:26: Yeah exactly.

00:01:27: Seventy four percent of organizations are successfully unlocking early AI value in pilot mode.

00:01:32: Wow!

00:01:33: But then they just hit a brick wall.

00:01:36: The execution completely freezes, they just cannot scale it across the enterprise.

00:01:40: Well and the obvious assumption there when a technology fails to scale is that the technology itself is flawed.

00:01:47: You hear these executives say you know well AI model wasn't accurate enough

00:01:51: Yeah or algorithms didn't understand our specific manufacturing process.

00:01:55: That kind of

00:01:55: thing Exactly but almost never the actual mechanism of failure.

00:01:59: They're hitting a wall because of an architectural crisis, not an algorithmic

00:02:03: one.".

00:02:03: Yeah

00:02:04: you have these brilliant predictive models generating highly accurate insights but those insights are essentially stranded.

00:02:26: AI without the digital thread is a genius with amnesia.

00:02:30: Oh, wow!

00:02:32: A genius with Amnesia?

00:02:34: I love that.

00:02:35: Yeah

00:02:35: Let's actually break down how that happens in practice, though.

00:02:37: Because it sounds like you're building this multi-million dollar state of the art race car which is...you know?

00:02:42: The AI model but your forcing to drive on a dirt road full of potholes.

00:02:47: Exactly!

00:02:48: The Dirt Road Is Your Disjointed Data Architecture.

00:02:51: You just can't expect high performance if foundation physically breaks machine and Brian Carroll offered really humorous but honestly painfully accurate reality check about what that broken foundation looks like?

00:03:03: Oh I saw this.

00:03:04: Yeah he pointed out that a true digital thread isn't finding

00:03:34: quote-unquote, go talk to Dave.

00:03:37: Your AI is doomed because a machine learning model cannot.

00:03:41: Go Talk To Dave.

00:03:42: No

00:03:42: it definitely can not grab coffee with Dave right?

00:03:45: It requires the continuous unimpeded flow of verified data.

00:03:49: if doesn't have that it suffers from that exact amnesia we just mentioned.

00:03:52: yeah it might predict a Machine failure.

00:03:55: perfectly but Because it Doesn't Have A digital Thread Connecting Into The Inventory System It has no idea that maintenance already ordered the replacement part yesterday.

00:04:04: Yeah, if your digital transformation relies on going talk to Dave you don't have a digital thread You just have a Digital scavenger hunt.

00:04:10: but let me- Let's push back On The Technology side of this for a second.

00:04:13: because Even If You Build A Perfect Pipeline Right And You Somehow Eliminate All The Excel Spreadsheets Doesn't This Still Relay Heavily On Human Operators?

00:04:24: Actually Inputting Truth?

00:04:25: Oh, absolutely.

00:04:26: And that is the hidden friction point that software vendors often just completely ignore.

00:04:31: Yeah.

00:04:32: Jacobo Lurek-Casal provided some incredible ground level insight on this.

00:04:36: he actually walked through more than fifty different manufacturing plants across Europe Wow!

00:04:41: Fifty?

00:04:42: yeah More Than Fifty.

00:04:43: and He pointed out That your factory might have a core operational problem that Absolutely No Software Architectures Ever Get A Fix Which Is a Fundamental Lack of Trust.

00:04:53: Okay, but how does a lack of trust actually break an AI model?

00:04:56: Because AI trains on historical human behavior.

00:04:59: Oh!

00:05:00: Right?!

00:05:01: Yeah, Chakubo observed the exact same patterns everywhere he went.

00:05:04: He saw an operator who had been doing specific machining process incorrectly for like three years But they actively hit it to avoid getting reprimanded.

00:05:12: Or...he saw shift supervisor, building manual workarounds.

00:05:17: They completely bypass the automated data collection sensors or a plant manager who knows that real equipment effectiveness numbers are terrible but they report a completely different padded number upstairs to the CEO just to protect their bonus.

00:05:32: Wow

00:05:33: so I mean if you feed three years of padded falsified numbers into an AI model The model is going to try and optimize the factory based on a total lie.

00:05:44: Precisely!

00:05:44: It's gonna recommend changes that just make absolutely no physical sense on the floor.

00:05:48: Exactly, you can deploy the absolute most expensive manufacturing execution system in the world but if operators don't trust this system or they're afraid it will expose their inefficiencies They'll feed garbage data.

00:06:02: And fundamentally changing how we evaluate these early projects.

00:06:06: Brent Roberts actually defined highly practical metric for success here.

00:06:11: Okay, he argues that the real test of a manufacturing AI pilot isn't whether it works once or even if it generates an immediate financial return?

00:06:19: Wait!

00:06:19: If we aren't measuring a pilot by its return on investment what are we measuring it by?

00:06:24: We're measuring its architectural momentum.

00:06:26: okay rent says The Real Test is whether the first pilot makes the next use case easier to deploy.

00:06:31: oh

00:06:32: I see.

00:06:32: yeah think about the mechanics of scaling.

00:06:34: if your required your engineering team to build this massive custom data pipeline from scratch, and then you're second project requires them to build another entirely separate Custom Pipeline From Scratch.

00:06:48: You haven't achieved scale

00:06:49: right here just starting over or...you've

00:06:51: built two isolated incredibly expensive pilot programs.

00:06:55: I see the logic there.

00:06:56: So if the foundation, like data architecture and trust of operators is getting stronger with each deployment you're actually building that enterprise momentum.

00:07:06: Exactly!

00:07:07: If final v seven dot xlsx is wrong way to operate what does it physically look when a manufacturer gets this right?

00:07:19: Because Ann Fairchild and Josh Angel recently highlighted this massive partnership between Siemens & Pringles, which was covered by the Wall Street Journal.

00:07:27: And I think that answers it perfectly.

00:07:29: Oh!

00:07:29: The Pringle's case study is just a masterclass in what happens when digital thread properly established and crucially trusted.

00:07:38: The

00:07:40: technical details here are staggering.

00:07:42: I mean, they didn't just build a dashboard.

00:07:44: They created a highly dynamic digital twin of how the potato dough physically moves through the production line.

00:07:51: right

00:07:51: and this system captures over two hundred data points per millisecond to monitor the dose consistency

00:07:58: Which is just insane?

00:07:59: Two hundred points in millisecond And it resulted in a forty plus percent return on investment.

00:08:04: Wow But the most important part is the mechanism of action.

00:08:08: Because the data was moving so fast, it allowed workers on the factory floor to make proactive adjustments to anomalies in real time before a whole batch of dough was ruined.

00:08:17: Right rather than manually testing the quality after it was already baked and packaged.

00:08:22: And that transition from merely predicting an error To actively helping a worker fix it in real-time.

00:08:29: That brings us to a broader concept gaining major traction right now which the agentic enterprise.

00:08:36: Okay, The agentic

00:08:37: Enterprise?

00:08:37: Yeah Sam Mahalingam recently broke down the functional formula for this transition and he views it as a three-part equation.

00:08:44: that I think clarifies A lot of the confusion in the market.

00:08:46: Let's

00:08:47: hear

00:08:47: First AI without context is just automation.

00:08:51: A robotic arm moving left to right is just automation.

00:08:54: Second, AI with context becomes intelligence.

00:08:58: Right.

00:08:58: So a camera telling you the robotic arm was moving too slowly.

00:09:01: that it's intelligence.

00:09:03: But third piece is critical leap Intelligence seamlessly connected to enterprise workflows creates action.

00:09:10: Let us use an analogy here actually because actions are pretty broad term.

00:09:13: Sure It sounds kind of like difference between autopilot and copilot Automation.

00:09:17: is cruise control Intelligence?

00:09:20: Maybe a weather radar telling you there is storm ahead, but the agentic enterprise as co-pilot that sees this storm automatically calculates three alternate flight path based on your current fuel load routes.

00:09:32: The best option to air traffic control and then simply asks you, the human pilot to just press a button to confirm the new heading.

00:09:39: That is a phenomenal analogy.

00:09:41: it perfectly describes how these systems actually function in the real world.

00:09:44: yeah It's not an autonomous machine acting entirely without supervision right?

00:09:48: Yeah Is an agent actively working within the system To execute really complex workflows?

00:09:54: and we see this exact mechanism happening with SAP Right now.

00:09:58: Sindhu Gangaharan shared how they're utilizing AI agents to modernize the automotive industry's ATD defect resolution process.

00:10:05: Oh, AT... Yeah for those who might not be in the automotive weeds The AT process-the eight disciplines of problem solving is notoriously manual.

00:10:13: When a defect happens engineers usually have to spend I don't know weeks digging through historical manuals maintenance logs past incident reports.

00:10:24: Just to figure out the root cause before they can even begin to propose a fix.

00:10:27: and it often completely stalls production lines while they investigate exactly.

00:10:31: but by deploying AI agents, They fundamentally alter that timeline.

00:10:35: The agent uses large language models To instantly read an understand decades of unstructured historical defect cases.

00:10:42: Wow It analyzes the root causes of similar past failures And then it proactively recommends corrective actions to the current engineering team.

00:10:51: It even goes a step further and generates the heavily formatted audit-ready reports required by original equipment manufacturers.

00:10:58: But, the crucial element there going back to our co-pilot analogy is that human engineers remain firmly in control of final decisions?

00:11:05: Yes

00:11:05: The AI does heavy lifting data retrieval synthesis sure but engineer signs off on physical fix.

00:11:12: And that shared context is the entire foundation of The Agendic Enterprise.

00:11:17: We're moving away from deploying isolated predictive algorithms in a vacuum and moving toward integrated systems where human workers and AI agents operate with shared objectives, and shared accountability.

00:11:29: You know before we move on to how this intelligence Is physically changing the hardware On the factory floor I just want to casually mention To our listeners If you are finding this deep dive valuable, make sure to subscribe so that we don't miss our future discussions into industrial tech.

00:11:45: Because the landscape is shifting incredibly fast right now.

00:11:47: It

00:11:47: really does!

00:11:48: Especially when looking at financial pressures driving these shifts...

00:11:51: Right which actually leads me on a very specific question for you… This agentic enterprise – The Pringles Digital Twin the SAP defect resolution.

00:12:00: I mean, it all sounds incredibly powerful on the factory floor but how do you actually get a conservative C-suite to buy in and write the massive checks required for this level of deep integration?

00:12:10: Well... The dynamic at the boardroom has completely changed over last twelve months.

00:12:15: Felix Belial Dockrell made very sharp observation about his reality.

00:12:19: What did he notice?

00:12:20: He pointed out that the CFOs are officially.

00:12:25: A year or two ago, a manufacturing director might have sold a small pilot program based purely on theoretical potential and you know slick presentation.

00:12:33: Yeah!

00:12:34: Nice PowerPoint Exactly.

00:12:36: But today slide decks And long-term projections simply do not survive contact with the modern manufacturing CFO.

00:12:43: So if projections don't work anymore what actual metrics are they demanding to

00:12:47: see?

00:12:48: The undeniable proof of AI's value has be drawn directly from ERP system the enterprise resource planning software.

00:12:56: That is the financial heartbeat of the company, right?

00:12:58: See if those don't care about a dashboard showing quote unquote predicted efficiency gains they want to see the physical before and after on the operations.

00:13:07: real world data.

00:13:08: oh They wanna see If The Promise Dates On Actual Purchase Orders Improved!

00:13:12: They Want To See The Physical Receipts Of Materials Saved!

00:13:16: If the AI's impact isn't definitively visible in the ERP system of record to the CFO, it simply doesn't count.

00:13:23: That is a harsh reality check!

00:13:25: If the data isn't moving the needle in the ERP you're essentially just showing this CFO of very expensive shiny new toy...

00:13:32: Exactly

00:13:33: But let's pause there because that raises really fascinating technical contradiction.

00:13:37: How so?

00:13:38: Well, we just established that the CFO demands all this data to be integrated into the central enterprise software.

00:13:45: Right?

00:13:45: But

00:13:46: physically on the factory floor AI is actually moving in the exact opposite direction.

00:13:50: it traces moving out of the centralized cloud and directly onto the heavy machinery itself.

00:13:55: We're seeing this explosive rise.

00:13:57: a physical AI

00:13:58: Yes The architectural shift toward intelligence at the edge

00:14:01: Exactly.

00:14:02: Max Wright pointed this out recently, he noted how machine vision in manufacturing is just fundamentally shifting.

00:14:08: We are moving away from centralized decision-making architectures where every image a camera takes on an assembly line Is sent to a remote server for analysis And instead we're pushing that computing intelligence directly onto the factory edge right onto the camera or the machine itself.

00:14:25: And because of this physical shift, there's a massive immediate need for implementation talent on the ground who actually understand how to install this hardware.

00:14:34: But we have to explain why this shift-to-the edge is happening right?

00:14:37: Especially when we just spent all this time emphasizing enterprise integration.

00:14:41: Right

00:14:42: Why move away from the cloud?

00:14:43: The absolute necessity here is speed.

00:14:45: Latency is the absolute enemy of manufacturing.

00:14:48: When a production line is moving at hundreds of units per minute, taking a high-resolution photo of apart sending that massive data file to a cloud server hundreds of miles away waiting for AI process it and receiving command back It simply takes too long.

00:15:03: Oh yeah By time the Cloud tells machine A part is defective Fifty more defective parts have already moved down the belt.

00:15:10: Wow!

00:15:11: And Jackie Tamm shared a brilliant example from the AWS Summit in Zurich that I think illustrates the mechanism of edge computing perfectly.

00:15:19: What was the application there?

00:15:21: Students From OST, That's Swiss University.

00:15:25: They took a standard injection molding quality control setup which is previously just running basic Python scripts and they upgraded it to Edge based AI anomaly detection.

00:15:35: So let's break down the physical difference that actually makes on the floor.

00:15:38: The differences is that the edge AI catches the faulty molded parts, the exact second.

00:15:42: they leave them old

00:15:43: instantly...

00:15:44: Instantly!

00:15:44: ...the computing power is physically sitting right next to the process.

00:15:49: it flags the structural error instantaneously rather than catching it hours later during a random batch inspection when the plastic has already cooled and you know the order supposed be shipping.

00:15:59: It all comes back to human reflexes.

00:16:00: I mean think about touching hot stove.

00:16:03: You don't want your hand to have to send an email up your arm, into your brain.

00:16:07: Wait for your brain to process the temperature and then wait for an e-mail back telling you muscles to contract.

00:16:12: Exactly!

00:16:12: By that time that round trip communication happens... ...you have a severe burn.

00:16:17: You want that localized reflex in your nervous system.

00:16:19: just pull away instantly without overthinking it.

00:16:23: That localized reflex is exactly what EDGE AI does on a factory line.

00:16:27: It reacts before damage scales.

00:16:29: That's perfect way of putting it.

00:16:31: The reflex must live at the edge.

00:16:33: And physical AI is rapidly moving past the experimental phase and into standardized deployment.

00:16:39: Yeah,

00:16:39: Andrew C highlighted that AWS has actually launched a dedicated Physical AI Solutions Library.

00:16:45: Oh really?

00:16:46: This isn't just conceptual code.

00:16:47: it's tailored specifically for robotics managing digital twins an optimizing Edge Fleet operations.

00:16:54: The major cloud providers are building this standardized architecture to push their intelligence down to the physical floor

00:17:00: Which brings us to the elephant in the room, because we're talking about reflexes at the edge.

00:17:05: Massive digital brains running agentic workflows and ERP.

00:17:09: Digital twins calculating hundreds of data points a millisecond.

00:17:13: yeah all this requires one major physical reality to exist astronomical amounts power and compute infrastructure.

00:17:21: We act like cloud is invisible but it's really just massive warehouses full incredibly PowerHungry servers.

00:17:29: It is the defining physical constraint of our era.

00:17:31: According to Dr.

00:17:32: Ferri Abul-Hasan, The concept AI sovereignty which a nation or company's ability control its own artificial intelligence isn't just about who writes best algorithms and owns proprietary data anymore.

00:17:45: it fundamentally a physical challenge for securing power, industrial cooling & physical space.

00:17:51: He shared the setup for T-Systems Industrial AI Cloud in Munich and it gives us a fascinating look at what actually takes to run these systems.

00:17:59: What are they

00:17:59: doing?

00:17:59: Instead of just pouring concrete from our massive new facility, They modernized an existing infrastructure footprint to run ten thousand NVIDIA Blackwell GPU's.

00:18:08: Ten Thousand Of The Newest Generation GPU Is Running Simultaneously.

00:18:14: The thermal output of that has to be absolutely staggering.

00:18:18: I mean, how does a mid-market manufacturer even begin to wrap their head around those kinds of resources?

00:18:23: Well they rely on these centralized hubs to provide the heavy lifting but the engineering required To keep those hubs from melting down is incredible at bet.

00:18:31: to manage the heat sustainably T systems are powering it entirely with renewable energy, and they're actually using flowing water from the nearby Iceback River for the cooling system.

00:18:40: Wait!

00:18:40: They're using riverwater?

00:18:42: Riverwater?

00:18:42: yeah Yeah.

00:18:43: And looking forward The waste heat generated by those GPUs will eventually be captured and used to heat a local city district.

00:18:50: That's

00:18:50: amazing.

00:18:51: that is what true AI sovereignty looks like in practice.

00:18:55: when physical resources are severely constrained You have to deeply integrate the data center into the physical environment

00:19:01: And we're seeing this exact type of heavy infrastructure map, expand rapidly across regions.

00:19:06: You might not immediately expect to be tech hubs either.

00:19:09: George Stritsanos highlighted the Firebird AI project currently underway in Armenia.

00:19:14: the Armedia project.

00:19:15: Yeah, this is a multi-hundred million dollar one hundred plus megawatt facility.

00:19:21: once it's completed It will be one of the world five largest AI GPU clusters and they're utilizing advanced Schneider electric infrastructure just to manage The incredibly high density power distribution And cooling required for those specific workloads.

00:19:36: It's a global build out really.

00:19:38: Yeah, Melton Chang recently toured completely prefabricated data centers that Snyder Electric is building in Spain designed to be deployed rapidly just to meet this incestible compute demand.

00:19:49: but This raises a huge red flag for me with all these massive hundred megawatt data centers popping up everywhere Aren't we?

00:19:55: Just gonna hit a hard wall at the power grid?

00:19:57: yeah You can't just plug ten thousand GPUs into the wall and expect the local utility not to brownout.

00:20:03: It's a highly valid concern, and it is bottleneck the industry is acutely aware of.

00:20:08: Jean-Christophe Moreau addressed this exact tension.

00:20:11: He argues that before utility companies just blindly start pouring billions into building new power generation plants they first need to look inward.

00:20:18: What do you mean?

00:20:19: Utilities must focus on gaining deep visibility in their existing grid capacity.

00:20:24: by applying digital intelligence essentially a digital thread for the power grid itself, utilities can identify localized constraints earlier and unlock existing capacity that is already sitting dormant in the infrastructure.

00:20:38: We have to optimize what we have before we blindly build more.

00:20:42: So the very AI systems that are demanding all this power also going be required to optimize the group to deliver it?

00:20:48: Exactly!

00:20:49: It's a closed loop...and boom, the GPUs that grid optimization.

00:20:55: The edge computing through the lens of the larger global manufacturing race?

00:20:59: The geopolitical shifts driving these investments are staggering.

00:21:03: Daniel Kepper shared some sobering statistics That really contextualize why this specific AI infrastructure Race is so critical for Western economies right now.

00:21:13: Let's hear the numbers.

00:21:15: just how drastic was the shift

00:21:17: if you look at the geopolitical math since the year two thousand The United States has lost four point five million manufacturing jobs.

00:21:25: Europe is lost for point six million.

00:21:27: Wow over that exact same two-decade time period China added fourteen point four million Manufacturing Jobs and India added nine point four Million.

00:21:36: Over nine million jobs lost in the West will nearly twenty four million were created in the East.

00:21:41: That's a staggering difference, but let me challenge the narrative here.

00:21:44: sure if we're spending all this Time talking about AI robotics an edge computing Aren't those technologies designed to automate jobs anyway?

00:21:52: Why does the West care about bringing manufacturing back if machines are just going do work.

00:21:56: That is the exact strategic pivot Maria Tupchiska points out, The response from the west isn't trying to recreate the labor intensive factories of the nineteen nineties.

00:22:05: right US manufacturing is indeed returning but it's returning in a completely different form.

00:22:11: It is shifting toward a highly localized, fiercely intelligent and high value added model.

00:22:16: The focus is no longer solely on finding the cheapest human labor.

00:22:20: it's about building highly automated resilient supply chains that cannot be disrupted by global shocks.

00:22:27: The new jobs require managing the AI not doing the manual lifting

00:22:31: And we are seeing massive undeniable capital backing up that shift in strategy.

00:22:37: Robert Little highlighted Cleveland Cliffs planned one billion dollar modernization of their Middletown Works steel plant in Ohio.

00:22:44: Right, this isn't just patching up an old furnace.

00:22:46: This includes five hundred million dollars and plan support from the Department of Energy And they are deeply integrating Palantir's AI for real-time process controls and production planning.

00:22:55: Wow They're pouring a billion dollars into a single American Steel Plant specifically to make it an AI driven facility.

00:23:02: It shows the sheer scale of the commitment to this new model And it's not isolated to heavy materials like steel either.

00:23:09: Stephanie Dorsey shared that Siemens is investing over two hundred million dollars in US manufacturing, specifically across Georgia and Texas.

00:23:18: Incredible!

00:23:19: Yeah this investment has aimed at creating jobs and rapidly expanding the critical electoral infrastructure needed to power these exact next generation AI ready data centers we've been talking about.

00:23:31: It's just this massive full-circle ecosystem.

00:23:34: You need the industrial AI to optimize manufacturing processes, but you desperately need advanced manufacturing to physically build the electrical infrastructure that powers the AI data centers.

00:23:44: Exactly!

00:23:44: We've covered an incredible amount of ground today starting from the traps of pilot purgatory and human lack of trust on factory floor diving into mechanisms for agentic enterprises and reflexes at edge all way to river cool datacenters and billion dollar strategic investments reshaping global supply chains.

00:24:02: It really highlights that industrial AI has completely outgrown its origins, it is no longer a software conversation happening in an IT department yeah...it's deeply physical structural transformation of how the world makes things.

00:24:16: and as US and European manufacturers heavily utilize autonomous AI and digital threads to rebuild local supply chains and overcome the loss of millions Will AI become a competitive wall that isolates regional industries into deeply protected silos?

00:24:34: Or will it ultimately become the new universal language, which allows global manufacturing knowledge to be shared instantly across

00:24:51: borders.