Latest / Best of LinkedIn: Digital Transformation & Tech / Best of LinkedIn: NVIDIA GTC 2026

Transcript

Best of LinkedIn: NVIDIA GTC 2026 · Best of LinkedIn: Digital Transformation & Tech

• 0:00
  Speaker 0: This episode is provided by Thomas Allgeier and Frennis, based on the most relevant LinkedIn posts about NVIDIA GTC.
• 0:06
  Speaker 0: twenty-twenty six.
• 0:07
  Speaker 0: Frennes supports enterprises with market and competitive intelligence decoding emerging technologies customer insights regulatory shifts in competitor strategies.
• 0:17
  Speaker 0: so product teams and strategy leaders don't just react but shake the future of AI
• 0:21
  Speaker 1: And man, shaping the future of AI is exactly what we need to talk about today.
• 0:26
  Speaker 1: Because the sheer scale of what's happening right now it was kind of hard to comprehend.
• 0:30
  Speaker 0: Yeah!
• 0:30
  Speaker 0: It really is and that's our mission for you.
• 0:33
  Speaker 0: in this deep dive We are stepping into a high level strategy room basically unpacking absolute top trends & insights from NVIDIA GTC And we're curating this specifically from the feeds of ICT and tech industry professionals over on LinkedIn.
• 0:49
  Speaker 0: So no fluff, just a real signal from the noise
• 0:51
  Speaker 1: Exactly!
• 0:52
  Speaker 1: If you are listening to this You really need understand why The fundamental unit value in Tech is shifting right beneath our feet.
• 1:00
  Speaker 1: We literally can't discuss future AI software or robotics without first looking at hardware layer
• 1:06
  Speaker 0: Because it's a tectonic shift.
• 1:08
  Speaker 0: I mean, just imagine standing in front of the demand pipeline so massive that it fundamentally breaks how we measure industrial output.
• 1:15
  Speaker 1: Yeah one trillion dollars right?
• 1:18
  Speaker 1: That is the projected order pipeline for Nvidia new Blackwell and the Vera Rupin platforms through twenty-twenty seven Which
• 1:24
  Speaker 0: Is Just A Staggering Number To Try And Wrap Your Head Around.
• 1:27
  Speaker 1: It Is David Daniszewski In Charles K. We're Actually Highlighting This On Their Feeds This Week.
• 1:32
  Speaker 1: Its Just One Trillion Dollars In Pure Raw infrastructure demand.
• 1:36
  Speaker 1: Wow,
• 1:37
  Speaker 0: okay.
• 1:37
  Speaker 0: so why the sudden massive explosion in infrastructure?
• 1:42
  Speaker 1: Well to understand that we have to look at what Karen Yu and McHugh Suley are calling the inference inflection
• 1:48
  Speaker 0: The afference inflection.
• 1:48
  Speaker 1: Yeah.
• 1:49
  Speaker 1: So for years the entire industry was just obsessed with the training phase right like pouring oceans of data into these massive GPU clusters Just to teach the AI how-to thing.
• 1:58
  Speaker 0: Right!
• 1:59
  Speaker 0: The learning phase
• 2:00
  Speaker 1: Exactly.
• 2:01
  Speaker 1: But that phases well it's maturing.
• 2:03
  Speaker 1: We've officially moved from the era of training models to the era executing them at scale.
• 2:08
  Speaker 0: Which completely changes the physical footprint of whole tech industry, you know when we historically thought about industrial revolutions?
• 2:16
  Speaker 0: We always pictured smokestacks right assembly lines stamping metal into car doors.
• 2:21
  Speaker 0: yeah very physical exactly.
• 2:23
  Speaker 0: You had a clear physical input and a physical output.
• 2:27
  Speaker 0: And then came The Digital Age and we built these traditional data centers.
• 2:32
  Speaker 0: But looking at what Laura Sophie Bother, Adrian McDonald and Santosh Pandey pointed out from the GTC floor those traditional data centers are kind of officially dead.
• 2:41
  Speaker 1: They're completely dead.
• 2:42
  Speaker 1: I mean a traditional data center was essentially just a massive barking garage.
• 2:46
  Speaker 1: your data Just sat there waiting to be retrieved.
• 2:49
  Speaker 0: right?
• 2:49
  Speaker 0: it was passive
• 2:50
  Speaker 1: totally passive.
• 2:51
  Speaker 1: but The new model these AI factories that Jensen Wong describes they don't store data.
• 2:56
  Speaker 1: They actually manufacture intelligence.
• 2:57
  Speaker 1: Oh
• 2:58
  Speaker 0: like.
• 2:59
  Speaker 0: So if the old data centers were parking garages, these new AI factories are continuous assembly lines.
• 3:04
  Speaker 0: Yes The smokestacks are now liquid cooling towers...the raw material is electricity and they actual widgets rolling off the belt or tokens
• 3:12
  Speaker 1: Exactly!
• 3:13
  Speaker 1: And whole metric of economic value has shifted.
• 3:15
  Speaker 1: We aren't talking about gigabytes anymore.
• 3:17
  Speaker 1: The metrics that matter now Are tokens per watt and costs per token.
• 3:21
  Speaker 0: But manufacturing those tokens so incredibly fast creates a physical bottleneck, right?
• 3:26
  Speaker 0: A massive
• 3:26
  Speaker 1: one because when a large language model is generating an answer spitting out words token by token it absolutely starves for memory bandwidth.
• 3:36
  Speaker 0: Okay let me see if I can break this down mechanically for a second.
• 3:39
  Speaker 0: If you are spitting up tiny pieces of information sequentially standard memory which usually grabs large heavy chunks of data all at once kind of slowly that would create a massive traffic jam.
• 3:51
  Speaker 0: You need a system that can continuously fire off tiny data bits at lightning speed.
• 3:56
  Speaker 1: That is the exact mechanical problem that NVIDIA solved, and honestly it's why their twenty billion dollar acquisition of GROC was basically the technical unlock of the entire conference.
• 4:07
  Speaker 0: Oh wow!
• 4:07
  Speaker 1: Yeah Anastasia Nisova and Padma Pavan Venomatla really detailed this on LinkedIn.
• 4:12
  Speaker 1: they explained how NVIDia integrated GROCs LPU technology to basically disaggregate the entire inference process
• 4:19
  Speaker 0: meaning they split it up to bypass those traditional bandwidth ceilings.
• 4:23
  Speaker 1: Precisely, yeah.
• 4:25
  Speaker 1: so NVIDIA handles the heavy compute which is called a pre-fill phase.
• 4:28
  Speaker 1: okay that's where it digests your massive complex prompt and does the deep mathematical lifting but once it understands the request It hands the task over to grokks
• 4:39
  Speaker 0: lpu Which handles high speed decode
• 4:42
  Speaker 1: Exactly because it's an entirely different architecture that just fires out those tokens without hitting a wall.
• 4:47
  Speaker 0: It's basically like having an industrial crane lift A massive boulder of raw material onto the factory floor and then instantly dropping it on to high-speed Frictionless conveyor belt, which shoots the finished widgets Out The Door.
• 5:00
  Speaker 1: That is great way to picture.
• 5:01
  Speaker 1: And that manufacturing of Tokens connects directly To what they are actually doing Because all this massive inference power Is fueling our transition from generative AI chatbots to agentic AI.
• 5:14
  Speaker 0: Right, digital workers!
• 5:15
  Speaker 1: Exactly we're seeing Open Claw having its Linux moment.
• 5:19
  Speaker 1: Jensen Wong actually stated that every company needs an open-claw strategy now because agents are becoming the new operating system for the enterprise.
• 5:28
  Speaker 0: Yeah Ronald Van Loon Michael Strickland and Stephanie Goodos were all discussing that shift.
• 5:32
  Speaker 0: it's moving from sauce to gas right?
• 5:34
  Speaker 0: Yes
• 5:35
  Speaker 1: Irithyth Kumar SS and Julian Medonca highlighted this perfectly.
• 5:39
  Speaker 1: It's the shift from software as a service to generative or egentic AI, as a Service
• 5:47
  Speaker 0: where instead of you clicking the buttons and pulling the levers You just define The goal in the AI agent executes all the steps
• 5:52
  Speaker 1: entirely In the background.
• 5:54
  Speaker 0: okay But I have To push back on This enthusiasm for fully Autonomous Agents For A second.
• 5:58
  Speaker 0: oh sure because Vivek Tandipani brought up a really good point about this.
• 6:01
  Speaker 0: If these agents are operating in multi-step execution loops, where a tiny error on like step two can compound massively before human ever detects it by step twenty.
• 6:13
  Speaker 0: Aren't we building massive terrifying control problem?
• 6:16
  Speaker 1: Oh!
• 6:16
  Speaker 1: It's huge concern.
• 6:17
  Speaker 0: Like how do govern something that executes faster than human oversight?
• 6:21
  Speaker 1: Well That is the exact tension enterprise market wrestling with right now.
• 6:25
  Speaker 1: Because you're Right Agents have dangerous freedoms.
• 6:28
  Speaker 1: They can run code, they access payroll files and send emails
• 6:31
  Speaker 0: which is a compliance nightmare
• 6:33
  Speaker 1: A total nightmare.
• 6:34
  Speaker 1: And that's why the enterprise security layer NEMA CLAW was such critical announcement.
• 6:38
  Speaker 1: Premdurendes Axel Ditman and Sanjay Basu were all over this on LinkedIn.
• 6:43
  Speaker 0: So what does NEMA Claw actually do to fix it?
• 6:46
  Speaker 1: It provides essential out of process sandboxing policy routing privacy guardrails.
• 6:52
  Speaker 1: Prem Noreen has actually noted that you can't just throw an agent in a traditional Docker container and call it safe.
• 6:58
  Speaker 0: Right, because the container protects the computer but doesn't govern what the agent decides to do inside of the network.
• 7:04
  Speaker 1: Exactly!
• 7:05
  Speaker 1: NemoClaw acts as external enforcement engine so enterprises can deploy these agents safely without them going rogue.
• 7:12
  Speaker 0: Even with security layers solved though there's still massive UX reality check here.
• 7:17
  Speaker 0: Valias Cajal noted something interesting about this.
• 7:19
  Speaker 0: What
• 7:19
  Speaker 1: did he say?
• 7:20
  Speaker 0: He pointed out While capabilities might get agents deployed in an enterprise, it's personality and trust that keep them used.
• 7:27
  Speaker 1: Oh!
• 7:27
  Speaker 1: That's incredibly true
• 7:28
  Speaker 0: because most agent demos right now are basically just command lines dressed up as chat interfaces.
• 7:35
  Speaker 0: if the digital worker doesn't have a bedside manner humans just reject It.
• 7:40
  Speaker 1: yeah.
• 7:40
  Speaker 1: And that trust becomes even more critical when we start talking about these autonomous systems stepping out of the web browser and interacting with The physical world.
• 7:48
  Speaker 0: okay yes Because building on the idea of agents executing digital tasks, what happens when they actually enter our physical space?
• 7:56
  Speaker 1: Well, Physical AI has officially arrived.
• 7:59
  Speaker 1: Pascal Bornet and Nidhi Warankar were highlighting Disney's Olaf robot... Did you see this?
• 8:04
  Speaker 0: I did!
• 8:05
  Speaker 0: The little walking robot
• 8:06
  Speaker 0: right?!
• 8:06
  Speaker 0: Yeah
• 8:07
  Speaker 1: but the crazy thing is it wasn't manually animated.
• 8:10
  Speaker 1: It was trained in a physics simulation first
• 8:13
  Speaker 0: Using Nvidia's Isaac Sim and Newton Physics engine.
• 8:17
  Speaker 0: Exactly
• 8:18
  Speaker 1: It had to literally learn how to balance and walk in a digital twin before they ever brought it into life,
• 8:24
  Speaker 0: which is mind-blowing.
• 8:25
  Speaker 0: But Tom Emmerich had a really grounded take on this.
• 8:28
  Speaker 0: He called it robot theater.
• 8:29
  Speaker 0: Robot
• 8:29
  Speaker 1: Theater?
• 8:30
  Speaker 0: Yeah, he was discussing how these cute entertainment focused robots are necessary stepping stone for marketing and consumer acceptance.
• 8:38
  Speaker 1: Oh I see
• 8:38
  Speaker 0: like before we see wide-scale industrial deployment of humanoid robots in warehouses.
• 8:44
  Speaker 0: We need the public to get comfortable with them.
• 8:46
  Speaker 1: that makes total sense yeah.
• 8:48
  Speaker 1: But to actually support that wide-scale deployment at the edge, you need critical infrastructure.
• 8:54
  Speaker 1: And that is where telecom is radically evolving.
• 8:56
  Speaker 0: Yeah Jordan Barr and CeCe Chong were discussing how cell towers are becoming robotic AI radios.
• 9:02
  Speaker 0: Right
• 9:02
  Speaker 1: creating an AI grid.
• 9:04
  Speaker 1: Telcos are totally transforming from just selling bandwidth dollars per gigabyte distributed AI compute grids that sell inference.
• 9:14
  Speaker 0: So if we draw a biological analogy here for the listener... Oh, let's hear it!
• 9:18
  Speaker 0: If the AI factory we talked about earlier is The Brain and the Agentec software is The Mind then the telecom-AIRN networks are basically becoming the nervous system.
• 9:28
  Speaker 1: Oh, that's perfect!
• 9:29
  Speaker 0: Instantly transmitting reflexes down to the muscles which are the physical AI robots at the edge?
• 9:34
  Speaker 0: Yes
• 9:35
  Speaker 1: because as Guy in Correlage provided AI isn't going to break telecom networks through bandwidth.
• 9:42
  Speaker 1: Tokens are actually tiny, eta-wise...
• 9:44
  Speaker 0: Right
• 9:45
  Speaker 1: But what it will stress is latency the uplink and this constantly always on behavior.
• 9:50
  Speaker 1: You need that nervous system to be lightning fast.
• 9:53
  Speaker 0: but synthesizing all of this you can design The best ai grid in the most advanced robots in the world?
• 9:58
  Speaker 0: They require massive amounts power right?
• 10:00
  Speaker 1: yeah And That Is the absolute bottleneck Of This Intelligence explosion.
• 10:05
  Speaker 1: Yeah, Maria Cortez noted that while AI is scaling its silicon speed power Is only scaling at grid speed
• 10:12
  Speaker 0: and grid speed as agonizingly slow
• 10:14
  Speaker 1: Extremely slow which is why some players are just bypassing the traditional grid entirely.
• 10:18
  Speaker 1: Daniel Shapiro shared how knee scale was building an eight gigawatt Islanded a i microgrid in West Virginia.
• 10:24
  Speaker 0: eight gigawatts An islanded microgrid so they aren't even relying on the public utility.
• 10:29
  Speaker 1: Nope entirely self-sufficient power Just to feed the compute.
• 10:33
  Speaker 1: And to manage that scale, Schneider Electric is actually developing digital twins of these gigawatt-scale factories.
• 10:40
  Speaker 1: Wow!
• 10:41
  Speaker 1: Deepak Sharma and Manish Gokal explained how they simulate the cooling and power dynamics in software before they even pour the concrete.
• 10:50
  Speaker 0: So we are simulating the power plants that will run the simulated factories that train the simulated robot.
• 10:56
  Speaker 0: Exactly
• 10:56
  Speaker 1: it's turtles all the way down.
• 10:59
  Speaker 1: But you know amidst this infrastructure talk There is a massive human element here.
• 11:04
  Speaker 1: Oh, for sure!
• 11:04
  Speaker 1: Stephanie Godos pointed out huge shift happening in Silicon Valley.
• 11:08
  Speaker 1: Employees are starting to negotiate token budgets instead of just salaries.
• 11:12
  Speaker 0: Wait really?
• 11:12
  Speaker 0: Token Budgets Yes
• 11:14
  Speaker 1: Because if you have a massive compute budget You can amplify your own output ten X and become one person department.
• 11:20
  Speaker 0: So completely changes.
• 11:21
  Speaker 0: what makes an employee valuable?
• 11:23
  Speaker 0: Who's Zhang actually articulated this beautifully?
• 11:25
  Speaker 0: What
• 11:25
  Speaker 1: do ya say?
• 11:26
  Speaker 0: He said that because AI amplifies technical execution so well, the true hard skills of the AI era are now totally intangible.
• 11:33
  Speaker 0: Oh like what?
• 11:34
  Speaker 0: Like curiosity having a moral compass and the courage to
• 11:37
  Speaker 1: act.
• 11:38
  Speaker 1: Wow That's powerful.
• 11:40
  Speaker 0: So I really want to pose this directly to you The listener think about your own career for a second.
• 11:46
  Speaker 0: If AI democratizes technical execution and levels the playing field of hard knowledge, does that mean your most valuable professional asset?
• 11:54
  Speaker 0: moving forward is simply your judgment?
• 11:56
  Speaker 0: Your character.
• 11:57
  Speaker 0: And you're humanity?
• 11:59
  Speaker 1: it absolutely doesn't and we need that humanity in the room which brings up a really sobering reality check That Anita Pandy delivered from the GTC floor.
• 12:08
  Speaker 0: right diversity warning.
• 12:09
  Speaker 1: exactly Amidst all these technological triumphs, she noted that fewer women are present in critical rooms shaping the future.
• 12:17
  Speaker 0: which signals a massive systemic failure.
• 12:19
  Speaker 0: Yes,
• 12:20
  Speaker 1: a failure to retain the brilliant minds that we desperately need to guide this technology.
• 12:25
  Speaker 1: We need diverse visionary leadership if you're going to navigate it safely.
• 12:28
  Speaker 1: That
• 12:29
  Speaker 0: is such critical point of land on!
• 12:30
  Speaker 0: We've covered so much ground today from AI factories manufacturing tokens to agentic digital workers to physical robots and eight gigawatt micro grids.
• 12:39
  Speaker 1: It's completely new world.
• 12:41
  Speaker 0: It really is.
• 12:42
  Speaker 0: And as we wrap up, I want to leave you with a final provocative thought to mull over on your own... Let's
• 12:47
  Speaker 1: hear it!
• 12:47
  Speaker 0: ...as we shift to an era where intelligence has continuously manufactured in these AI factories What happens when the cost of generating a brilliant answer becomes cheaper than electricity required power-the-server?
• 12:59
  Speaker 0: Oh wow In a world of infinite answers The only true scarcity left will be ability to ask right questions.
• 13:06
  Speaker 1: That Is A Wild Thought To End On.
• 13:08
  Speaker 1: If you enjoyed this episode, new episodes drop every two weeks.
• 13:12
  Speaker 1: Also check out our other editions on ICT and Tech Digital Products & Services Artificial Intelligence Cloud Sustainability in Green ICT DefenseTech and HealthTech.
• 13:22
  Speaker 1: Keep
• 13:23
  Speaker 0: asking those right questions.
• 13:24
  Speaker 0: Thank You so much for joining us And don't forget to subscribe.