Best of LinkedIn: Cloud Insights CW 27/ 28

00:00:00: This episode is provided by Thomas Allgaier and Frennus, based on the most relevant LinkedIn posts about cloud in calendar weeks twenty-seven and twenty eight.

00:00:08: Frenness has built a sovereign Cloud Market Reader for ICT providers with weekly hot news monthly reports quarterly leadership presentations an AI podcast for field teams.

00:00:20: you can find more info in the description.

00:00:22: so imagine realizing your cloud bill isn't You know, based on a fixed subscription you bought last year.

00:00:27: But instead it's based on what your software just decided to like.

00:00:30: think about overnight.

00:00:32: Yeah welcome to the deep dive.

00:00:33: if you're listening To this you probably already know that The infrastructure underneath our entire digital world is undergoing honestly A seismic shift right now

00:00:41: for sure and today Our mission is really just to cut through all the noise for you.

00:00:45: We're delivering the absolute top cloud trends we are seeing across LinkedIn, specifically from professionals in the ICT and tech industries.

00:00:52: Right

00:00:52: so were going look at how AI token economy has completely breaking traditional IT budgets And then will get into why the definition of a sovereign Cloud have suddenly become this massive geopolitical boardroom priority.

00:01:05: Yeah!

00:01:05: And we'll wrap up by looking out how physical deployment on new AI infrastructure physically reshaping the European landscape.

00:01:14: We'll have to cover

00:01:14: it is yeah and I mean you really can't talk about building all this new digital infrastructure until You look at how organizations are actually paying for right?

00:01:22: Yeah, because The billing models we've relied on For what a decade They fundamentally changed.

00:01:28: Oh, completely!

00:01:29: There is a massive debate happening on LinkedIn right now between site reliability engineers and phenops teams because AI is basically forcing phenops to become an absolute core enterprise discipline.

00:01:41: Exactly Like Matt Baxter brought up a brilliant point recently.

00:01:45: He said that engineering teens are still wrongly treating AI tokens like predictable sauce subscription

00:01:52: Which is just such a dangerous mindset.

00:01:53: I mean, treating AI tokens like a sauce subscription... ...is basically trying to pay for your electricity with a flat rate Netflix plan.

00:02:00: It completely ignores the underlying mechanics You know?

00:02:03: You aren't paying for access anymore.

00:02:05: you are paying for raw consumption Tokens or just raw compute.

00:02:08: Yeah and we're in this massive wave of token inflation right now Like Baxter pointed out that Anthropics' Fable model currently costs twice as much as their Opus Model.

00:02:18: Oh wow!

00:02:19: And then if look at Google their new deep thinking reasoning loops are spiking costs by like up to six times.

00:02:27: Wait,

00:02:29: Six Times?

00:02:30: Yes!

00:02:30: Let's actually unpack that for a second before you because That isn't just the slight price increase.

00:02:34: Like why does our reasoning loop burn so much more compute?

00:02:37: Well it really comes down.

00:02:39: what model is doing under the hood?

00:02:41: right In traditional query You ask question The model generates an answer and pay for those specific input and output tokens.

00:02:48: That's it.

00:02:49: Makes sense, but a deep thinking reasoning loop.

00:02:52: It operates totally differently.

00:02:53: the AI queries itself it iterates on its own logic It tests hypotheses in the background and it rejects bad answers before it finally shows you The actual result.

00:03:02: okay So all of that invisible background work just consumes massive amounts of raw compute.

00:03:06: exactly.

00:03:07: You're basically paying for the AI to talk to itself

00:03:09: which perfectly frames pre-arretti's observation.

00:03:12: I think yeah.

00:03:13: she summarized it beautifully noting that the next enterprise IT bill won't look anything like your standard cloud bill.

00:03:20: No it won't!

00:03:20: It's going to look a massive ledger of prompts, dynamic retrieval processes and autonomous agent actions that are just quietly adding up in the background.

00:03:29: Yeah so managing those token costs isn't this administrative finance job anymore?

00:03:35: if you manage an engineering budget today this is becoming your core discipline.

00:03:39: And that leads to the obvious friction point, right?

00:03:42: Like how do you manage a cost that fluctuates wildly with every single user prompt?

00:03:46: Right well Victor Garcia shared a really actionable solution for this exact problem.

00:03:51: on LinkedIn he argues that we need to apply site reliability engineering or SRE principles directly to FinOps.

00:03:59: oh interesting sre for finance.

00:04:01: yeah His point is that we have to stop making engineers do manual tagging of cloud resources.

00:04:05: We have to automate it, but more importantly he says we need to treat a sudden budget spike exactly like a live system outage.

00:04:13: Okay wait let me stop you right there.

00:04:14: How does that actually work in practice?

00:04:17: Because an outage means the website is down customers are locked-out The company is actively bleeding revenue.

00:04:24: Yeah Of course

00:04:25: A budget spike just means the finance team gets a nasty surprise at the end of month.

00:04:29: Right?

00:04:30: How do you treat those the exact same way?

00:04:32: It's about changing the culture of consequence, basically.

00:04:35: Kershia is saying that if an AI agent goes rogue over the weekend and processes say ten thousand unnecessary retrieval loops You don't just send a angry email on Monday Right!

00:04:46: ...you call the emergency break... You'd do a blameless post-mortem right then in there Just like would have the database crashed find the root cause in the prompt architecture and fix it immediately so that never happens again.

00:04:58: I see, okay yeah... That level of urgency makes total sense when costs can literally multiply by six in a matter

00:05:04: hours Exactly.

00:05:05: And its interesting because cloud providers know this is massive pain point.

00:05:10: Jonah Mengol recently highlighted that AWS actually trying to automate this exact investigative process.

00:05:16: Oh

00:05:16: really?

00:05:17: Yeah, they have a new Phenops agent in preview right now.

00:05:20: It uses natural language so teams can just ask questions about cost anomalies like They could say why did my compute bill double yesterday?

00:05:28: And I get immediate investigations and automated reports.

00:05:31: see i'm going to push back on that A bit.

00:05:33: yeah isn't it natural-language phenops agent?

00:05:36: Just Like putting a shiny new AI dashboard over an old fundamentally broken forecasting problem?

00:05:43: That's

00:05:43: the fair point.

00:05:43: Like if my underlying architecture is unpredictable, having it chatbot politely explained to me that I blew my budget doesn't actually solve my problems.

00:05:52: you know... No!

00:05:53: That is a highly valid pushback and actually academic research is backing up on that exact point.

00:05:58: Yeah Jansy Simons just published his master thesis findings on this.

00:06:02: Oh what did he find

00:06:03: He discovered while Phenops is maturing operationally, meaning you know teens have the dashboards.

00:06:09: They know how to track where the money goes.

00:06:10: It completely lacks strategic alignment with business schools.

00:06:13: Wow

00:06:14: so they just don't know why they're spending it

00:06:16: right.

00:06:16: most organizations are flying totally blind strategically?

00:06:19: They know they spend a million dollars on AI tokens But they can't tell you if had actually moved the needle

00:06:24: which leads right into The truly uncomfortable question posed by Josh Squires on LinkedIn.

00:06:29: he asks Does any of this massive AI spend actually make human employees better at their jobs?

00:06:35: And can we even measure that.

00:06:37: Yeah, That's the real question

00:06:38: because if you can't prove the value You know The fancy fine ops dashboard telling you where the money went doesn't matter it all.

00:06:44: It's the ultimate reality check.

00:06:45: exactly If you can' t measure the strategic value Of a compute...the only concrete Value you have is defensive.

00:06:53: Yes

00:06:53: and that transition perfectly explains our next major trend for This deep dive.

00:06:58: because AI compute is so incredibly strategic and wildly expensive.

00:07:03: where that data physically lives, who controls it has escalated entirely.

00:07:08: Oh

00:07:08: for sure!

00:07:09: It's no longer just a basic IT compliance checkbox...it's a geopolitical boardroom priority.

00:07:15: We are seeing digital sovereignty become a Board-level design principle

00:07:18: now.

00:07:19: Right.

00:07:19: Wanda Matasewska observed this shift really clearly.

00:07:23: She pointed out cloud dependency used to be about avoiding vendor lock-in or maybe optimizing your long term costs.

00:07:32: Right, the usual stuff.

00:07:33: But now it is fundamentally shifted to geopolitical risk management.

00:07:38: but here's the problem with that.

00:07:39: I think The terms sovereign cloud gets thrown around in so many marketing materials That its kind of starting to lose its actual technical meaning.

00:07:47: Oh a hundred percent Yeah,

00:07:49: but Alexander Walbridge broke it down brilliantly into three distinct definitions that honestly Every tech leader listening to this needs to understand.

00:07:58: Let's go through them.

00:07:58: first You have data residency.

00:08:00: that just means where the server physically sits.

00:08:03: So you put a data center in Frankfurt?

00:08:04: Yeah, of data residency most global providers offer there.

00:08:07: right

00:08:07: That's the easiest one

00:08:08: exactly yeah.

00:08:09: second you have data sovereignty.

00:08:11: This dictates whose legal jurisdiction actually applies to that data.

00:08:15: and third you have operational sovereignty.

00:08:18: And what does that mean?

00:08:19: practically

00:08:20: This means local citizens with local clearances are the ones controlling systems ensuring foreign engineers cannot access underlying infrastructure.

00:08:29: Ah,

00:08:30: got it!

00:08:31: That legal jurisdiction piece though, that is where the geopolitical tension really flares up.

00:08:36: Like in Europe The US Cloud Act Is constantly brought up In these board level discussions as this massive risk factor.

00:08:42: Well yeah it comes up all of time.

00:08:44: But

00:08:44: Chris Souter argues that the cloud act is often just used As a scare tactic by local hardware vendors.

00:08:50: He pointed out A really remarkable statistic on This.

00:08:52: What was that?

00:08:53: AWS has had zero disclosures of enterprise or government data stored outside the US to the U.S.

00:09:00: Government.

00:09:01: Wait, really?

00:09:02: Zero?

00:09:02: Zero!

00:09:03: Souter basically calls The Cloud Act a theoretical risk especially when you compare it with actual daily threats running on aging vulnerable on-premises infrastructure like we frequently see in UK's National Health Service for example.

00:09:17: I hear that argument...I do But I have a very non theoretical counter example from Andreas Welch.

00:09:23: They kind of completely flips that logic,

00:09:25: okay?

00:09:25: Let's hear it.

00:09:26: so the US government recently ordered anthropic to block foreign nationals From accessing its fable five model specifically citing national security concerns.

00:09:36: Wow

00:09:36: just like that.

00:09:37: yes think about the precedent It sets.

00:09:40: The hypothetical blocking of a foundational technology became a reality in Like A single afternoon.

00:09:47: yeah That's huge.

00:09:49: If your entire enterprise tech stack relies on a model or infrastructure governed by a foreign provider, you are just incredibly exposed to sudden geopolitical shift.

00:10:00: That is a fantastic point, actually.

00:10:02: And I think it demands really clear analogy relying strictly on data residency without operational sovereignty as well.

00:10:10: It's like renting a highly secure bank vault in your own city right?

00:10:12: Okay Yeah But the Bank Manager who lives another country can legally change the locks Without telling you

00:10:17: Oh that great analogy.

00:10:18: You

00:10:18: feel completely secure because The Vault Is just down the street.

00:10:21: yeah but you lack the ultimate control over their

00:10:23: access.

00:10:24: i love that and Paul Walker Really drove this specific concept home On LinkedIn recently.

00:10:29: he argues that true cloud sovereignty isn't about the physical metal of a server at all.

00:10:33: Right, what is it about?

00:10:35: It's defined entirely by identity governance!

00:10:39: The Identity Control Plane – your Active Directory or your Identity and Access Management System…is this system which decides who gets to see WHAT.

00:10:46: Exactly

00:10:47: So if that Identity control plane runs on infrastructure you don't operate...you've basically locked ALL THE DOORS TO YOUR BANKFAULT….but handed someone else the master key.

00:10:56: Which brings us to a really uncomfortable juncture for the European tech industry, honestly.

00:11:01: Because if operational control and identity governance are the master keys how do tech leaders actually measure this control?

00:11:08: Right And maybe more importantly can the native european cloud providers even build the locks?

00:11:15: well Let's talk about how the EU is trying to measure this first.

00:11:19: Tom Popemans and Mark Aksnich highlighted the EU's SEAL framework.

00:11:23: What's

00:11:23: a SEAL Framework?

00:11:24: It's a standardized benchmarking tool, it scores providers on scale from zero-to-four.

00:11:31: A score of zero means absolutely no sovereignty.

00:11:34: You're essentially relying on foreign public cloud with no local controls.

00:11:39: A score of four means full EU control, where the infrastructure operations and legal jurisdiction are entirely localized.

00:11:46: So it's designed to give enterprise buyers a real quantifiable metric rather than just relying on like marketing jargon.

00:11:53: Exactly

00:11:54: I mean, it is a great framework in theory but when you look closely at the software layer It exposes as pretty stark reality Federer company that's recently published some really eye-opening research on what he calls The EU cloud managed services gap.

00:12:07: Oh yeah?

00:12:08: Yeah When You Look past the raw compute and actually look At the Software Services That Developers Actually Use On A Daily Basis The Numbers Are Highly Revealing.

00:12:16: AWS offers over two hundred managed services.

00:12:20: If you look at the deepest EU native catalog, which is currently OVH cloud they have just forty five and Hetzner sits out around eight.

00:12:28: Wow!

00:12:28: Just eight?

00:12:29: Let's pause.

00:12:30: explain why that matters to if your listening maybe might not be a developer.

00:12:34: The massive gap isn't in basic storage or raw processing power.

00:12:38: No Not At All.

00:12:39: The Gap Is In Higher Level Abstractions.

00:12:42: We're talking about things like advanced observability pipelines, managed serverless databases or machine learning orchestration.

00:12:49: Right these are the tools that allow developers to deploy code quickly without having to manually build and maintain The underlying database architecture.

00:12:58: Exactly so if you don't have those manage services your engineering team just has to build them from scratch which takes thousands of hours.

00:13:05: Yeah.

00:13:06: So here's my question then If European providers are missing over a hundred and fifty of these crucial managed services, aren't enterprises just going to look at their SEAL score shrug and default right back to the global hyperscalars out of pure convenience?

00:13:20: Probably.

00:13:21: I mean speed-to-market usually wins out over compliance ideals.

00:13:24: Right

00:13:25: it does.

00:13:25: And that convenience is exactly what creates the risk here.

00:13:29: Harry Milonis actually calls this The one billion dollar sovereign compliance trap Oof

00:13:35: A Billion Dollar Trap.

00:13:36: Yeah He points out that AWS recently announced a massive initiative to deploy their own forward deployed engineers directly into enterprise IT departments, basically to help them build agentic AI solutions.

00:13:50: My Lonus warns the hyperscalers are brilliantly shifting the lock-in vector here... Well, historically providers locked you in by trapping your data.

00:14:02: Right?

00:14:02: It was just way too expensive to move petabytes of data out their cloud.

00:14:07: The egress fees were insane!

00:14:08: Exactly.

00:14:09: but now... By sending engineers to build your AI agents using there are two hundred proprietary managed services they're permanently weaving core business processes directly into the specific ecosystem.

00:14:22: it's not datalock-in anymore.

00:14:24: logic and process lock-in.

00:14:25: Wow,

00:14:26: so they write the proprietary code that just binds your business operations to their infrastructure?

00:14:30: Yep If you ever try to leave for a sovereign alternative Your entire automated workforce just breaks

00:14:36: Completely breaks.

00:14:37: That is a brilliant And truly terrifying strategy.

00:14:41: But You know despite these massive software capability gaps We are actually seeing The physical deployment of sovereign AI hardware Accelerating in an unprecedented rate right now.

00:14:52: Oh

00:14:52: absolutely This is really where the geopolitical strategy meets the physical supply chain.

00:14:57: Like Hamers and HondaGroot just announced, The deployment of the Netherlands first truly sovereign AI cloud.

00:15:03: Yeah this huge.

00:15:04: It's a massive collaboration.

00:15:05: it's between Volelt Dell North Sea data centers And its powered entirely by NVIDIA B- three hundred GPUs.

00:15:12: They are launching their first phase in Amsterdam this October.

00:15:15: Okay, but the truly staggering part of this announcement is their expansion plan right?

00:15:19: Oh for sure.

00:15:20: They are laying the groundwork for a two hundred and fifty thousand GPU AI Gigafactory in Rotterdam.

00:15:26: Two hundred fifty thousand GPUs.

00:15:28: We have to put that scale in perspective.

00:15:29: for you the power and cooling requirements For a quarter of a million advanced AI chips running simultaneously.

00:15:35: That's comparable to the energy demands of a small city.

00:15:38: Yeah, you can't just plug that into his standard local grid.

00:15:41: And this isn't even an isolated event.

00:15:43: Trisha Vangala shared another massive deployment example.

00:15:47: Paula Cloud just secured a five hundred eighty million euro deal to deploy two hundred and eighty waterless cooled micro data center units across five different European countries.

00:15:58: Wow!

00:15:59: Yeah, they're bringing twenty-nine thousand GPUs online

00:16:02: And we really should explain why that waterless, cool detail is so critical there.

00:16:06: Traditional data center air conditioning completely fails when you pack that many dense GPUs into a tight space.

00:16:12: they just run way too hot

00:16:14: Right!

00:16:14: You need advanced liquid cooling to keep the silicon from melting.

00:16:17: Exactly What PolyCloud is doing it highly strategic here.

00:16:21: They are deploying these high density micro units directly at the site of renewable energy farms specifically solve power grid bottleneck.

00:16:29: We're watching massive physical print expanding across the continent right now.

00:16:33: It's incredible to watch and we really have to remember that this physical expansion isn't just a commercial enterprise trend driven by you know chat GPT wrappers or whatever, yeah!

00:16:42: This is a defense mandate.

00:16:44: Ralph Gaidwell published his study recently showing that a staggering seventy two percent of European aerospace and defense cloud usage currently relies on US hyperscalars.

00:16:55: That's a massive vulnerability.

00:16:57: It is!

00:16:58: If you were building next-generation military tech, relying on foreign infrastructure... ...is just an unacceptable vulnerability

00:17:05: Which is exactly why we are seeing highly targeted initiatives emerge to capture that specific market.

00:17:12: Tobias Demel highlighted Oracle's new defense ecosystem for instance.

00:17:16: What's that about?

00:17:17: it specifically designed to accelerate secure completely local deployment paths for military tech and defense contractors in Europe.

00:17:25: It makes sense, it requires a complete architectural shift though like James Warnett made great observation comparing two different sovereign deployments Sharon AI Australia and the CSCS Super Computing Center Switzerland.

00:17:36: Okay what did he find?

00:17:38: He pointed out that when you were dealing with defense or highly sensitive corporate research The overarching architectural challenge isn't just about standing up raw compute anymore...the hardest questions now sit squarely at data layer.

00:17:49: Let's expand on that a bit actually.

00:17:51: When you have a massive GPU cluster, you rarely just one entity using it right?

00:17:58: Exactly!

00:17:58: You have multiple tenants different departments and contractors all sharing the exact same physical silicon.

00:18:05: The architectural nightmare is ensuring high-speed data feeding into those GPUs while maintaining incredibly strict encryption in isolation.

00:18:14: So Tenant A's AI model cannot accidentally ingest Tenant B's classified data.

00:18:20: That would be a disaster.

00:18:21: Yeah, raw compute without rigorously governed data is essentially useless for sensitive workloads

00:18:28: Which transitions us perfectly into our final theme today?

00:18:31: You have the massive Gigafactories.

00:18:33: yeah The advanced data layers but deploying all these secure sovereign GPUs really only half of battle.

00:18:39: yep

00:18:40: Once the infrastructure is live and humming, how do you actually govern the autonomous AI agents running on top of it?

00:18:46: And this is exactly where the industry's really struggling right now.

00:18:49: We aren't talking about simple scripts that automate a spreadsheet anymore.

00:18:52: we are talking about Autonomous Agents making complex multi-step decisions.

00:18:56: Fikki Makija argues that regulated modernization you know, bringing legacy industries like banking or healthcare into the AI age.

00:19:05: It actually stalls on governance not on agentic AI capabilities.

00:19:09: Oh

00:19:10: I see!

00:19:10: The cloud can be incredibly fast but regulatory audit trails and risk sign-offs are painfully slow.

00:19:17: She points out that you need incredibly clean boundaries between what an AI agent can decide on its own And what a human must explicitly approve.

00:19:30: David Linthicom outlined three crucial security shifts that tech leaders, well they just can no longer ignore them.

00:19:36: Okay what are they?

00:19:37: First zero trust has to be the absolute baseline.

00:19:40: it's not an afterthought.

00:19:41: you bolt on later right.

00:19:42: second organizations must start implementing quantum sales cryptography Right now because data intercepted today could easily be decrypted by quantum computers tomorrow.

00:19:51: Oh That's a scary thought yeah.

00:19:53: and third there is an urgent need to protect What he calls non-human identities.

00:19:57: See, that third point is fascinating to me.

00:19:59: A non-human identity is essentially the digital credential given to the autonomous AI agent acting as your digital workforce.

00:20:07: Yeah

00:20:08: Think about it If an agent spins up pulls financial data makes a purchasing decision and then just spins down in three seconds.

00:20:17: How do you authenticate?

00:20:18: Exactly.

00:20:19: These digital workers need their own dynamic, strictly governed access policies And

00:20:24: the platform vendors are scrambling to build those governance tools right now.

00:20:29: Chris Wolf detailed how VMware Cloud Foundation is evolving specifically for this reason.

00:20:34: Oh what're they doing?

00:20:35: They are building in centralized policy enforcement To provide end-to-end audit trails For agentic actions And building an audit trail for AI is incredibly difficult.

00:20:46: I can imagine!

00:20:46: It's not just logging a human clicking button, it's logging the entire dynamic retrieval path and probabilistic logic that agent used to reach its decision.

00:20:56: Yeah because if you cant prove exactly what an agent did and more importantly why they did it You cannot legally deploy it in regulated enterprise.

00:21:06: That traceability is the absolute foundation of trust in this new era.

00:21:10: I mean, if we connect everything that's discussed today from shifting token economics of phenob to geopolitical drive for sovereign data centers all the way into the governance of agentic AI We are looking at an incredibly complex highly regulated future.

00:21:26: The days of moving fast and breaking things in the cloud are completely over.

00:21:29: Visibility and control, our now just as critical a speed innovation.

00:21:33: Which brings me to final thought I want leave you with.

00:21:36: is we wrap up this deep dive?

00:21:38: We're rushing at breakneck speed.

00:21:39: build these massive sovereign AI Gigafactories across the globe right?

00:21:43: Yeah But imagine scenario for second.

00:21:46: If fully autonomous AI agent running inside perfectly sovereign completely disconnected environment hallucinates it makes multi-million euro mistake.

00:21:54: yeah

00:21:56: Who exactly holds the liability?

00:21:59: That is a question.

00:22:00: Is it the creator of the original AI model, is that sovereign cloud host providing the isolated infrastructure?

00:22:07: or does the Liability fall on the digital agent itself and a team?

00:22:11: they gave at the prompt?

00:22:13: It's very murky landscape Honestly, the rules haven't been written yet.

00:22:17: No

00:22:17: they haven't!

00:22:18: If you enjoyed this episode new episodes drop every two weeks.

00:22:21: also check out our other editions on ICT and Tech digital products & services artificial intelligence sustainability in green ICT defense tech and health.

00:22:30: Thanks so much for joining us and exploring these shifts with us today.

00:22:33: don't forget to subscribe And we'll see next time.