Best of LinkedIn: Cloud Insights & Sovereignty CW 39/ 40

00:00:00: Provided by Thomas Allgeier and Freeness, based on the most relevant LinkedIn posts about cloud insights in sovereignty.

00:00:06: In calendar weeks thirty-nine and forty.

00:00:09: Freenes has built a sovereign cloud market radar for ICT providers with weekly hot news monthly reports quarterly leadership presentations an AI podcasts for field teams.

00:00:19: you can find more info in the description

00:00:21: alright.

00:00:21: so let's just get right into it because the landscape is shifting

00:00:25: fast.

00:00:25: yeah exactly I mean, imagine handing a corporate credit card to a highly enthusiastic incredibly fast robot and just letting it go shopping for servers while you sleep.

00:00:37: Right which sounds like a nightmare.

00:00:38: It does but that is essentially what's happening in enterprise cloud architecture right now.

00:00:43: So if your an IT leader or tech professional listening the steep dive Your job description is fundamentally changing.

00:00:49: Today we're tracking massive shift across digital products services And really all boils down into single theme Which is control.

00:00:57: Yeah, it's basically about survival in a totally new landscape.

00:01:00: We're talking about reclaiming control over where your data physically sits raining and these exploding AI infrastructure costs And just dictating exactly Where you most critical enterprise workloads actually run

00:01:13: because organizations are finally realizing that the old model You know Just handing over the keys to a single massive public cloud provider

00:01:20: It's just no longer sustainable

00:01:21: right.

00:01:21: they want The power back.

00:01:23: In.

00:01:23: looking at the professional landscape Over the past two weeks this isn't like theoretical policy debate anymore.

00:01:29: Let's start with the foundation of that control, Sovereign Cloud.

00:01:32: I mean it feels like this went from white papers and government halls to massive real-world procurement contracts almost overnight.

00:01:38: Oh The maturity curve has been incredibly steep And we're seeing massive validation Of That.

00:01:44: There was a post by Antonin Sully Steffen.

00:01:47: He highlighted really watershed moment.

00:01:49: Basically AWS just became first cloud provider cleared for NATO restricted workloads across all thirty two member states.

00:01:56: Wait, all thirty two states?

00:01:58: I have to imagine that wasn't just a quick rubber stamp.

00:02:00: Oh not even close.

00:02:01: no this was a grueling multi-year assessment against the NATO D- Thirty Two directive.

00:02:06: it Was actually carried out by The Spanish National Cryptologic Center Wow yeah.

00:02:11: and It covers fifteen adlb wise regions with seven of those physically sitting in mainland Europe.

00:02:16: But the the architectural reality Antonin points out is what Actually matters.

00:02:21: for you listening sovereignty requires enterprise level delivery

00:02:26: Right.

00:02:26: Meaning you can't just throw a bare metal server in the local data center, slap national flag on building and call it sovereign.

00:02:32: Exactly

00:02:32: And that is the trap so many organizations fall into If a sovereign platform isn't fully managed like if its lacks native ready to use databases AI tools security monitoring automated backups all of that.

00:02:47: Then government agency or enterprise ends up having build operate patch infrastructure themselves

00:02:53: totally defeats the entire purpose of moving to the cloud in the first place.

00:02:58: You end up needing these highly specialized database and security engineers that frankly just aren't available on the market right now?

00:03:05: No, The talent shortage is real And that shifts the operational risk directly back into the agency's lap Right.

00:03:11: so projects take longer Vulnerabilities Just stay unpatched and this system Is ultimately less secure.

00:03:18: Exactly True sovereignty means retaining full legal and physical control over the data, but without sacrificing the enterprise scalability that actually makes the cloud valuable.

00:03:28: And we are seeing that exact same demand for Enterprise-grade Sovereignty really bleeding into the commercial sector too.

00:03:35: Christophe Negarier shared a massive move by Arvato Systems.

00:03:39: they just migrated a medicine verification platform directly to the Oracle EU sovereign Cloud.

00:03:45: Okay.

00:03:46: Yeah,

00:03:46: think about the stakes here.

00:03:47: this is a system serving fifteen European countries and it's relied upon by over one hundred thousand health care professionals.

00:03:54: I mean that is very definition of critical infrastructure.

00:03:57: if a system like that goes down or even just faces serious latency issues pharmacies literally cannot verify or dispense prescriptions Like supply chain for life saving medication

00:04:07: Just halts.

00:04:08: Exactly They are processing hundreds of millions of verification requests a year.

00:04:13: So they needed massive compute flexibility to scale up during peak hours, but they were legally bound to meet strict European data residency and protection requirements.

00:04:23: And you know it's not just health care feeling this pressure.

00:04:25: the financial sector is making super aggressive moves right now.

00:04:29: Oh definitely

00:04:30: Bernard Wagner posted about The DZ Bank Group...they just kicked off a pilot partnership with Stackit

00:04:36: A local european provided?

00:04:37: Yeah But I imagine a bank of that size didn't just casually sign up for a pilot on a whim.

00:04:42: No, no they spent six full months doing a rigorous bankability assessment before even starting the pilot.

00:04:49: They had to guarantee stack it could meet their extreme regulatory resilience requirements.

00:04:54: See?

00:04:54: That's six month assessment timeline is really telling.

00:04:57: banks are starting to look at sovereign cloud as Essentially in necessary hedge there modeling out scenarios involving sudden regulatory shifts or even you know geopolitical withdrawal of foreign cloud providers.

00:05:09: Like a backup plan?

00:05:11: Exactly, it's an insurance policy.

00:05:13: It ensures continuous operation regardless of what happens on the global stage.

00:05:17: Okay let me push back here though because I look at these moves and kind of wonder isn't sovereign AI or sovereign Cloud just becoming a convenient marketing label like slapping secure sticker on screen door.

00:05:31: You might buy a cloud platform hosted inside your borders, but the hardware —the silicon chips and AI models— they're all still designed and manufactured by foreign entities.

00:05:41: Right!

00:05:42: And that skepticism is entirely justified... I mean, Marin Marcus brings up this exact paradox in really sobering post….

00:05:49: He uses a harsh real-world lesson to illustrate the point.

00:05:52: looking at Ukraine's domestic flamingo cruise missile program

00:05:57: Wait how does a Cruise Missile Supply Chain connect with Enterprise Cloud Architecture?

00:06:01: So it highlights the illusion of independence.

00:06:05: When allies wouldn't supply Ukraine with sufficient long range missiles, they had to build their own right?

00:06:10: But building a missile requires incredibly complex components that they just couldn't source locally.

00:06:15: I see

00:06:16: Yeah and Marcus points out that Europe talks endlessly about digital independence from American or Chinese suppliers.

00:06:23: but true sovereignty means Building domestic capability across The entire stack.

00:06:28: you cannot Just buy the top layer software

00:06:31: Right.

00:06:31: You need domestic foundation models, domestic operating systems the silicon chips

00:06:36: The physical data centers?

00:06:38: The energy grid to power them and the engineering talent run them

00:06:41: all.

00:06:42: So you can't just like procure sovereignty off a shelf.

00:06:45: it has to be built from ground up

00:06:47: Exactly And most companies in honestly most countries cannot independently provide even a fraction of that technology stack.

00:06:56: It requires decades of painful sustained investment.

00:06:59: However we are seeing some really fascinating attempts to build sovereignty directly into the underlying code itself.

00:07:06: Really?

00:07:07: Yeah, like how well

00:07:09: look at The Post from Jordy Mikkelbrink about Kern.

00:07:13: it's a new open source European programming language currently in development and what makes it revolutionary is How It Treats Security.

00:07:20: It enforces it as compiler

00:07:22: rules

00:07:23: using capability based authority and compile time taint tracking.

00:07:27: okay hold on compile time tank tracking.

00:07:32: Yes, for those of us who don't write custom programming languages on the weekends what does that actually mean?

00:07:37: Fair question.

00:07:38: think if it like putting a permanent fluorescent dye On any piece of untrusted data.

00:07:44: The exact moment in integer application likes say users submitting a form on a website.

00:07:49: Okay

00:07:50: That data is now tainted with this die.

00:07:53: If that DieData later tries to execute a sensitive command like deleting a database.

00:08:00: The system catches the die and blocks the action before code is even allowed to run.

00:08:04: Security isn't an afterthought or some policy.

00:08:06: you check later, it's literally baked into physics of language!

00:08:11: Wow that is incredible.

00:08:13: so they are trying to build a sovereign cloud stack starting from basic language syntax all way up to container level?

00:08:21: Yeah its very early days for projects.

00:08:24: But it proves that true digital sovereignty goes so much deeper than just the geographic coordinates of a data center.

00:08:30: And, The rules governing those datacenters are about to change drastically anyway.

00:08:34: Anthony Monthe posted critical reminder.

00:08:37: January twelfth twenty-twenty seven That is day EU Data Act goes into full effect legally eliminating cloud switching and data egress charges.

00:08:46: Oh!

00:08:46: That legislation will cause an absolute earthquake in cloud architecture Because for over decade egress costs, you know the exorbitant fees providers charge to move your own data out of their cloud.

00:08:58: Yeah it's a trap!

00:08:59: Exactly they've been the primary artificial mechanism in forcing vendor lock-in.

00:09:03: so if the EU Data Act basically eliminates the cost of moving data in twenty twenty seven The market is suddenly wide open.

00:09:11: You can actually move your workloads to a sovereign provider without paying a massive financial penalty.

00:09:16: Right...the friction has gone.

00:09:17: But honestly that creates completely new arguably more dangerous problem.

00:09:22: Just because data is free to move doesn't mean it's free to process.

00:09:26: And with AI agents now entering the picture and making autonomous decisions, we are seeing infrastructure costs just spiral completely out of control

00:09:34: which brings us perfectly into our next major shift actually.

00:09:37: Fin Ops' and The Sheer Difficulty Of AI Cost Control.

00:09:41: I was reading Nicholas Fondrini breakdown this.

00:09:44: he coined a phrase that genuinely terrified me.

00:09:46: he called it the agentic cloud tax.

00:09:49: It's this rapidly compounding hidden infrastructure debt.

00:09:52: Oh yeah, divisible costs of AI like what you pay per token for the LLM to generate text?

00:09:58: That's just the tip of the iceberg right.

00:10:00: because when a user asks an AI agent to perform a task that single business request can trigger a massive chain reaction behind the scenes.

00:10:09: The Agent might autonomously query three different vector databases, spin up an API call a storage bucket.

00:10:16: It multiplies the consumption of background infrastructure.

00:10:20: it goes back to that enthusiastic intern with a corporate credit card.

00:10:23: yes oh you need a summary of last quarter sales.

00:10:26: i just spun up a dedicated database server to run the calculation.

00:10:30: like how do you stop this autonomous spending without completely killing the innovation that ai is supposed to bring?

00:10:36: well you have to automate the oversight But you absolutely have to keep a human at the final point of friction.

00:10:42: Victor Garcia broke down an incredible system from BMW called CLIA that tackles this exact issue.

00:10:48: BMW has fourteen thousand separate cloud accounts across their organization.

00:10:53: Fourteen

00:10:53: thousand?

00:10:54: I mean just the sprawl of shadow IT alone, disparate dev teams and microservices.

00:10:59: no human team could possibly monitor that manually.

00:11:03: It can't They really can't.

00:11:05: So BMW's CLIA system checks every single one of those fourteen thousand accounts, Every single day.

00:11:11: It analyzes three hundred sixty-five days of historical spend data predicts what tomorrow should cost for each specific account and then flags any anomaly.

00:11:20: The system runs its full check in twenty minutes.

00:11:23: And here is the kicker it costs them about fifty dollars a month to operate.

00:11:27: Wait wait I have to pause you there.

00:11:29: how is fifty dollars A month mathematically possible In the cloud?

00:11:32: For that level of compute?

00:11:34: because they aren't running a massive heavy server twenty-four seven just to watch other servers.

00:11:39: They are using highly optimized, serverless architecture.

00:11:43: They spin up the code from the complex math for twenty minutes and shut it all down.

00:11:48: You literally only pay for the exact seconds of compute you use.

00:11:52: but honestly The real genius at this system isn't that cheap monitoring.

00:11:55: It's filtering mechanism.

00:11:58: Right because most anomaly detection tools just scream all day long about minor fifty cents spikes and noisy services, right?

00:12:05: Exactly.

00:12:05: And that causes catastrophic alert fatigue.

00:12:08: engineering teams Just set up a rule to send those alerts straight to their email trash folder.

00:12:13: But BMW's system aggressively filters out the noise.

00:12:17: It only sends the major context-breaking overspends directly to this specific account owner and The machine never makes the final call on shutting service down.

00:12:26: Got it.

00:12:26: It just spots the pattern break and a human engineer looks at the context to decide if that spike was a mistake or Just you know, a successful new product launch.

00:12:34: so humans handle The Context machines handle the scale but there are tactical architectural ways To control these AI costs of the source too.

00:12:43: Guy Ben-Beruk shared a brilliant engineering fix for LLM routing.

00:12:47: His team set up system that intercepts prompts and routes the routine simple AI requests to smaller open source models running on Amazon SageMaker.

00:12:56: Oh, that's smart

00:12:57: Yeah!

00:12:58: And then they only send highly complex tasks to expensive frontier models like Claude on Amazon Bedrock

00:13:05: See?

00:13:05: The cost model shift there is vital to understand.

00:13:08: With a managed service like bedrock you are paying per token.

00:13:12: Every word costs money.

00:13:14: With SageMaker, you are renting the server so that you pay per instance hour regardless of how many words it generates.

00:13:21: By using continuous batching to keep that SageMakers server constantly busy they move seventy-four percent their AI requests off the per token meter.

00:13:29: Exactly!

00:13:31: The busier that server gets,the cheaper each individual token effectively becomes.

00:13:35: They modeled four point one million dollars in annual savings just by acting as a smart traffic cop for their prompts.

00:13:42: That is what happens when FinOps becomes a core engineering habit, rather than just an accounting exercise at the end of month.

00:13:49: But for workloads that absolutely must run on those per token APIs you need immediate guardrails.

00:13:56: Steph Gooch pointed out critical really common flaw in how teams manage this.

00:14:02: cloud budgets only tell you to have problem after money's already gone.

00:14:07: Yeah budget alert was just receipt from bad decision

00:14:10: literally.

00:14:12: So to catch a runaway AI agent caught in a recursive loop, you need cloud watch alarms set directly on the bedrock token metrics.

00:14:19: In real time.

00:14:21: But she gets a crucial piece of operational advice here an alarm that pages nobody.

00:14:27: It's just a blinking light on the dashboard.

00:14:29: You have to explicitly name a responder.

00:14:32: The telemetry must be connected to actual human intervention or it's completely useless.

00:14:38: So once an organization actually has those FinOps guardrails firmly in place, that's when they can start safely scaling these AI agents into production.

00:14:47: And the scale we are seeing is staggering.

00:14:51: Matt Renner posted about BNP Paraba.

00:14:53: They just signed a five-year deal with Google Cloud to deploy Gemini powered agentic AI to sixty-five thousand employees.

00:15:01: Sixty-five Thousand Users I mean, the logistics of rolling that out are massive.

00:15:06: What's crucial there is the governance.

00:15:09: every AI agent acting on behalf Of an employee is strictly authenticated only given access to specific necessary resources and all interactions with core banking systems Are continuously monitored?

00:15:20: Yeah they have to be.

00:15:22: They're Strictly enforcing their public cloud boundaries To ensure The AI doesn't you know hallucinate its way to sensitive customer data.

00:15:30: By the way, agentic AI is moving so fast right now.

00:15:32: If you want to keep tracking how these trends evolve make sure you hit subscribe So you don't miss our future deep dives.

00:15:38: Yeah The pace of development is genuinely startling.

00:15:40: We are moving past platforms that merely assist engineers straight-to-systems That act entirely autonomously.

00:15:46: Dovlett Jekisif posted about cloud genie shipping a truly autonomous cloud ops mode.

00:15:52: Autonomous meaning like absolutely zero human input

00:15:55: almost zero.

00:15:57: the agent autonomously observes the cloud infrastructure Decides on its own what compliance violation or configuration drift to fix.

00:16:03: next writes in the necessary code gathers The evidence-to-proof it works Runs, this security tests and opens.

00:16:10: The poll request the human engineers only required for the very final step reviewing the Code And clicking accept.

00:16:17: see here's where gets really interesting But also a bit terrifying from an operational standpoint.

00:16:22: are we just Click?

00:16:24: Opsing are way into an unmanageable mess.

00:16:27: Like if developers are just logging in to a web portal and manually deploying these autonomous agents, how on earth does an IT leader.

00:16:34: keep track of what systems they're running where?

00:16:37: That is the exact engineering danger Elliot Layton Woodruff addressed.

00:16:41: he issued very strict warning for the community basically stop deploying AI Agents from the Azure Portal

00:16:47: Right!

00:16:47: Click OPs.

00:16:48: Manually clicking through Web Menus To Build Infrastructure Is The Total Enemy Of Scale.

00:16:53: It creates an untrackable, unrepeatable environment.

00:16:56: I mean you would never manually configure a core production database through a web interface.

00:17:00: so why would you do it for an autonomous AI agent that has permission to alter your infrastructure?

00:17:06: Engineers absolutely must use infrastructure as code or IAC like Terraform or Bicep.

00:17:13: Definitely!

00:17:14: You have to ensure the your agents infrastructure is written in code version controlled and repository securely deployed exactly like any other critical piece of cloud architecture.

00:17:26: And because these AI workloads demand such strict governance, such intense sovereignty requirements and extreme cost predictability we are seeing a massive shift in where they're ultimately hosted.

00:17:38: the default answer is no longer just put everything into public Cloud.

00:17:42: The private and hybrid Cloud is back on the spotlight in major way.

00:17:46: Oh absolutely.

00:17:47: It is a profound shift in the hosting landscape, and it's driven entirely by economics and risk profile of AI.

00:17:55: Michael Curry sparked really fierce debate about this reality.

00:17:58: He pointed out that while VMware certainly costs more today following its recent acquisition changes than it did twenty years ago... ...it still delivers absolutely unmatched private cloud capability.

00:18:10: If an enterprise needs heavy compute Kubernetes orchestration and private AI tools all tightly integrated into one platform, he argues VMware is still the heavyweight champion.

00:18:20: Sure but the shifting economics of that exact licensing model are pushing the rest of the market to innovate rapidly.

00:18:27: Brad Parks posted a great counter perspective about HPE Morpheus.

00:18:32: They just added a native KVM based hypervisor to their platform

00:18:36: And kvm as an open source virtualization technology.

00:18:39: yes

00:18:41: By building a robust enterprise-grade control plane around native KVM, HPE can shift the licensing model entirely.

00:18:48: Instead of volatile per core licensing they offer predictable per socket pricing.

00:18:53: Let's unpack that really quickly because the hardware reality there is fascinating.

00:18:57: Modern server motherboards are packing more and more individual processing cores into a single physical socket.

00:19:04: Exactly so.

00:19:06: if your software vendor charges you per core Your licensing bill skyrockets.

00:19:10: every single time you upgrade your hardware to denser, more modern chips.

00:19:14: It literally punishes you for modernizing.

00:19:16: but by charging per socket HPE essentially ignores how dense the chip is.

00:19:21: Parks noted this pricing shift can cut virtualization costs By up to ninety percent as enterprises consolidate their compute.

00:19:29: Ninety percent

00:19:30: ninety percent

00:19:31: That's huge

00:19:32: massive.

00:19:32: it gives companies the freedom To keep there most critical workloads on VMware but migrate the rest to a highly capable, much cheaper alternative.

00:19:40: All managed under one unified hybrid

00:19:43: dashboard.".

00:19:44: And that level of flexibility is driving massive enterprise migrations?

00:19:49: Look at the sheer scale of what Alakumar posted regarding Nestle!

00:19:53: They migrated two hundred and eighty thousand employees to the SAP Sforhana private cloud... ...and they shifted over fifty-thousand users in single global wave….

00:20:04: I mean, the engineering stakes of moving fifty thousand enterprise users in a single wave are terrifying.

00:20:11: If a migration of that scale fails or if they ERP system goes down global supply chains grind to a halt factories stop shipping stops billing stops.

00:20:20: it's wild.

00:20:21: It is A phenomenal feat and They did it to balance two things that historically fight each other in large enterprises Global standardization having a single erp template so the whole company speaks The same data language.

00:20:34: and local flexibility, allowing regional markets to adjust workflows without breaking that core system.

00:20:40: But stepping back in looking at all of this the move to private AI?

00:20:43: The return on-premise hardware... This feels like a massive pendulum swing!

00:20:49: A decade ago the absolute mandate from every consulting firm was put everything into public cloud sell data centers.

00:20:56: Now it feel's as if the mandate is pull it all back onto our own basement so we can control the AI.

00:21:00: Are we just going backward...?

00:21:02: You know, from the outside it definitely looks like a retreat.

00:21:05: But Jason Reece framed this perfectly in his post.

00:21:07: It's not a retreat!

00:21:09: IT IS AN EVOLUTION.

00:21:10: The opportunity for IT leaders isn't about migrating everything to one single public cloud anymore.

00:21:16: So the hybrid reality has actually arrived?

00:21:19: It

00:21:19: really is.

00:21:19: Yeah Its' about bringing the Cloud services TO THE CUSTOMER.

00:21:23: Whether that workload sits across the Public Cloud Multi-Cloud Setup A Hybrid Environment Or Dedicated On-Premises Hardware You place the workload exactly where the architecture creates differentiated value.

00:21:34: Makes sense!

00:21:36: If a web application needs raw infinite scale to handle global traffic spikes, it goes in the public cloud.

00:21:42: if an AI model need strict data sovereignty and highly predictable per hour compute costs It goes to an air-gapped private Cloud or local sovereign provider.

00:21:51: So think

00:21:52: about the cloud architecture you are currently running at your company.

00:21:55: Your job isn't just migrating legacy servers anymore.

00:21:58: You're job is orchestrating workflows across a highly fragmented, highly specialized landscape of different clouds

00:22:05: which leads to final thought.

00:22:06: I want leave everyone with kind combining the major themes we've explored today.

00:22:11: We know that EU Data Act will kill cloud egress fees in twenty-twenty seven making it financially free move massive data sets between providers.

00:22:20: And we know from the Cloud Genie and BMW examples that autonomous AI agents are becoming highly capable of monitoring infrastructure costs in writing their own deployment code.

00:22:30: Oh wow, I see where this is going!

00:22:32: So what happens when your FinOps AI agent realizes a sovereign cloud provider just dropped their GPU pricing by twenty percent?

00:22:42: Because there were no egress fees to worry about... The agent decides on its own to seamlessly migrate your entire enterprise AI workload overnight, to save the company money.

00:22:52: That is wild!

00:22:54: If the AIs making architectural decisions writing the Terraform code and moving sensitive data across international borders entirely autonomously who's really in control then?

00:23:05: That is a wild thought.

00:23:06: to end on... The Autonomous Migration If you enjoy this episode, new episodes drop every two weeks.

00:23:12: Also check out our other editions on digital products and services, defense tech AI and agentic systems green ICT and sustainable AI and health tech.