Best of LinkedIn: Cloud Insights CW 15/ 16

00:00:00: This episode is provided by Thomas Allgeier and Frennis, based on the most relevant LinkedIn posts about cloud in calendar weeks fifteen and sixteen.

00:00:07: Frennes has built a sovereign cloud market radar for ICT providers with weekly hot news monthly reports quarterly leadership presentations an AI podcast for field teams.

00:00:17: you can find more info in the description.

00:00:20: so um i want you to imagine something for second.

00:00:23: imagine a hundred billion dollars building your own stay-of-the-art data centers.

00:00:29: Oh, wow!

00:00:30: Okay Yeah right you lay your own fiber optic cables across the ocean.

00:00:33: You control your entire hardware supply chain.

00:00:35: all of that only to realize you still have to hand over thirty five billion dollars To a third party cloud provider just to survive The next few

00:00:43: years.

00:00:43: I mean That is just wild.

00:00:44: it Is and that is exactly what meta is doing Right now.

00:00:47: so welcome back to our deep dive.

00:00:49: Today, we're looking at the absolute top cloud trends lighting up the feeds of ICT and tech professionals over the last two weeks.

00:00:56: And there's just a massive shift happening.

00:00:58: Definitely a shift.

00:00:59: Yeah Looking at all these posts...I'm noticing a real change in tone.

00:01:03: It feels like The Wild West era Of you know.

00:01:05: Just put everything into the Cloud.

00:01:07: We'll figure it out later.

00:01:08: That is completely over.

00:01:09: People seem almost anxious.

00:01:12: They are!

00:01:15: For good reason.

00:01:16: The narrative has completely shifted.

00:01:17: It used to be like how fast can we migrate and now it's a, How tightly can we control what?

00:01:22: We've actually built right looking at the landscape where the underlying architecture of the internet is basically being fundamentally rewired.

00:01:29: Yeah So from the post in weeks fifteen and sixteen I've broken this down into three big themes for us to tackle.

00:01:35: i love A good structure.

00:01:36: What if we got

00:01:37: first Sovereign Cloud Momentum, which is moving from just abstract compliance to concrete procurement.

00:01:44: Then the Phenops evolution where we're shifting from cost cutting to strategic investment.

00:01:49: and finally AI cloud convergence like how AI is fundamentally rewiring cloud architecture in talent?

00:01:55: Okay let's unpack this because I agree that era of clouds for the sake of cloud it's done.

00:02:01: We are entering an Era of strict intentional control.

00:02:05: so lets start with first theme, that geopolitical pressure because the concept of a sovereign cloud is I mean it's everywhere in these notes.

00:02:14: Oh absolutely!

00:02:14: Everywhere

00:02:15: for a long time digital sovereignty just felt like a legal buzzword right?

00:02:19: Like something compliance teams argued about in a basement somewhere while The Engineers kept building on AWS.

00:02:27: but James Patrick posted something that framed this perfectly.

00:02:30: He described Sovereign Cloud not as some, you know political slogan but as the floor under everything else for critical infrastructure.

00:02:38: I really like that phrasing The Floor Under Everything.

00:02:40: Exactly You think about hospitals energy grids national defense.

00:02:44: They cannot sit on digital foundations governed by laws That aren't their own

00:02:48: Right and the scale of vulnerability there is staggering once you actually map it out.

00:02:52: Yeah Paula Felcon shared a data point.

00:02:55: Honestly, you should keep every European CTO awake at night.

00:02:58: Oh boy!

00:02:59: What was it?

00:03:00: He noted that eighty-five percent of Europe's cloud infrastructure is currently controlled by U.S.

00:03:05: hyperscalars.

00:03:06: Eighty five

00:03:06: percent?!

00:03:08: I mean...that basically all of it.

00:03:09: Basically yeah.

00:03:10: and just to clarify for anyone listening outside the infrastructure bubble when we say hyper scalars We're talking about the massive tech giants right Amazon Web Services Google Cloud Microsoft Azure The

00:03:22: big three

00:03:23: Exactly.

00:03:23: They operate at a scale that completely dwarfs traditional data centers.

00:03:29: Falcone points out, they're relying heavily on these few American companies.

00:03:33: it creates an architectural single point of geopolitical failure.

00:03:36: Wow!

00:03:37: A single Point Of Geopolitical Failure.

00:03:39: Yeah

00:03:39: and he cited a very visceral example He brought up when Iranian drones hit EWS Data Centers in the Gulf.

00:03:46: Now Europe's digital economy didn't burn that day obviously.

00:03:49: But as Falcom put it, Europe essentially watched its landlords building catch fire and realized um...it had absolutely no other place to live.

00:03:58: Wait I have to push back there or stop you there because my immediate reaction If relying on a foreign hyperscaler is such a massive risk, why don't these European hospitals and energy grids just you know pull everything back on premise?

00:04:12: Like build their own walls.

00:04:13: Yeah!

00:04:14: Buy your physical servers put them in secure building in Berlin or Paris.

00:04:19: lock the doors call it today.

00:04:21: doesn't that solve it?

00:04:22: It sounds like logical move.

00:04:24: sure but Benjamin Herman completely dismantles this argument.

00:04:29: He points out that retreating to on-premise solutions does not magically grant you independence.

00:04:34: It doesn't!

00:04:35: No, look at what just happened with Broadcom's acquisition of VMware.

00:04:39: Oh

00:04:40: right the massive price hikes.

00:04:41: I saw headlines about companies facing What a thousand percent increase in their licensing costs overnight?

00:04:48: Yep A thousand percent In some cases.

00:04:49: But exactly how does it happen if You own The physical servers like its your metal

00:04:54: because you might own the metal but you don't own the software That makes them metal useful.

00:04:58: VMware makes hypervisors.

00:05:00: That's essentially the software layer that lets you chop up one big physical server into dozens of smaller, highly efficient virtual machines.

00:05:09: Over last decade companies built their entire internal networking Their security protocols and storage routines all around VMWare's proprietary code.

00:05:18: So they're totally locked in.

00:05:19: Exactly.

00:05:20: They stayed on-premise specifically for independence but just traded one dependency to another.

00:05:25: When Broadcom bought VMware and changed the pricing model, these companies couldn't just leave.

00:05:31: Ripping out a hypervisor ecosystem?

00:05:33: I mean that takes years of engineering.

00:05:35: Yeah you can't just flip a switch Right.

00:05:37: You pick a vendor with zero competition And zero exit options...you're tracked.

00:05:42: Herman argues true sovereignty isn't about where your physical servers live.

00:05:46: Geography is almost irrelevant if you are locked in.

00:05:48: Wow Sovereignty is entirely about your ability to exit the room.

00:05:53: He says if your exit strategy takes longer than ninety days, you weren't independent.

00:05:57: That is a massive paradigm shift.

00:05:59: independence Is the ability to leave.

00:06:01: so geography alone?

00:06:02: It's just a trap.

00:06:03: You could have a server in your own city But if a foreign entity Holds the software licenses or the encryption keys Or can just financially squeeze you you aren't sovereign at all.

00:06:14: What's fascinating here is how Rick Godes framework ties into this.

00:06:18: he argues that sovereignty has become bloated subjective political term, and we really need to strip it down of the studs.

00:06:27: Strip

00:06:27: it down how?

00:06:27: He says sovereignty should only be defined by two strict things.

00:06:31: first exclusive data access Meaning no one else can touch your data period

00:06:36: okay makes sense.

00:06:37: And second service continuity meaning No One Else Can Interrupt Your Service.

00:06:41: that's It.

00:06:42: That Is The Entire Checklist.

00:06:43: Its Not Just About Geography.

00:06:44: And if we apply that checklist, it exposes some massive blind spots.

00:06:48: because Enrico Sinurelli brought up a brilliant operational point related to this.

00:06:52: He asked CTO's very uncomfortable question.

00:06:54: Oh I read about disaster recovery.

00:06:57: Yes he asks where does your Disaster Recovery Infrastructure run?

00:07:02: Because he sees companies spending tens of millions of euros building a fully sovereign, hyper-secure primary cloud.

00:07:09: But then to save money they run their disaster recovery backups on US hyperscaler.

00:07:14: It completely defeats the purpose.

00:07:16: Right Your last line of defense The system you activate when everything goes wrong still depends on foreign soil.

00:07:22: You've built an impenetrable fortress, but the emergency exit leads directly into the territory you were trying to avoid.

00:07:28: That's such a good analogy!

00:07:29: And to solve this, Senority noted that companies like Cubbit and SES, Elemento & Storepool are now launching fully sovereign disaster recovery packs.

00:07:39: These are systems deployable in hours that close the exact geographic and legal loophole.

00:07:44: So if sovereignty is about controlling where your data lives, ensuring you have power to walk away That naturally forces a completely different conversation on how actually pay for all of this.

00:07:53: Oh yeah The money.

00:07:54: Right You can't just swipe a corporate credit card And let servers run forever anymore.

00:07:58: Your architectural choices are suddenly dictating your financial freedom.

00:08:02: Which brings us our second theme This massive shift In Finops or cloud financial operations.

00:08:09: I really want to dive into that because the fine-offs evolution is fascinating right now and i found this brilliant historical analogy from mylan k in The Sources That Totally Reframes How You Should Think About Cloud Efficiency.

00:08:22: It's about the Jevons Paradox, have you ever heard of it?

00:08:24: Is that the coal efficiency theory?

00:08:26: Yes!

00:08:26: Back in eighteen sixty five an economist named William Stanley Jevins realized something incredibly counterintuitive Engineers were making coal engines much more efficient.

00:08:37: Logically, you'd think that means society would use less coal right?

00:08:41: Because the engines need less fuel to do the same amount of work.

00:08:43: Sure!

00:08:43: That makes sense.

00:08:44: But the exact opposite happened Coal consumption absolutely excluded because suddenly coal power was cheap and efficient enough that every single industry wanted to use it.

00:08:55: Making it efficient didn't reduce demand.

00:08:57: It induced demand.

00:08:59: Wow...that is a phenomenal parallel with cloud computing.

00:09:02: Right Mylan argues that making cloud infrastructure more efficient shouldn't just be about cutting costs and, you know handing a smaller monthly bill to the CFO.

00:09:12: Why?

00:09:13: Phenops?

00:09:13: one point oh was reactive.

00:09:15: it was Just finding waste in turning off servers no one Was using but phenops.

00:09:19: two point Oh is offensive.

00:09:21: if You optimize your code And make Your Cloud architecture four times cheaper To run The engineers don't Just bank the savings.

00:09:27: No they spin up Four Times as Many Machine Learning Experiments

00:09:30: Exactly.

00:09:32: The question shouldn't be how much did we save, the question is what strategic bets can we build now that we couldn't afford before?

00:09:39: And acting on that requires a total structural shift in how organizations view their engineering teams.

00:09:45: Nikolai Velchev made a pretty bold statement in his post.

00:09:48: he said Phenopsis not a finance function

00:09:51: Not a Finance Function.

00:09:52: What Is It Then?

00:09:53: He argues it's about reviewing monthly bills.

00:09:56: Finops is economic control because architecture decisions are financial decision.

00:10:00: Oh,

00:10:01: I see

00:10:01: every time an engineer writes a line of code that queries the database They're spending the company's money.

00:10:07: if you can't answer exactly what it costs to serve one individual customer or process One transaction You have absolutely no control over your business margins.

00:10:17: Yeah mark Harris echoed that exact sentiment too.

00:10:20: he noted that Finops' evolving to link cost risk and value natively into the engineering process.

00:10:26: Uncontrolled cloud spend isn't just a finance problem anymore, it introduces security exposure and operational

00:10:32: risks.".

00:10:32: But here is the real friction point in the industry right

00:10:35: now... Let me guess.

00:10:36: Engineers absolutely hate staring at cost dashboards!

00:10:41: They despise it if you tell a senior developer to Stop building a new feature so they can stare at a dashboard and tag their resources for the finance team.

00:10:49: They will ignore you.

00:10:50: Yeah, we know this.

00:10:51: yeah That's not gonna happen.

00:10:52: You cannot slow engineering down to a crawl just to maintain the budget.

00:10:56: So how do we actually embed?

00:10:58: This financial control into the architecture without Just ruining the developer experience in slowing everyone down?

00:11:03: well John ended out provided A really compelling answer For this.

00:11:07: he argues that fan ops needs is self-healing architecture powered by agentic AI

00:11:12: agentic ai

00:11:14: ya.

00:11:15: For the last five years we've relied on passive tools, right?

00:11:18: A tool scans the cloud flags a wasteful process sends an alert to a Slack channel and then it just waits for human engineer to fix.

00:11:26: And The Human ignores the Slack Channel

00:11:28: Exactly!

00:11:29: By the time The Human actually looks at dashboard three days later infrastructure has already dynamically changed.

00:11:35: The Alert is stale.

00:11:37: money's gone.

00:11:38: So the Dashboard isn't really problem...the human bottleneck.

00:11:42: John argues.

00:11:43: we need continuous investigation with AI agents that don't just alert but actually execute.

00:11:49: Agentic AI means the system has memory, bounded execution rights and ability to turn runtime waste into pre-deployment prevention rules.

00:11:56: Wait really so it can write its own rules?

00:11:58: Yes The system notices a server is over provisioned It scales down autonomously within safe limits And writes a rule.

00:12:05: So doesn't happen again.

00:12:06: Okay if sounds like science fiction let me bring in Stein to pro.

00:12:10: He posted about a weekend project that proved how shockingly accessible this automation has become.

00:12:15: Oh, I love good weekend projects.

00:12:17: over

00:12:17: a single weekend using a workflow Automation tool called n eight and the AI model.

00:12:22: Claude he built for functional phenops AI agents in one week One Weekend you literally said it is scary simple to create these agents.

00:12:33: now they scan his environments identify cost anomalies And handle the remediation automatically

00:12:39: democratizes financial control.

00:12:41: You don't need a massive enterprise software budget to build these guardrails

00:12:44: anymore.".

00:12:44: And we desperately needed that democratization because Venkat Reddy Chintalapudi highlighted a terrifying statistic in his post, he said fifty-six point five percent of tech teams are still managing their Kubernetes costs manually.

00:12:58: He calls it trying to count raindrops and thunderstorms.

00:13:01: Manual Kubernetes cost management is projected fifty billion dollars

00:13:09: trapped in unoptimized environments.

00:13:12: And we really need to explain why Kubernetes specifically breaks traditional finance?

00:13:17: Yeah, please do because it sounds crazy!

00:13:19: Well,

00:13:20: traditional cloud billing was built for static servers.

00:13:23: You rent a server you run at twenty four hours a day get a bill for twenty-four hours and compute.

00:13:27: easy right but kubernetes is an orchestration tool designed for massive chaotic distributed systems.

00:13:34: It chops applications into tiny ephemeral things called pods.

00:13:37: Okay.

00:13:38: A pod might spin up for thirty seconds just to handle a spike in website traffic from like a marketing email and then it vanishes.

00:13:45: It ceases to exist.

00:13:47: I see the problem,

00:13:48: yeah you cannot track the financial footprint of millions of thirty second ephemeral pods with a traditional monthly invoice.

00:13:54: You absolutely need policy as code and AI automation To manage that level of chaos

00:13:59: which provides The perfect bridge to our final theme.

00:14:02: because if AI is stepping in to manage Our cloud costs what happens when AI becomes?

00:14:06: The primary workload driving those cost In the first place?

00:14:09: If we connect this to the bigger picture We're looking at A total architectural rethink.

00:14:15: The community is calling it the AI Cloud Convergence.

00:14:19: Tell me more about that?

00:14:20: Christian Dussel had this massive realization.

00:14:23: walking out of KubeCon Amsterdam, he realized that the entire last decade Of cloud native development you know all the containerization All the microservices was actually just preparation for this exact moment.

00:14:37: Preparation For AI.

00:14:38: Yes!

00:14:39: AI inference Is going to be the biggest distributed systems challenge in human history.

00:14:43: And we should probably clarify the difference between training and inference here, because it dictates the infrastructure.

00:14:49: Training is when you feed a massive amount of data into a model to teach.

00:14:53: that takes massive centralized supercomputers running for months.

00:14:57: but inference is what happens after the model was trained When user actually asks at question generates an answer in real time

00:15:04: and inference requires unprecedented compute power spread across multiple global nodes with incredibly low latency.

00:15:11: Because no one wants to wait for an answer!

00:15:13: Exactly, you can't have a user in London waiting for a server in California to generate the next word in the sentence it has to be distributed.

00:15:21: that complexity is why Kubernetes has quietly become the backbone of modern AI infrastructure Not because it was originally built for artificial intelligence, but because its literally the only tool robust enough to handle the chaotic scale of global inference.

00:15:37: Here's where it gets really interesting... The physical cost running that infrastructure is staggering!

00:15:44: Which brings us back into our mind-bending statistic I mentioned at very beginning of this show…

00:15:48: The meta one?

00:15:49: Yes.

00:15:49: Meta is the undisputed king building custom data centers.

00:15:53: They spend north of a hundred billion dollars designing their own servers or networking gear, everything.

00:15:58: Yet Oliver D pointed out that Meta just committed an additional twenty-one billion dollars to external cloud provider called CoreWeave.

00:16:06: Wow!

00:16:07: That brings Metta's total commitment with CoreWeav to over thirty five billion dollars through twenty thirty two.

00:16:12: You really have to pause and think about the mechanics.

00:16:16: why does a company that literally writes book on custom hardware need to rent billion dollars of external compute.

00:16:24: Because the compute shortage is an existential threat, it's a bottleneck of advanced GPUs like NVIDIA H- one hundreds combined with this sheer electrical grid requirements to power these massive AI

00:16:36: clusters too much power

00:16:37: exactly demand for inference so unbelievably high that they old build versus by debate dead.

00:16:44: you can't choose.

00:16:45: just built your own data centers anymore.

00:16:47: have do both simultaneously at maximum scale.

00:16:52: That macroscopic stale is hard to wrap your head around, honestly.

00:16:55: But let's shift to what this convergence means for the individual engineer or architect.

00:16:59: listening to this right now like... What does it mean?

00:17:01: For your career!

00:17:02: Yeah

00:17:02: that's why everyone wants to know.

00:17:03: Both Yonkaiang Age and Vinod Kumar address this directly in their posts.

00:17:08: The days of kicking just one specialized path are over.

00:17:11: You mean being a Kubernetes administrator Or an AWS network architect?

00:17:16: Exactly The silo era is dead.

00:17:19: The modern engineer must stack skills to remain relevant.

00:17:24: Young Kang ran a poll of over four hundred and fifty engineers And the data showed that the most successful professionals in the market right now are combining three things cloud infrastructure DevOps automation, and AI integration.

00:17:38: Oh Vennad Kumar calls this the future stack Right?

00:17:41: Yes!

00:17:41: The Future Stack.

00:17:42: Young Kang noted something else That stood out To me.

00:17:44: he said Prompt to deployment is no longer an advanced highly specialized skill, it's the new baseline entry level requirement.

00:17:53: Just get in a door?

00:17:54: Right!

00:17:54: Meaning if you're engineer and can't take AI generated prompt validate code containerize it and deploy onto a working Kubernetes cluster autonomously your going fall behind incredibly fast

00:18:07: The tooling just advancing so rapidly that the baseline expectations for human output are compounding.

00:18:12: I

00:18:12: hear that, but let me play devil's advocate here.

00:18:15: Think about what we've discussed today.

00:18:17: We have sovereign cloud mandates forcing companies to juggle local providers?

00:18:21: We have phenops AI agents running around autonomously terminating resources and we have kubernetes clusters struggling To manage massive ai inference workloads.

00:18:33: aren't companies just creating a massive tangled web of technical debt?

00:18:37: How do you govern all of this without the entire system collapsing under its own weight?

00:18:41: That is a crucial question, and it's driving major behavioral shift to market right now.

00:18:46: David Green released a Market Report showing that tech teams particularly in German markets which are highly sensitive to compliance and sovereignty they're actively moving away from complex, sprawling multi-cloud setups.

00:18:59: They are!

00:19:00: Wait I thought multi cloud was the holy grail for the last five years?

00:19:03: Everyone said you have to use AWS Azure and Google simultaneously so that they never get locked into one vendor.

00:19:08: It's a conventional wisdom But in reality treating multicloud as a lockin avoidance strategy largely failed.

00:19:14: Why is it that?

00:19:15: Because it often just duplicated operational overhead its skyrocketed network egress costs and It created massive security nightmares because you had engineers trying to secure three completely different ecosystems at once.

00:19:28: That

00:19:28: sounds like a headache.

00:19:29: A huge one.

00:19:30: so now teams are consolidating.

00:19:32: Less is more.

00:19:34: they're moving toward what green calls intentional simplified environments.

00:19:38: They were being much more selective about where workloads go based on strict intent rather than just spreading them out for the sake of it.

00:19:45: Dimitri Furman made a fantastic point about this too, he said that treating the public cloud like another traditional data center is exactly why so many digital transformation programs completely stall-out.

00:19:58: He quoted a colleague who bluntly said don't wrap the crap.

00:20:02: I love that.

00:20:03: It's such good phrasing, meaning don't take your broken overly complex legacy processes put them in a virtual machine and just lift and shift them into the cloud.

00:20:13: you have to design natively for the cloud specific capabilities exactly.

00:20:17: That means actually letting go of old organizational exceptions and prioritizing only what actually differentiates Your business.

00:20:24: if running a custom email server doesn't make your product better Just stop running it.

00:20:29: Bringing all of this together raises a really profound friction point for the future.

00:20:34: We started today talking about the absolute necessity of digital sovereignty, right?

00:20:38: Controlling your data controlling your infrastructure and maintaining The ability to walk away from a vendor.

00:20:44: Right.

00:20:45: then we explored how fine ops dictates that you're architecture decisions are fundamentally financial Decisions.

00:20:50: every pod spun up is money spent.

00:20:53: finally We look at how agentic AI and automated workflow tools are rapidly being deployed to autonomously manage those complex distributed architectures, the handle inference.

00:21:03: So when you put those three truths in a blender what comes out?

00:21:07: A very challenging philosophical an operational question.

00:21:10: if your architecture decisions or your financial decision on highly autonomous agents are increasingly making this architecture decisions in real time two optimized costs and managed inference workloads who ultimately controls the digital sovereignty of your enterprise.

00:21:24: Is it the human executive sitting in a boardroom setting the policy?

00:21:27: is that cloud provider hosting infrastructure and providing base models or AI algorithm itself dynamically shifting workloads across borders at millisecond speeds just to chase a fraction of sent inefficiency.

00:21:42: That's brilliant, and honestly slightly terrifying thought.

00:21:46: to leave on.

00:21:47: The moment you hand keys for an autonomous agent The very definition of sovereignty gets incredibly complicated.

00:21:55: Are you still independent if you don't fully understand the minute-by-minute decisions your infrastructure is making on your behalf?

00:22:01: It's definitely something to chew on before your next architecture review!

00:22:05: If you enjoyed this episode, new episodes drop every two weeks.

00:22:08: Also check out our other editions on ICT and tech Digital products & services Artificial intelligence Sustainability in green ICT Defense Tech And health.

00:22:18: Thank You so much for joining us On This Deep Dive.

00:22:20: Keep building, keep questioning and don't forget to subscribe.

00:22:23: Catch you on the next one!