Best of LinkedIn: Cloud Insights CW 23/ 24

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

00:00:06: And twenty four.

00:00:07: Frenness has built a sovereign Cloud Market Radar for ICT providers with weekly hot news monthly reports quarterly leadership presentations and AI podcasts for field teams.

00:00:18: you can find more info in the description.

00:00:20: welcome to The Deep Dive.

00:00:23: so i want you imagine first second that it's Twenty-twenty six.

00:00:27: You're a CIO, you are sitting at your desk looking for financial projections and suddenly realize something completely counterintuitive.

00:00:34: Let me guess moving cutting edge AI workloads back to an on premise mainframe is actually cheaper than running them in the public cloud?

00:00:40: Exactly Yes!

00:00:41: Mainframes Which I mean it sounds completely absurd.

00:00:44: Well its totally absurd if look into lens of like last decade.

00:00:48: Yeah But that's exactly reality we are barreling toward right now Because The Public Cloud originally promised us infinite scale Zero hardware maintenance pennies on the dollar for compute power.

00:00:58: But right now we are seeing this massive structural pullback.

00:01:02: It's wild.

00:01:03: everyone suddenly seems to be scrambling to buy physical servers again You know repatriate their workloads and lock there data inside these highly specific Geographically isolated vaults,

00:01:14: right?

00:01:15: And if you're an ICT or tech professional listening to this your probably feeling the whiplash?

00:01:19: Oh absolutely so.

00:01:21: today We're gonna unpack why this is happening.

00:01:24: We've spent the last couple of weeks analyzing the most critical shifts dominating the top cloud trends on LinkedIn for calendar weeks, twenty-three and twenty four.

00:01:32: And there's a very clear mystery here

00:01:35: The Cloud was supposed to eat the world?

00:01:37: Yeah exactly.

00:01:38: So why is cloud first suddenly being replaced by what David A recently called a cloud selective strategy?

00:01:45: Well it's because The questions that tech leaders are asking have fundamentally changed.

00:01:50: As David pointed out, the boardroom conversation is no longer about if an organization should move to the cloud.

00:01:55: They're now looking at their entire portfolio and asking which specific workloads actually belong in a public multi-tenant environment Which ones will perform better on bare metal?

00:02:07: And crucially...which one's absolutely require sovereign heavily localized architectures.

00:02:14: Yeah, let's pull on that thread of sovereign architecture because it feels like the biggest immediate driver pulling workloads out.

00:02:23: And I mean, this isn't just about companies getting nervous about their intellectual property.

00:02:27: No not at all!

00:02:28: It's being forced by massive regulatory shifts especially over in Europe.

00:02:32: it is entirely regulatory driven if you look at the analyses shared by Kies Huger-Vorst and Martins Wick.

00:02:38: recently they broke down the EU's tech sovereignty package... ...and The Upcoming Cloud & AI Development Act.

00:02:45: Exactly Kennedy.

00:02:47: what's fascinating here is how the legal definition of cloud infrastructure has completely transformed Regulators, they aren't treating the cloud as just an IT service anymore.

00:02:56: They are treating it is critical national infrastructure.

00:02:59: Yeah It's being put on the exact same regulatory tier As The National Energy Grid Or Like The Telecommunications Network.

00:03:06: And as Ilias Chance was pointed out This means that old playbook Is Just Dead Because, you know a few years ago.

00:03:12: You could satisfy regulators just by proving your data physically resided in a data center and Frankfurt or

00:03:18: Paris?

00:03:18: Right!

00:03:18: Data residency was enough

00:03:20: Exactly But chances noted that buyers are now legally mandated to demand absolute jurisdictional and operational control.

00:03:28: Okay let's unpack this because is where I get a bit confused.

00:03:32: If a European government essentially says we don't trust U.S or Chinese cloud providers anymore, We are only going to use local ones aren't we just swapping?

00:03:41: A massive highly efficient foreign monopoly for smaller less-efficient domestic one?

00:03:47: I mean how does creating a walled garden actually help in enterprise scale?

00:03:50: well

00:03:51: That is the exact trap that policymakers are trying to avoid and OV young brought up a really crucial point regarding public sector procurement here.

00:03:59: true digital sovereignty isn't just about looking at the passport of the company's CEO.

00:04:03: If you lock yourself into a local vendor and can never leave, You aren't sovereign...you're just a hostage to a different

00:04:10: master.".

00:04:11: So Young argues that true sovereignty requires contestability

00:04:15: Meaning The power To realistically fire your supplier.

00:04:18: Exactly That.

00:04:19: And deep interoperability

00:04:21: Which brings up incredible historical analogy that David Eves shared in this context.

00:04:26: It completely changed how I look at this problem.

00:04:28: Oh, the NATO interoperability crisis?

00:04:30: Yes

00:04:31: so back in the nineteen fifties NATO realized they had this massive problem.

00:04:35: They were a unified alliance on paper but every member country used different rifle calibers Different aviation fuel grades entirely different communication protocols

00:04:46: Total nightmare for logistics.

00:04:47: right?

00:04:48: if French troops ran out of ammo British troops couldn't hand them their magazines.

00:04:52: Now, NATO didn't solve this by forcing every country to buy their weapons from one single American manufacturer.

00:04:58: They solved it by establishing incredibly strict standardized

00:05:01: protocols.".

00:05:02: And Eves argues we have do the exact same thing with a cloud today.

00:05:06: If you enforce strict architectural and API standards digital sovereignty can coexist with operational flexibility.

00:05:13: You don't need one giant provider...you need highly standardized ecosystem

00:05:18: Exactly.

00:05:19: But wait, hold on.

00:05:19: Let me push back on this a bit because the US hyperscalars are not just giving up on Europe.

00:05:24: I was reading Guido Hockenberg's breakdown of four levels of sovereignty under Kanan And Amazon recently announced they're dropping €七 point eight billion to build dedicated European sovereign cloud.

00:05:37: Yeah it is a lot money.

00:05:38: Seven

00:05:38: point eight million.

00:05:39: Are you telling that kind of money still doesn't buy them?

00:05:42: full sovereignty status in eyes of EU?

00:05:45: How does hardware sitting on European soil count?

00:05:48: This is where the geopolitical legal clash happens, and Hockenberg laid this out perfectly.

00:05:53: Level one sovereignty just means that data physically sits in EU.

00:05:56: Okay so Amazon clears it easily?

00:05:58: They clear it easily.

00:05:59: but level three sovereignty which will be required for highly sensitive workloads mandates that cloud provider must be owned legally controlled from within Europe.

00:06:08: Ah I see!

00:06:09: Amazon cannot touch level III regardless of how many billions they spend Germany or France.

00:06:14: why?

00:06:15: because they are headquartered in Seattle Which means the US cloud act still applies to the parent company

00:06:22: meaning us federal authorities could theoretically subpoena that European data and Amazon would be legally bound by u.s.. Law, to hand it over which directly violates european law.

00:06:33: exactly The physical location of this server doesn't matter if the corporate entity holding the keys is subject To a foreign jurisdiction.

00:06:40: And paul if I can point out the massive market implications of this legal paradox right now U.S.

00:06:45: Hydroscalers hold roughly sixty-five percent of the European cloud market.

00:06:50: When CAD is fully enacted, it's going retroactively turn that sixty five percent market dominance into a profound compliance nightmare for any enterprise running critical infrastructure.

00:07:00: Wow and we are already seeing enterprises executing on this reality like its not just theory anymore, Dr.

00:07:06: Fariable Hassan shared a breakdown of Volkswagen's new dual cloud strategy.

00:07:11: Yeah

00:07:11: that was great example.

00:07:12: For their standard applications sure they use the public cloud but for highly critical sensitive workloads They are physically separating those systems and deploying t-cloud private to ensure absolute resilience and legal sovereignty.

00:07:26: And we're seeing it on the channel partner side too, Janet Smith and Michael Schulte highlighted that Capgemini just became SAP's first sovereign cloud partner across several European countries.

00:07:37: right this is serious capital expenditure moving away from traditional public clouds into specialized sovereign architectures.

00:07:46: but you know regulations in compliance are really only half of story here because even if your operating a company outside Europe completely untouched by CADDA, there is another gravitational force pulling workloads out of the public cloud.

00:07:59: The AI economics?

00:08:00: Yes!

00:08:01: The sheer jaw-dropping cost of running artificial intelligence...

00:08:04: It's completely breaking the traditional economics at the Cloud.

00:08:07: The numbers coming right now are just staggering.

00:08:10: Anand Kalyan and Nagesh Jezwal shared a major statistic from Broadcoms' Private Cloud Outlook twenty-twenty six report.

00:08:16: Oh yeah I saw this.

00:08:17: Listen to this shift.

00:08:19: Private Cloud has officially overtaken Public Cloud for production AI inferencing.

00:08:25: Right now, fifty-six percent of enterprises are opting for private on-premise deployments for their AI models while public cloud usage for inference is dropped to forty one per cent.

00:08:36: Okay here's where it gets really interesting because repatriating AI into a private cloud sounds like fantastic way save money on paper.

00:08:44: But to me, it sounds like buying a high-performance sports car because you want to save money on your Uber fares but suddenly realizing that now have to pave your own roads, refine your own high octane gasoline and hire full time pit crew.

00:08:57: That is brilliant way of putting it!

00:08:58: I mean our enterprise IT team's actually ready for the massive operational burden?

00:09:02: You hit on the multi million dollar catch right there And Anant Kalyan validated exactly this concern.

00:09:07: When you pull a massive workload out into public cloud and get home Operational load moves with it.

00:09:13: For the last ten years, The public cloud abstracted away all of the truly miserable parts of IT.

00:09:18: The thermal cooling...the hardware load-gallancing....The three AEM incident response.

00:09:23: Paving your own roads demands a level operation maturity that a lot infrastructure teams simply haven't maintained.

00:09:29: But the

00:09:29: token costs for AI are so high they're basically being forced to do it anyway right?

00:09:34: John Schultz pointed out that the unpredictable infrastructure spikes associated with large-scale AI have simply broken the public cloud pricing model for these companies.

00:09:43: Exactly, and Rami Alka Fajie highlighted some incredible IBM data that puts this into perspective.

00:09:49: Seventy two percent of organizations are facing higher than expected cloud costs right now.

00:09:55: The financial pressure is so intense that executives are actually reading mainframes, literally legacy iron as better than the cloud on a cost per transaction basis for specific AI workloads.

00:10:04: See?

00:10:04: That just blows my mind... Mainframes!

00:10:07: We're talking about technology that peaked in the nineteen eighties.

00:10:10: How does a mainframe beat a hyperscale cloud cluster For cutting edge AI?

00:10:15: It comes down to data gravity.

00:10:17: If you were massive bank Your core systems of record every transaction Every account balance run on a main frame.

00:10:23: If you want an AI to analyze that data in real time, moving petabytes of data back and forth over a network into public cloud for inference is astronomically expensive.

00:10:33: Not to mention the latency!

00:10:35: Right...the

00:10:35: latency makes real-time analysis impossible.

00:10:38: The AI has run where it physically lives.

00:10:41: Moses Acosta framed this perfectly.

00:10:43: He said AI is forcing a hard reset on cloud strategy.

00:10:47: It is no longer about simply being Cloud first.

00:10:50: it Is about establishing a strict economic control model and we're seeing the big tech vendors Acknowledge This reality, and adapt their software to it.

00:10:58: Douglas Phillips was talking About The recent Microsoft build conference where they Showcase foundry local

00:11:03: which is a massive paradigm shift right.

00:11:05: Foundry Local allows agentic ai Autonomai agents that can make decisions And take actions To run full on-premises, zero data movement outside the enterprise environment.

00:11:14: And if we connect this to the bigger picture it's not just about software agents reading spreadsheets.

00:11:20: Alex Romero brought up The Rise of Physical AI which completely redefines where infrastructure needs to sit.

00:11:27: Oh like robotics

00:11:28: Exactly We are seeing the convergence of AI cloud architecture and robotics.

00:11:34: Think about an advanced manufacturing floor.

00:11:37: The machines in that factory aren't just stand-alone hydraulic tools anymore.

00:11:40: They are cloud connected intelligent agents.

00:11:43: they perceive their environment, they reason through production problems and they act.

00:11:48: And you can't run a robotic arm relying on a public Cloud server three states away.

00:11:52: the roundtrip latency alone would cause the machine to crash into a wall before the server even processed the stop command.

00:11:59: Exactly!

00:12:02: real-time physical parsing require the infrastructure to sit inside the factory on the edge.

00:12:07: So between sovereign data laws, this sheer cost of API tokens and latency demands for physical AI... The public cloud is no longer a default destination for

00:12:16: everything.".

00:12:22: For the workloads that do stay in the cloud or operate across these incredibly complex hybrid environments, how on earth are companies controlling this spend?

00:12:33: Because your engineers are suddenly making massive financial decisions every time they execute a script.

00:12:38: Which brings us to our third major theme – The Rapid Force Evolution of Phenops Financial operations and the Cloud used.

00:12:48: as Ajay Joshi pointed out, it was essentially a finance team looking at a dashboard at the end of the month gasping at the AWS bill and then yelling at engineering to turn off some vital servers.

00:12:58: Yeah but that doesn't work when an AI agent can spin up ten thousand concurrent processes in three seconds.

00:13:04: Azilar San Natarjan shared a wild statistic.

00:13:07: Ninety-eight percent of Phenops teams right now are tasked with managing AI spend but almost none of them have the granular tooling required

00:13:17: Because treating AI-span like traditional cloud compute spend is a recipe for bankruptcy.

00:13:22: Traditional Compute is billed by the hour or second, AI is billd by token.

00:13:27: Every single

00:13:29: word every snippet of code red generated costs money.

00:13:34: The optimization can't happen at end of month.

00:13:36: it has to happen in engineering level.

00:13:40: Breno Capuana shared some highly practical advice on this, specifically regarding developer tools like GitHub Co-Pilot.

00:13:47: If you give co-pilot access to your entire enterprise code base –like a massive context window–you are paying for millions of tokens.

00:13:55: every single time a developer asks it fix the simple typo…

00:13:58: Which adds up incredibly fast!

00:14:00: Capilano noted that using AI tools more doesn't have to mean burning more money.

00:14:04: You cut the token drama by training your engineers to use much smaller context windows, deliberately selecting cheaper, smaller models for simple tasks and using agent mode with incredibly strict financial

00:14:15: guardrails.".

00:14:16: And we're seeing Phenops move directly into the integrated development environment at IDE.

00:14:21: Banna G mentioned a new AWS Phenop's agent which is currently in preview...

00:14:25: Oh right!

00:14:26: This tool brings cost anomaly investigations straight into the daily workflows of the engineers.

00:14:31: You don't have to wait for finance to flag an issue, The engineer gets a real-time prompt suggesting and architectural optimization.

00:14:38: right there on their screen it is shifting financial accountability all the way left.

00:14:43: And speaking of those architectural optimizations Lefterist Kara Georgia shared A brilliantly simple engineering hack On LinkedIn that I absolutely loved.

00:14:52: He needed To build a serverless read API and he built it in ten minutes, but he completely skipped the compute layer.

00:14:59: Wait

00:14:59: really?

00:15:00: Yeah!

00:15:00: He didn't use an AWS Lambda function at all—he just used native integrations to connect API Gateway directly to DynamoDB Which

00:15:07: is brilliant from a cost perspective.

00:15:09: Exactly because AWS Lambda has cold starts.

00:15:12: It takes time to spin up the compute environment And you pay for every millisecond that runs.

00:15:17: By bypassing it entirely, Lefteris eliminated this spinoop latency... ...and cut out entire billing layers.

00:15:23: As we put it sometimes deleting a layer is the optimization.

00:15:27: And there are dozens of these micro-optimizations emerging at the platform level too.

00:15:31: Victor Garcia listed several major cost saving updates that dropped right around the Finoff's X conference, for example Google Clouds.

00:15:39: BigQuery just introduced fluid scaling with per second billing meaning you weren't paying for heavy compute cycles this second year data query finishes.

00:15:48: He also mentioned Azure's global PTE reservations.

00:15:51: For you listening, a provisioned throughput unit or PTU is basically a guaranteed reservation for AI compute power.

00:15:59: But traditionally if you reserved that power in specific regions say US East and your demand shifted to Europe You just wasted all of the money.

00:16:07: Now Azure's making those reservations region agnostic letting companies shift their computer allowance globally To chase demand and eliminate waste at spend

00:16:15: All these tactics raise critical question about how we view our relationship with hyper scalers.

00:16:20: George K. Matthew made a very sharp point, cloud commitment maturity isn't about letting your procurement team chase the highest discount percentage on a three-year contract.

00:16:29: it's about balancing technical efficiency with financial flexibility based on your actual business reality.

00:16:35: right

00:16:35: and Guy Bartram added a crucial layer to this.

00:16:38: Cloud portability is fundamentally an architectural decision not a contractual one

00:16:43: meaning that the cloud providers don't lock you in with lawyers.

00:16:45: You lock yourself in with lazy engineering

00:16:48: exactly.

00:16:49: The Hyperscalers recently made it free to transfer your data out.

00:16:53: But if you're application relies on fifty proprietary AWS or Azure services that don't exist anywhere else, Your architecture is what keeps you trapped.

00:17:02: and let's not forget the macroeconomic reality here...the hyperscalers are currently investing roughly six hundred billion dollars in AI capital expenditure.

00:17:11: Someone has to repay that six-hundred billion And they fully intend for it be you.

00:17:15: So what does this all mean?

00:17:16: We're obsessing over exactly where our data lives, like renting a highly secure titanium vault in a specific European city.

00:17:24: But with these APIs and interconnected systems are we just leaving the Vault door wide open?

00:17:29: That is the ultimate paradox of this entire industry shift And Tim Reigns articulated perfectly.

00:17:34: He argued that Data Residency's very often just security theater

00:17:38: Security Theater Meaning it looks good on an audit report but doesn't actually stop a breach.

00:17:43: It satisfies regulators and creates a feeling of comfort for the board, But it completely ignores the reality how threat actors operate today.

00:17:51: Attackers do not care about physical geography as a server.

00:17:55: They don't care if your hardware is sitting in Frankfurt Dublin or an Ohio cornfield.

00:17:59: they care about connectivity

00:18:01: Because they aren't physically breaking into data center.

00:18:04: They are attacking across networks using compromised credentials And open APIs

00:18:10: phishing, credential stuffing API exploitation.

00:18:13: As Reigns points out every time you deploy a new AI system it calls external APIs.

00:18:17: that exposes internal APIs and radically expands your attack surface.

00:18:22: putting data in sovereign data center solves the political problem an illegal problem.

00:18:27: It does absolutely nothing to solve security problems.

00:18:30: The real architectural challenge, and the actual security work that protects the business is rigorous identity management strict network segmentation.

00:18:38: And flawless API governance.

00:18:41: if you don't have those fundamentals dialed in your multimillion dollar sovereign cloud it's just a locally hosted data breach waiting to happen.

00:18:48: That is a deeply sobering thought but absolutely vital for any architect or engineer designing these systems right now.

00:18:55: You cannot just check the compliance box, wipe your hands and assume that engineering is secure because the server has European passport

00:19:02: Exactly.

00:19:03: And as we wrap up this deep dive into insights from past couple of weeks This raises an important question.

00:19:08: drawing on the perspectives of Enrico Signoretti & Lars Neumann.

00:19:12: They pointed out that industry effectively dividing IT leaders in two distinct camps

00:19:18: Oh, the cowardly versus courageous CIOs.

00:19:20: Yes!

00:19:21: There is The Cowardly CIO This leader who looks at overwhelming complexity of new regulations and uses it as an excuse to avoid making hard architectural decisions.

00:19:32: They treat vendor lock-in just a necessary technical tradeoff And cross their fingers hoping prices don't spike.

00:19:39: Then there was The Courageous CIO.

00:19:41: This leader treats Stack Lock in not as an IT problem, but a massive systemic risk to the survival of business.

00:19:49: They design the exit strategy before the kill switch is flipped.

00:19:52: They engineer portability first?

00:19:53: Yes

00:19:54: because while achieving true digital sovereignty managing AI token costs and securing a sprawling hybrid architecture might require intense engineering come at premium price.

00:20:04: The question you have ask yourself this What is the true long-term cost of not being sovereign?

00:20:10: what does the ultimate cost of outsourcing your fundamental business continuity to a foreign tech stack that you cannot control.

00:20:16: If you enjoyed this episode new episodes drop every two weeks.

00:20:20: also check out our other editions on ICT and Tech, Digital Products & Services, Artificial Intelligence Sustainability in Green ICT, DefenseTech and Health.

00:20:29: Thank you so much for joining us on this deep dive into the forces reshaping.