Best of LinkedIn: Cloud Insights & Sovereignty CW 41/ 42

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

00:00:05: In calendar weeks forty one and forty two.

00:00:08: Franus has built a sovereign cloud market radar for ICT providers with weekly hot news monthly reports quarterly leadership presentations and AI podcast for field teams.

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

00:00:21: so today We're doing an executive briefing on the top digital products and services trends across LinkedIn.

00:00:27: And it's really curated specifically for you that tech professionals managing digital transformation right now,

00:00:33: Right?

00:00:33: Because we are cutting right through the noise surrounding cloud architecture AI economics in governance to find the actual signal here were starting with a real bedrock of modern infrastructure which is well where your data physically lives and the engineering reality of who truly controls it.

00:00:50: Okay, let's unpack this because the word sovereignty gets thrown around in boardrooms so casually now?

00:00:56: It's kind

00:00:59: become just a legal checkbox.

00:01:01: But we need to move it into the realm of hard-coded engineering reality.

00:01:05: Bertrand Legra actually surfaced really brilliant framework for this.

00:01:09: Yeah,

00:01:09: The Six Questions

00:01:10: Exactly.

00:01:11: He argues that calling a cloud sovereign actually hides six distinct highly technical questions.

00:01:17: You have know where data sits whose legal jurisdiction reaches who physically holds encryption keys

00:01:23: And whether infrastructure keeps running if vendor relationship catastrophically fails

00:01:28: Right and whether you can actually build on it.

00:01:31: And finally, if its politically acceptable to rely on that specific foreign provider.

00:01:37: what's fascinating here is the most crucial distinction LaGrosse makes is fundamentally separating sovereignty from resilience

00:01:43: Which people mix up all of time?

00:01:45: Constantly.

00:01:47: Resilience is availability under failure like a fiber cut or server fire You solve with redundancy.

00:01:53: Sovereignty Is Availability Under Decision In executive order or a sudden sanction.

00:01:59: Right?

00:02:00: And that tension is playing out live at the hardware level right now, Fadi D'Souz flagged this massive development where AWS's running alive tests to actually cut off their European sovereign cloud from global AWS backbone.

00:02:13: They're physically severing it

00:02:15: Literally intentionally severing digital connection To prove datacenters can operate independently inside EU operated solely by EU residents without calling home to US servers.

00:02:27: I mean, it is a serious engineering flex to sever the backbone of a hyperscaler and prove that lights stay on.

00:02:33: Wait

00:02:33: but let me push back on this exact point though... If The U.S.

00:02:36: Cloud Act still applies because the parent company's American aren't we just buying a highly resilient data center and slapping a sovereign sticker?

00:02:44: That is the big debate!

00:02:46: Right Because the law historically doesn't care where server rack physically sits It cares who owns the subsidiary.

00:02:52: So if a subpoena lands in Seattle ...the data in Frankfurt is still compromised

00:02:56: And that exact legal friction is driving a completely different architectural approach.

00:03:01: Harold Yos highlighted the model being built with the Thales-operated Google Sovereign Cloud.

00:03:05: Oh, The Separation of Duties?

00:03:06: Exactly!

00:03:07: Thales which an independent German owned entity holds operative control and the sole encryption keys.

00:03:13: Google provides underlying hypervisors in storage but they are COMPLETELY blind – They mathematically cannot decrypt workload.

00:03:20: So their building operator US government legally can't compel?

00:03:24: Precisely because the parent company doesn't have the keys.

00:03:28: But, The problem extends way past the provider's corporate structure right?

00:03:32: It is buried in your own application code.

00:03:34: Harry Malonis warned that most of these shiny multi-cloud policies that Corporate Boards are rubber stamping Are going to fail.

00:03:40: a Dora exit test.

00:03:41: And we should clarify, a Dora exit test driven by the EU's Digital Operational Resilience Act is not just a paperwork drill.

00:03:49: No!

00:03:49: Not at all.

00:03:50: You have to prove you can physically move your critical workloads elsewhere if a vendor collapses.

00:03:54: Exactly and Milona's points out organizations are confusing provider sovereignty with architectural excitability.

00:04:01: Provider Sovereignty Just means that hyperscaler cant read data but Architectural Exitability Means you Can actually pack up & leave

00:04:09: Right, if you're hard-coded to AWS IAM.

00:04:12: You can't exit

00:04:13: Yep If your microservice authorization is irreversibly tied there or your AI reasoning loops are bound to Amazon bedrock?

00:04:21: You physically cannot exit that cloud...you move the data The application breaks.

00:04:25: This really ties into the core argument from Peter DeLorenzo.

00:04:29: He says true sovereign team means having the engineering capability To switch providers..if a code traps you....You want Sovereign!

00:04:35: You just negotiated A longer lease.

00:04:37: Just uh New landlord.

00:04:39: Exactly, and Bert DeSmet notes the regulatory heat is turning up.

00:04:43: The EU-CADA proposal level two sets audited sovereignty requirements.

00:04:47: We're talking mandatory local support models complete software bills of materials.

00:04:52: So major US hyperscaler sovereign editions might actually fail these audits?

00:04:57: They might because self declaration isn't enough anymore.

00:05:00: Auditors want to see the architecture.

00:05:02: But despite all these hurdles that deployment momentum is very real.

00:05:07: Stephen Wagner from SAP stated, digital sovereign trees basically an engineering challenge requiring strong local ecosystems.

00:05:15: And we're seeing them stand up?

00:05:17: We are.

00:05:18: Ash Thanky noted the Brandenburg Medical University is using the AWS European Sovereign Cloud to connect a regional health network and Harriet Reese shared that Starling is the first UK bank to train AI on a UK sovereign cloud via SIVO

00:05:33: which ensures domestic financial data never leaves domestic infrastructure during training.

00:05:38: And if true sovereignty requires you to be able to cleanly exit a cloud environment, You really have to prioritize architectural freedom above all else.

00:05:47: This is triggering massive resurgence of private cloud To avoid commercial lock-in.

00:05:52: Public and private Cloud Parity Is making workload placement an architectural choice again Not a constraint.

00:05:58: But how do you justify that capital expense to a board that spent a decade demanding public cloud?

00:06:04: Brian White highlighted, A perfect catalyst.

00:06:07: Broadcom's decision to deauthorize insight enterprises as a VMware reseller

00:06:12: Yeah That was huge.

00:06:14: So what does this all mean?

00:06:16: It's like renting a fully furnished apartment.

00:06:19: You've lived there for years.

00:06:21: Then without warning the landlord decides To triple the rent specifically on The couch you sit On every day.

00:06:26: I love

00:06:26: that analogy.

00:06:27: Right Your application hasn't changed.

00:06:30: Not one line of code, but your financial reality is ruined by a vendor decision.

00:06:35: so you have to preserve your freedom to move your own furniture.

00:06:38: The market reaction that economic shock has been staggering!

00:06:43: Andres Cleaver shared survey data showing nine out ten VMware customers are looking for an exit.

00:06:47: Wow...

00:06:47: ninety percent

00:06:49: Yes, and sixty percent want a multi-hypervisor strategy.

00:06:52: They refuse to swap VMware lock in for a different single vendor login.

00:06:57: but obviously running multiple hypervisors introduces massive operational complexity.

00:07:02: Oh absolutely

00:07:03: Which is why providers like HPE are pushing hybrid by design cloud ops platforms Like Morpheus To abstract that complexity And manage it all from a single control plane.

00:07:13: The hardware vendors are racing to catch this Exodus too.

00:07:16: Michael Redmond noted Nutanix is actively building landing zones where licensing practically maps onto VMware tiers based on the workload.

00:07:23: They're trying to make the escape hatch as frictionless as possible.

00:07:27: if we connect this to The Bigger Picture, David Linthicum delivered a pretty stark warning.

00:07:32: he argues public cloud has silently become exponentially more expensive and it isn't inherently more secure.

00:07:39: so his message is stop taking your architecture strategy from the vendor holding the invoice.

00:07:44: exactly It flips the cloud first.

00:07:46: always assumption to, Cloud only where it makes mathematical sense.

00:07:50: And we are seeing industry specific use cases for this.

00:07:53: Todd Mason pointed out that a twenty four seven live television broadcast production is way better suited for purpose built private multi-tenant cloud than public one.

00:08:02: The latency tolerance just demand dedicated hardware

00:08:06: An edge computing another mechanism.

00:08:08: Eugene Park highlighted Dell Private Cloud Edge virtualization.

00:08:13: Instead of round-tripping real time inventory computer vision data from a store in Chicago to Virginia and back...

00:08:19: They just put the compute in the store.

00:08:20: Right, it drastically lowers AI inferencing latency And cuts down on truck rolls for IT support.

00:08:26: But let's look at the human toll here.

00:08:28: Your infrastructure is now spanning audited sovereign zones Edge locations & stores In multi hypervisor private cloud.

00:08:36: Human teams are completely overwhelmed.

00:08:39: The dashboards are blinking red too fast

00:08:41: Way Too Fast.

00:08:43: We're just driving this convergence of DevOps, ML Ops and AI into agentic cloud ops.

00:08:48: AI is stepping in to operate the Cloud.

00:08:51: It's a huge paradigm shift.

00:08:54: Carthike NJ and Jill Ferris highlighted The AWS well-architected Agent entering public preview.

00:09:00: In the old model an engineer finds An issue reads docs writes terraform code And deploys it.

00:09:06: right now the AI agent identifies the anomaly explains the trade-offs and generates the exact prescriptive fixes.

00:09:15: The engineer just approves

00:09:16: it.".

00:09:16: It's wild, like AI isn't suggesting a blueprint name where its pouring concrete.

00:09:22: Thomas Fontres pointed out Nokia.

00:09:23: AI agents are cutting cloud migration cost assessments from days down to literally minutes.

00:09:29: They map dependencies instantly.

00:09:32: Here is really interesting though If the AI recommends The Fix, writes the code and evaluates the cost.

00:09:38: At what point does human cloud architect just become a rubber stamp?

00:09:42: Are they blindly clicking approve?

00:09:44: Mark Vigueroso hosted this exact psychological trap.

00:09:49: We are risking repeating old Cloud lift-and-shift mistakes with AI agents.

00:09:54: If you bolt an agent onto broken deployment process... You get automated chaos at light speed!

00:10:00: What's The Fix?

00:10:01: Governance first The traditional change advisory board, the cab is dead in an automated world.

00:10:08: You have to repulse it with hard-coded automated

00:10:10: guardrails."

00:10:11: And Yulia Petrishanthian echoed that.

00:10:14: she said autonomous operations require a persistent infrastructure context layer.

00:10:20: you can't just give an AI API access to Terraform and let it loose.

00:10:24: No!

00:10:24: It'll optimize one service and break three others.

00:10:27: Exactly...it needs semantic understanding.

00:10:30: This is forcing disciplines.

00:10:32: Ravi Sankara noted DevOps, ML Ops and AI are converging into one governed operating model.

00:10:38: And Olivia Caruso pointed out that IBM Bob is now available.

00:10:41: on-premises and air gapped which proves trust in governance are officially the AI product not just model size.

00:10:49: if a bank can't put its code in public AI The vendor has to bring the AI to the Bank

00:10:54: Exactly!

00:10:55: Operating those models leads to ultimate reality check which is economics.

00:10:59: You can't let AI agents manage your cloud without governing the explosive compute costs.

00:11:05: Speaking of managing things real quick, if you are finding these insights valuable should definitely make sure to subscribe to The Feed so that you don't miss future deep dives into digital transformation trends.

00:11:15: Alright lets talk FinOps and AI cost.

00:11:19: This conversation has matured SO fast.

00:11:22: Joe Daly noted FinOps practitioners have stopped debating whether to cap token usage.

00:11:27: Now It's about how to build flexible CAFs.

00:11:31: Like setting limits based on job roles?

00:11:33: Right!

00:11:34: A junior PM doesn't need the same token budget as a senior developer, and hitting a CAP doesn't shut you off.

00:11:39: it just starts a conversation.

00:11:41: Tatiana Boyarshan & Satya Depal Singh shared some great lessons from FinOpsx Amsterdam on this.

00:11:47: because measuring AI economics purely by tokens is so flawed?

00:11:51: Oh completely...

00:11:52: Its like measuring cost of meeting with the number words spoken…it gives you a number but it ignores whether any actual work got done.

00:11:59: That is the perfect way to look at it.

00:12:01: Singh argues we have to shift to measuring the cost per successful business outcome.

00:12:05: if a cheap model requires thirty minutes of human debugging, It's actually more expensive than premium models that get it right instantly.

00:12:12: This raises an important question though How do you forecast these outcomes?

00:12:17: Diana L argues cloud forecasts based purely on historical spend math are just flawed

00:12:23: now.

00:12:23: Historical Math Is Stale Math.

00:12:25: Exactly.

00:12:27: They have to include direct engineering input on upcoming architecture changes.

00:12:31: And the visibility tooling is finally evolving for this.

00:12:34: Hassan Kajahusani noted infracost as shifting Kubernetes.

00:12:37: pod cost estimates left, engineers see the financial estimate directly in their pull request before deployment.

00:12:44: That's

00:12:45: huge and Ben Schachter noted advantages adding NVIDIA MIG support multi-instance GPU.

00:12:51: So instead of renting a whole massive chip and wasting power you slice it up.

00:12:56: This means companies can finally allocate costs for partial GPU usage.

00:13:00: It's all about precision,

00:13:02: but Nicholas Fondrini reframed this beautifully.

00:13:04: He says sometimes spending extra upfront on architecture is a Phenops risk premium

00:13:09: like buying insurance against lock-in.

00:13:12: Yes Spending a bit more to reduce commercial lock in cyber risk and concentration risk

00:13:16: which really summarizes the core thread of this entire deep dive

00:13:20: it really does.

00:13:21: Whether we're talking about Dora Exitability moving off VMware, governing AI agents or tracking AI ROI the ultimate goal for tech leaders right now is taking back control of their architecture and their economics.

00:13:34: Taking Back Control I love that but it leaves me with a final maybe slightly terrifying thought for you to ponder.

00:13:41: if agentic ai eventually abstracts away all Will our ultimate vendor lock-in simply shift from the cloud provider to the AI agent itself?

00:13:56: That is a very scary thought.

00:13:58: Right, if you enjoy this episode new episodes drop every two weeks.

00:14:03: also check out or other additions on digital products and services defense tech AI in agentic systems green ICT and sustainable AI and health tech.