Latest / Best of LinkedIn: Digital Transformation & Tech / Best of LinkedIn: Cloud Insights & Sovereignty CW 31/ 32

Transcript

Best of LinkedIn: Cloud Insights & Sovereignty CW 31/ 32 · Best of LinkedIn: Digital Transformation & Tech

• 0:00
  Speaker 0: provided by Thomas Allgaier and Frennus, based on the most relevant LinkedIn posts about cloud insights in sovereignty.
• 0:06
  Speaker 0: In calendar weeks thirty one and thirty two.
• 0:09
  Speaker 0: Frenness 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.
• 0:20
  Speaker 0: You can find more info in the description.
• 0:22
  Speaker 1: Right, so imagine you log into your cloud dashboard on a Friday afternoon fully expecting Your standard infrastructure bill
• 0:30
  Speaker 0: like you always do
• 0:31
  Speaker 1: exactly but instead you discover that A single autonomous AI agent which got stuck In this like continuous retry loop On a data extraction task just burn through your entire quarterly compute budget and three hours.
• 0:44
  Speaker 1: Wow
• 0:44
  Speaker 0: Yeah.
• 0:45
  Speaker 1: That is the new reality of AI.
• 0:47
  Speaker 1: fine ops, so we are distilling the top digital products and services trends that are just you know dominating the conversation for tech professionals across LinkedIn right now.
• 0:55
  Speaker 1: yeah
• 0:55
  Speaker 0: in The mission For this analysis Is to give You a well highly practical fluff free blueprint Of the three tectonic shifts happening In cloud architecture Right Now?
• 1:03
  Speaker 1: Because Things Are Moving Fast.
• 1:05
  Speaker 0: They Really Are.
• 1:06
  Speaker 0: We're Unpacking the AI cost reckoning, the geopolitical reality driving sovereign cloud and why the enterprise is pragmatically returning to hybrid architectures.
• 1:20
  Speaker 1: Okay let's unpack this.
• 1:21
  Speaker 1: I mean we have to begin with the bottom line here.
• 1:23
  Speaker 1: absolutely every single enterprise out there is rushing to build and deploy AI into production but traditional cloud cost management is frankly completely failing to track the actual spend.
• 1:36
  Speaker 0: Yeah, Anjakush recently shared this really great analogy about exactly this.
• 1:41
  Speaker 1: Oh right!
• 1:42
  Speaker 1: The wardrobe one.
• 1:43
  Speaker 1: yeah
• 1:43
  Speaker 0: she said.
• 1:44
  Speaker 0: managing your cloud used to be like decluttering your wardrobe.
• 1:47
  Speaker 0: you know for years You just moved everything into the cloud.
• 1:49
  Speaker 0: new project more data.
• 1:50
  Speaker 0: Just shove it in the wardrobe
• 1:51
  Speaker 1: which obviously got messy
• 1:53
  Speaker 0: and expensive.
• 1:54
  Speaker 0: But now with AI, it's like someone just backed up a truck and dumped a thousand completely unlabeled boxes right into the Milvett same wardrobe.
• 2:01
  Speaker 1: It's such a good way to picture that.
• 2:03
  Speaker 1: what's fascinating here is how fundamentally the consumption model has changed.
• 2:08
  Speaker 1: The old Phenops playbook-the one you likely spent the last decade perfecting was built around stable units
• 2:15
  Speaker 0: Like compute hours in storage.
• 2:17
  Speaker 1: Exactly!
• 2:17
  Speaker 1: You paid for compute hours, storage gigabytes...it was all predictable metrics Traditional phenops paying for a rental car by the day.
• 2:24
  Speaker 1: You know exactly what the invoice will be, but AI Phenops is like paying for taxi where the driver... The AI agent can just autonomously decide to take a four hundred mile detour without even asking you
• 2:38
  Speaker 0: and your build per mile?
• 2:39
  Speaker 1: Yep!
• 2:39
  Speaker 1: Build Per Mile.
• 2:41
  Speaker 1: You aren't just managing the resource anymore.
• 2:43
  Speaker 1: uh..you actually have to govern the agents logic itself
• 2:46
  Speaker 0: which explains exactly why the old tracking mechanisms are breaking down
• 2:50
  Speaker 1: Right, and John Schlegel-Milch pointed out the underlying mechanism causing this failure.
• 2:56
  Speaker 1: In the AI world a quote unquote token is not a consistent unit across different providers or models.
• 3:02
  Speaker 1: yeah thousand tokens on one model does not equal a thousand tokens On another.
• 3:07
  Speaker 1: And even more critically The exact same AI task can cost ten times more today than it did yesterday, simply depending on how many times that autonomous agent loops through its internal retries to find an answer.
• 3:22
  Speaker 0: So wait the exact same prompt executed a week later could generate internal reasoning pathway.
• 3:30
  Speaker 0: Precisely!
• 3:31
  Speaker 1: That
• 3:31
  Speaker 0: is terrifying from a governance perspective.
• 3:33
  Speaker 0: It
• 3:33
  Speaker 1: really is, and because of that completely unpredictable execution path organizations are seeing this massive AI expense hitting their invoice but they lack the telemetry to say you know who initiated the task or whether the appropriate model was even used
• 3:48
  Speaker 0: Or what actual business outcome it even returned Exactly.
• 3:50
  Speaker 0: And we're seeing the hyperscalars actively adjust their billing to capture these two.
• 3:55
  Speaker 0: Oh
• 3:55
  Speaker 1: absolutely Yeah.
• 3:56
  Speaker 0: Dope Moore shared some vital specifics regarding Microsoft Copilot Co-Work's new consumption billing.
• 4:02
  Speaker 0: They recently shipped to their architecture from a free preview, to a metered currency system.
• 4:08
  Speaker 0: so your standard thirty dollar month seat license covers standard features.
• 4:12
  Speaker 0: but if you want co work task execution that runs on copilot credits which cost a penny per credit
• 4:18
  Speaker 1: and this is where the scale of enterprise AI becomes just massive financial vulnerability because A penny sounds like absolute margin error.
• 4:27
  Speaker 0: Yeah, it sounds like nothing.
• 4:28
  Speaker 0: but
• 4:29
  Speaker 1: multiply that single penny by thousands of employees each running dozens of autonomous multi step tasks every single day.
• 4:37
  Speaker 0: the spend just scales automatically and silently with adoption.
• 4:41
  Speaker 0: so to deploy this safely you really have to master mapping your line-of-business spending policies directly.
• 4:47
  Speaker 0: two specific Azure subscriptions
• 4:49
  Speaker 1: You have too
• 4:50
  Speaker 0: because if you simply use a single shared credit pool for One department's heavy AI usage say a marketing team generating thousands of images could accidentally drain the pool and
• 5:00
  Speaker 1: completely starve out critical financial analysis tasks in another Department.
• 5:04
  Speaker 0: Exactly, you have to isolate the plast radius of these AI costs.
• 5:07
  Speaker 1: it really requires A total architectural shift in governance And You know It's not just about setting up guardrails against runaway meters either.
• 5:21
  Speaker 0: Yeah, I looked at the cost performance analysis to your.
• 5:24
  Speaker 0: Mizrahi ran on AWS bedrock and The numbers are just a perfect example of this.
• 5:28
  Speaker 0: yeah
• 5:28
  Speaker 1: That was fascinating
• 5:29
  Speaker 0: right.
• 5:29
  Speaker 0: he didn't Just look at the billing dashboard.
• 5:31
  Speaker 0: his team actually merged cloud watch Performance metrics with their building data To see what they were you know really buying?
• 5:38
  Speaker 0: And
• 5:38
  Speaker 1: What did they find?
• 5:39
  Speaker 0: They found that Their us regional endpoint Was carrying A ten percent Cost premium over the global end point But would Actually sit out to me wasn't the Premium?
• 5:48
  Speaker 0: it was the latency?
• 5:50
  Speaker 0: Oh, yeah.
• 5:50
  Speaker 0: They found a four and half times P-Ninety nine to average tail latency ratio on that premium endpoint
• 5:58
  Speaker 1: which is such a critical metric To understand.
• 6:01
  Speaker 1: I mean in plain English A high p ninety nine tail latency Ratio means That while ninety nine percent of your AI requests might execute In normal time frame that one percent Of outliers you know the worst case scenarios are taking Four and a half Times longer to
• 6:16
  Speaker 0: execute And an AI workload.
• 6:19
  Speaker 0: If an agent gets stuck waiting on one of those extreme outliers to process, your token meter might just keep spinning or you're dependent applications timeout.
• 6:28
  Speaker 0: You are literally paying a premium for end point that is significantly less predictable under load.
• 6:33
  Speaker 1: Exactly and the team didn't need some magical new Tokenomics framework applied core engineering discipline, pairing a financial post-metric with the technical performance metric.
• 6:44
  Speaker 0: So what does this all mean for the actual Phenops teams on the ground?
• 6:47
  Speaker 0: Are they just like token counters now?
• 6:49
  Speaker 1: Well as Max Gill noted, Phenop's teams are actively evolving into first responders for AI deployments.
• 6:55
  Speaker 0: First responders?
• 6:56
  Speaker 0: I liked that?
• 6:57
  Speaker 1: Yeah because before security team or executive board even knows who owns specific AI initiative The budget impact lands at desk of people managing cloud spend.
• 7:07
  Speaker 0: So, it's shifting the focus.
• 7:08
  Speaker 0: Right!
• 7:09
  Speaker 1: Shifting focus from tracking baseline cost of compute to calculating actual business value per AI outcome is now number one skill these teams need to build.
• 7:19
  Speaker 0: It's moving from what did this cost?
• 7:22
  Speaker 0: To was specific agent output worth token burn?
• 7:26
  Speaker 1: Precisely
• 7:27
  Speaker 0: By the way, if you are finding this deep dive helpful as you architect your own infrastructure make sure to hit subscribe so that we don't miss our future analyses on how these tech-conic trends evolve.
• 7:37
  Speaker 1: Good call!
• 7:38
  Speaker 1: And y'know...if we connect it with a bigger picture optimizing those AI endpoints globally introduces massive legal trap because you might find cheaper faster server in different regions but routing data there may violate national law which brings us into the geopolitical reality of sovereign cloud.
• 7:55
  Speaker 0: Here's where it gets really interesting, because suddenly what your data physically sits is no longer just some mundane IT question for network architect to answer.
• 8:05
  Speaker 0: No not
• 8:05
  Speaker 1: at
• 8:06
  Speaker 0: all!
• 8:06
  Speaker 0: It has a boardroom mandate but let me push back on this bit right?
• 8:10
  Speaker 0: Isn't Sovereign Cloud just marketing buzzword.
• 8:13
  Speaker 0: hyperscalars used to sell localized servers.
• 8:15
  Speaker 1: You know its tempting to write off as Marketing But the conversation fundamentally shifted.
• 8:21
  Speaker 1: Bader Alaw by Don mapped out four converging forces that prove this is a permanent structural change in how the internet operates.
• 8:28
  Speaker 1: Okay First, geopolitics Infrastructure you do not legally and physically control Is well leverage.
• 8:35
  Speaker 1: someone else holds over your economy Right.
• 8:38
  Speaker 1: Second data is officially recognized as a national asset.
• 8:42
  Speaker 1: It's The new digital oil
• 8:43
  Speaker 0: Which feeds directly into his third force right?
• 8:46
  Speaker 0: Yeah The global AI race.
• 8:47
  Speaker 0: Exactly Because if you are exporting Your nation's raw data for free sending it across the ocean to train foundational models governed in another country, you are actively losing your competitive edge.
• 8:57
  Speaker 0: Yeah You were basically shipping out the crude oil and buying back the refined gasoline at a massive premium.
• 9:03
  Speaker 1: That's exactly that.
• 9:04
  Speaker 1: And The fourth force is regulatory action.
• 9:06
  Speaker 1: Governments have stopped issuing gentle advisory guidance.
• 9:10
  Speaker 1: They're now enforcing strict legal mandates regarding data residency and operational autonomy.
• 9:17
  Speaker 0: We see real world boardroom panic resulting from this too.
• 9:21
  Speaker 0: Jory Malmberg highlighted a massive shockwave in the European market recently.
• 9:26
  Speaker 1: Oh, The Dutch Central Bank?
• 9:28
  Speaker 0: Yes!
• 9:28
  Speaker 0: The Dutch central bank moved its core cloud infrastructure to a platform owned by the Schwartz Group which is literally the parent company of the little supermarket chain
• 9:38
  Speaker 1: Which just sounds completely absurd on the surface like why's a central bank trusting a supermarket with it data?
• 9:44
  Speaker 0: Right It sounds wild But when you look at the mechanism that makes perfect sense.
• 9:48
  Speaker 0: European retailers generate massive amounts of data, but they refuse to host it on U.S.
• 9:53
  Speaker 0: hyperscalers
• 9:54
  Speaker 1: because They don't want a hand their operational intelligence To the same tech giants that operate competing retail businesses.
• 10:00
  Speaker 0: exactly so The Schwartz group simply built their own enterprise-grade cloud called stackit.
• 10:06
  Speaker 0: It's full european fully sovereign.
• 10:08
  Speaker 0: Wow and when the Dutch central bank made the move every board in Europe read the news, and instantly asked their CTOs why they weren't securing their data.
• 10:18
  Speaker 0: The exact same way
• 10:19
  Speaker 1: Sovereign Cloud became a boardroom imperative in like a matter of months
• 10:23
  Speaker 0: And it's happening at the highest levels of industrial manufacturing too.
• 10:27
  Speaker 0: Yeah We saw announcements from Damia Lucas Anil Morali and Christina Weber that Airbus has officially selected Scaleway & SAP Rise to build out there sovereign cloud foundations
• 10:39
  Speaker 1: Which makes sense.
• 10:39
  Speaker 1: Right,
• 10:39
  Speaker 0: because when you are designing classified aerospace technology You simply cannot risk a foreign entity legally compelling access to your digital supply chain.
• 10:48
  Speaker 1: Absolutely.
• 10:49
  Speaker 1: But we do have to provide a crucial architectural reality check here Because there is a lot of misunderstanding about what sovereignty actually means technically speaking?
• 10:58
  Speaker 0: Yeah for sure.
• 10:59
  Speaker 1: Daniel Bacellic broke this down brilliantly in his analysis.
• 11:03
  Speaker 1: A lot of enterprise leaders assume that avoiding foreign government access, specifically the U.S.
• 11:08
  Speaker 1: Cloud Act is simply a matter of geography
• 11:11
  Speaker 0: like they just think.
• 11:12
  Speaker 0: Just put this server in Frankfurt or Paris and we are safe.
• 11:15
  Speaker 0: right.
• 11:16
  Speaker 1: but it's not about The postcode isn't?
• 11:17
  Speaker 0: no?
• 11:18
  Speaker 0: It is entirely about the architecture.
• 11:20
  Speaker 0: We have to look at this impartially.
• 11:22
  Speaker 0: the US cloud act requires a warrant from a judge tied To his specific investigation its illegal design.
• 11:29
  Speaker 0: But more importantly, look at how something like the AWS European Sovereign Cloud actually functions.
• 11:35
  Speaker 0: Yes, the servers operate physically within Europe but The critical mechanism is that encryption keys are controlled locally by a completely separate EU entity.
• 11:45
  Speaker 0: No US person and no us based Amazon employee has logical or physical access to customer data.
• 11:52
  Speaker 1: So even if a foreign government legally compels the cloud provider to hand over data, all they get is encrypted gibberish.
• 11:58
  Speaker 0: Exactly!
• 11:58
  Speaker 0: A legal request against data you literally cannot decrypt or reach... ...is unenforceable by design.
• 12:04
  Speaker 0: The core insight here Is that sovereignty is cryptographic problem Not geographical one.
• 12:09
  Speaker 0: It's key management policy not just border on map.
• 12:13
  Speaker 1: It's math, not just laws.
• 12:15
  Speaker 1: If they don't have the key and data regardless of where the parent company's headquarters happens to be.
• 12:21
  Speaker 1: And this realization explains third massive shift we are analyzing today.
• 12:25
  Speaker 1: because enterprise organizations now demand strict cryptographic data sovereignty Because they absolutely need localized AI cost controls To prevent runaway token.
• 12:36
  Speaker 1: spend The dream a public cloud only world is officially dead.
• 12:41
  Speaker 1: Wow!
• 12:41
  Speaker 0: The reality
• 12:42
  Speaker 1: for the modern enterprise is deeply, undeniably hybrid.
• 12:46
  Speaker 0: Wait didn't the tech industry just spend the entire last decade trying to escape the on-prem data center?
• 12:51
  Speaker 0: Yeah We spent billions migrating workloads out of the basement and now you're telling me we are really going back.
• 12:56
  Speaker 1: If we connect this to the bigger picture it feels like a massive step backward but It's actually an necessary evolution.
• 13:03
  Speaker 1: Brian White pointed out that enterprises aren't just building one unified future right now.
• 13:07
  Speaker 1: They're effectively funding three parallel operating models simultaneously, you have your traditional virtual machines running the core legacy business logic.
• 13:17
  Speaker 1: You have your containerized environments driving modern stateless cloud native applications.
• 13:22
  Speaker 1: and Now you have AI quickly creating a third incredibly resorts hungry compute model.
• 13:29
  Speaker 0: And you simply cannot shove all three of those models into a single public cloud bucket and expect it to be efficient or secure, they have drastically different needs for data gravity latency in compliance
• 13:40
  Speaker 1: Exactly!
• 13:41
  Speaker 1: And your FunCon shared the perfect example how this pragmatic shift is playing out at scale.
• 13:46
  Speaker 1: AT&T is currently undergoing massive migration on their critical workloads to AWS Outposts.
• 13:51
  Speaker 0: This completely shatters that myth.
• 13:54
  Speaker 0: moving off-premises By moving to Outposts, AT&T is literally having AWS drop fully managed hyperscaler server racks directly inside AT&Ts own legacy data centers.
• 14:05
  Speaker 1: It separates the control plane from the data plane.
• 14:08
  Speaker 1: The developers get the exact same suite of cloud native tooling they're used to, the EC-II compute environments and S-III storage APIs And manage it all through a standard AWS console.
• 14:21
  Speaker 1: But physical data never leaves AT&T's highly regulated physically secure buildings.
• 14:26
  Speaker 0: So you got software agility without ever compromising on your physical data residency or network latency.
• 14:33
  Speaker 1: But there is a massive catch to this hybrid reality.
• 14:37
  Speaker 1: Oh?
• 14:37
  Speaker 1: Yeah, when your data is constantly moving between local edge devices on-premise datacenters and the global public cloud Your core networking fundamentals become more critical than ever.
• 14:48
  Speaker 1: Right?
• 14:48
  Speaker 1: The plumbing?
• 14:49
  Speaker 1: Both Vishakasadwani and Timur Ijlal emphasized this recently.
• 14:53
  Speaker 1: Every year, the industry obsesses over a new framework or a new generative model.
• 14:57
  Speaker 1: But at the end of day packets still need to get from point A to point B securely.
• 15:01
  Speaker 1: when your DNS breaks so you're routing tables fail absolutely nobody cares how advanced your AI agent is.
• 15:06
  Speaker 0: The plumbing fundamentally has.
• 15:08
  Speaker 1: And to truly understand the staggering mechanical scale of modern networking, look at the data Amaya Kumar Jaganathan shared regarding Amazon CloudFront's performance during World Cup final.
• 15:19
  Speaker 0: Oh I saw those numbers and they are honestly difficult even comprehend!
• 15:23
  Speaker 0: They delivered a peak of one hundred seventeen terabits per second of traffic.
• 15:27
  Speaker 0: One hundred
• 15:27
  Speaker 1: and seventeen terabytes PER SECOND!
• 15:30
  Speaker 1: It is
• 15:30
  Speaker 0: insane!
• 15:31
  Speaker 1: Across over forty different global broadcasters simultaneously...
• 15:35
  Speaker 0: ...and didn't achieve that by just building one massive super server in the middle of the ocean somewhere.
• 15:41
  Speaker 1: No, The mechanism behind it was flawless edge networking.
• 15:44
  Speaker 1: they embedded points-of-presence basically highly optimized local caching servers directly inside Local internet service provider networks around the world.
• 15:54
  Speaker 0: right.
• 15:55
  Speaker 1: so when millions of fans requested the exact same high definition video segment Of a goal being scored at the exact Same millisecond that data didn't have to travel across an Ocean submarine cable.
• 16:05
  Speaker 1: Oh It was served from a box sitting just a few miles from their homes, entirely bypassing the global internet backbone.
• 16:12
  Speaker 0: Which is just the ultimate architectural lesson for this hybrid reality!
• 16:16
  Speaker 0: You need the Global Intelligence and Centralized Management of The Cloud to coordinate the massive scale but you absolutely need localized edge-level execution... ...to deliver performance & maintain legal control.
• 16:29
  Speaker 0: Exactly!
• 16:30
  Speaker 0: Whether your are delivering a four K video stream or soccer match or running a highly sensitive AI agent analyzing sovereign financial data, the physical architecture has to map exactly to this specific use case.
• 16:42
  Speaker 1: And this raises an important question as we look at where all of these converging trends are heading?
• 16:47
  Speaker 1: We have talked extensively about the enterprise obsession with tracking token spend optimizing free knobs and securing hybrid infrastructure but what happens when these autonomous AI agents become so advanced that they start negotiating in provisioning their own hybrid cloud resources?
• 17:03
  Speaker 0: That is a wild thought.
• 17:05
  Speaker 0: AI agents dynamically buying and spinning up their own servers based on latency requirements
• 17:11
  Speaker 1: Exactly, And furthermore analysts are increasingly pointing out that human labor currently makes Fifty to one hundred percent of the true investment in AI today.
• 17:20
  Speaker 1: You know, data preparation pipeline engineering compliance checks right.
• 17:24
  Speaker 1: if we are entirely fixated on tracking the pennies of token spend On a fine-offs dashboard Are we completely missing?
• 17:31
  Speaker 1: The forest for the trees?
• 17:33
  Speaker 1: or we hyper focused on the wrong side Of the ledger while the real costs they're just hiding in plain sight?
• 17:39
  Speaker 0: It really brings us right back to Anja Kush's messy wardrobe analogy.
• 17:43
  Speaker 0: If you do not have your infrastructure organized, Your sovereign boundaries defined and spending policies mapped out today What happens when the AI decides it wants to start going shopping for new compute resources on your corporate credit card?
• 17:55
  Speaker 0: Right You'll get a foundation right before agents take the wheel Absolutely!
• 17:59
  Speaker 0: If you enjoyed this episode New episodes drop every two weeks.
• 18:03
  Speaker 0: Also check our other editions on digital products & services Defense tech AI and agentic systems green ICT and sustainable AI, and health tech.