Best of LinkedIn: Cloud Insights CW 21/ 22

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

00:00:07: And twenty two.

00:00:08: Frenes 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.

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

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

00:00:22: Really excited to have you with us today as we dig into some pretty massive shifts happening right now.

00:00:26: Okay,

00:00:27: let's unpack this because the conversations were seeing across the ICT and tech industry on LinkedIn are just.

00:00:34: They are intense.

00:00:35: We're looking at a fundamentally different landscape.

00:00:37: Oh, absolutely the old playbooks are well they basically being thrown out of window right.

00:00:41: so our mission today is to cut through The noise and look at three major trends dominating calendar weeks twenty one.

00:00:47: in twenty two we were going To Look At New Geopolitical Reality Of Sovereign Cloud How Finoffs Is Evolving From Just You Know Passive Dashboards Into Actual Engineering Workflows.

00:00:57: And Finally This Sudden Really Harsh Reality Check Around AI Token Billing.

00:01:02: Yeah, that AI token bill is.

00:01:04: it's definitely causing some panic in the finance departments right

00:01:07: now.

00:01:07: Oh for sure.

00:01:08: but let's start with sovereignty because cloud architecture is no longer just an IT discussion?

00:01:14: Is

00:01:15: not at all?

00:01:15: I mean setting up a cloud environment today is essentially foreign policy executed through APIs.

00:01:20: its geopolitics

00:01:22: Right.

00:01:22: It makes me think of Imagine you rent this highly secure state-of the art vault in Europe, right?

00:01:29: You've got biometric locks zero trust laser grids.

00:01:32: The works yeah, you feel totally secured.

00:01:34: Yeah impenetrable

00:01:36: Exactly.

00:01:37: but then you find out the contractor who built the Vault is headquartered in the US and Because of that they are legally obligated by their local government to keep a master key.

00:01:46: Wow And under certain laws They could be forced use that key completely bypassing whatever European laws you thought were protecting your IP.

00:01:54: That is a perfect analogy.

00:01:56: And it's not a hypothetical movie plot either, that is the exact tension keeping enterprise architects awake right now.

00:02:03: Roger S and Emily K brought up some fantastic points about this on LinkedIn.

00:02:06: Yeah The illusion of geographic boundaries has basically shattered.

00:02:09: We used to think you know just build a data center in Frankfurt keep the data in Germany You check the compliance box Right

00:02:16: simple data residency.

00:02:18: But roger s points out that jurisdiction beats geography every single time.

00:02:23: he specifically brought up the U.S.

00:02:25: Cloud

00:02:25: Act, which is a huge deal.

00:02:27: it's massive because The cloud act explicitly says us law enforcement can compel any American company to hand over data they control regardless of where that physical server sits on the planet.

00:02:39: so if the parent companies in Seattle The U.S.

00:02:41: courts view a server in Frankfurt exactly the same way they view a Server in Virginia?

00:02:46: Exactly, and Emily Kaye pointed out how this totally undermines things like the AWS European Sovereign Cloud model.

00:02:53: AWS is aggressively pushing that showing off new European hospital-and utility customers.

00:02:58: But the service is still one hundred percent owned by Amazon.com Inc.

00:03:02: Creating legal separation on paper, like spinning up a European subsidiary doesn't actually grant legal immunity in practice if the ultimate parent company gets a foreign subpoena.

00:03:11: That puts Europe at really precarious position.

00:03:13: but I mean they can just unplug from US hyperscalars tomorrow.

00:03:17: The infrastructure gap is way too big.

00:03:19: No of course not.

00:03:20: you cant slip and switch.

00:03:22: The EU strategy is shifting.

00:03:24: Guido Hackenberg shared some really crucial updates on the upcoming Cata rules, that's the Cloud and AI Development Act from the European Commission.

00:03:31: Yeah those rules are going to tie sovereign cloud eligibility directly to providers.

00:03:38: legal entity in ownership right?

00:03:39: Yes

00:03:40: they're linking it directly to ownership Which means if Cata enforces these strict entity requirements, we could literally see major US hypersailers pushed out of highly sensitive EU government workloads.

00:03:53: We're talking healthcare defense judicial data.

00:03:56: So the question changes from where's the physical hard drive to?

00:04:02: If a foreign warrant lands on the parent company's desk, are you forced to hand over our data?

00:04:06: Exactly.

00:04:06: And this is already impacting the market.

00:04:08: Evan Kirstel highlighted a massive M&A ripple effect in The Netherlands recently.

00:04:12: Oh

00:04:12: right!

00:04:12: The Kindrel and Solvenity deal.

00:04:14: Yeah...the

00:04:15: Dutch government blocked Kindrel from acquiring solvenity…and Kindrel is no joke.

00:04:21: they have deep IBM Enterprise DNA.

00:04:24: But Solventy operates Digid which is digital identity backbone for Dutch citizens.

00:04:29: So it's critical national infrastructure,

00:04:31: right?

00:04:32: And blocking that deal proves that sovereignty today is entirely about jurisdictional ownership.

00:04:38: if we connect this to the bigger picture I know That something you always say true sovereignty is really about operational control under extreme pressure.

00:04:45: Exactly If We Connect This To The Bigger Picture It'S About Control.

00:04:48: Robert Felsen and Anne Leslie had this brilliant exchange about it.

00:04:52: fousin says sovereignty doesn't start at the database layer

00:04:58: Meaning like identity and telemetry.

00:05:00: Yeah, how identity is established?

00:05:02: How metadata is handled?

00:05:03: because if your core workload Is on a sovereign server but you're telemetery You know the data detailing who is logging in And when it's routed through a global network In a non-sovereign jurisdiction

00:05:15: then your whole security posture is compromised

00:05:17: Exactly.

00:05:18: Telemetry leakage defeats the whole purpose, and Anne Leslie took it a step further.

00:05:23: She challenged this obsession we have with redundancy.

00:05:26: Multi-region architectures don't protect you if The real threat is A geopolitical disruption of the shared control plane.

00:05:32: So It's not just about can We technically migrate our data?

00:05:35: No?

00:05:35: It's About do You Have the legal And technical authority to execute an exit strategy when things go wrong Without a foreign jurisdiction?

00:05:43: locking your admin accounts?

00:05:45: Wow Yeah, that really reframes the whole issue.

00:05:48: And we are seeing industry react with massive capital deployment.

00:05:52: Elke and Andrew posted that telecom an SAP just won a two hundred fifty million euro contract to build dedicated AI cloud for the German government.

00:06:01: Yeah clear signal that European states or funding infrastructure where they actually hold the master key... ...and

00:06:06: it's not just them!

00:06:07: Erindam Choudhury mentioned TCS launching sovereign secure Cloud for Europe.

00:06:11: Dr Marian Janek highlighted Google Cloud and Thales signing a memo for an independently operated German region.

00:06:18: Right, the hyperscalers are realizing they need localized partners to provide what George Politis called Silicon Trust actual hardware backed verifiable exit paths.

00:06:28: but building all this highly localized hardware back infrastructure that burns in incredible amount of capital

00:06:33: oh it is staggeringly expensive

00:06:36: which transitions us perfectly into our second theme because and looking at the LinkedIn consensus from weeks twenty-one to twenty two, traditional fan offices basically broken.

00:06:48: Yeah!

00:06:48: The traditional dashboard approach is just.

00:06:50: it's dead... It separates financial insight from engineering action.

00:06:54: Ankoj Madan made this brilliant argument for a total shift in methodology.

00:06:58: Because

00:06:58: right now what happens?

00:06:59: Finance sees a twenty five percent spike at end of month.

00:07:02: Panics and emails engineering to go hunt for abandoned resources.

00:07:06: Exactly, it's entirely reactive.

00:07:08: Madonna argues that Finoffs needs to shift left.

00:07:11: It belongs embedded directly inside internal developer platforms Inside the pull request.

00:07:16: So you embed namespace quotas and budget alerts directly into the PR?

00:07:20: Yeah If an engineer tries to spin up resources That blow past a budget The PR gets flagged right there.

00:07:27: They self-correct before the meter even starts running!

00:07:30: That makes so much sense.

00:07:32: But even when teams try to actively manage costs, aren't they often using tools that just misunderstand the environment?

00:07:39: Rogavendra R brought up a huge blind spot regarding Azure and AWS.

00:07:43: Oh, this is such an expensive mistake!

00:07:46: Rogavandra pointed out that third-party phenops dashboards frequently treat azure mechanics exactly like aws mechanics which

00:07:52: I mean on the surface sounds logical.

00:07:54: They're both cloud compute platforms

00:07:56: right but under the hood The discount structures are completely different.

00:07:59: aws relies heavily on compute commitments Like savings plans But azures biggest lever Is the azure hybrid benefit?

00:08:06: Which is software license portability thing?

00:08:08: right exact

00:08:08: Exactly.

00:08:09: It lets you take the on-premise Windows and SQL Server licenses that already own, apply them to cloud.

00:08:14: It can drop your pay as you go rate by up to eighty five percent.

00:08:18: Wait... Eighty Five Percent?

00:08:20: Yes!

00:08:21: But if your central dashboard is normalizing everything look like AWS.

00:08:25: it ignores portability so your biggest potential discount remains totally buried.

00:08:31: That leads me ask are Finops teams actually celebrating wrong savings?

00:08:36: Because Mohammed Bruechner mentioned teams cheering a ten percent compute discount while completely ignoring shelfware.

00:08:42: Yeah, the double-paying problem!

00:08:44: Bruekner calls it technology value management.

00:08:47: If your engineers don't configure those Azure Hybrid benefit tags in their infrastructure as code Microsoft just charges you with full on demand rate

00:08:55: Which includes this software license fee.

00:08:57: Exactly So are renting software by the hour that your procurement team already bought a permanent license for Ouch?

00:09:03: That hurts just to think about.

00:09:05: And then Nicholas Fundrini brought up another hidden cost, security fine ops.

00:09:08: Telemetry and sign costs are apparently scaling faster than the infrastructure

00:09:12: itself.".

00:09:13: Yeah because with microservices and zero trust every container is constantly talking and authenticating... ...and compliance says you have to log it all.

00:09:21: so you pay to ingest it and you pay again to store it!

00:09:24: But there was a really tactical fix shared for this right?

00:09:27: by Alan Visser.

00:09:29: Yes, Allen Visser suggested using XPath filtering right at the source using the Azure Monitor agent.

00:09:35: so instead of capturing all the noise and filtering it in the cloud you filter it at the node?

00:09:39: Precisely!

00:09:40: It's like putting a bouncer at the club door.

00:09:42: instead of bussing everyone to the Club just to kick ninety percent of them out You drop the noise before it hits your Siam workspace.

00:09:49: That is such a good analogy And it proves that if we can't react to a bill We have shape architecture Like Satya Brata Chowdhury's example with AI fine-tuning.

00:09:59: Oh,

00:09:59: the GPU request example?

00:10:01: Yes!

00:10:01: He outlined a scenario where data science team asks for eight H one hundred GPUs For two week fine tuning job.

00:10:08: Normally an automated platform just provisions that and tags it for billing.

00:10:11: right

00:10:11: Right but Satya brought up points out that fennups needs to step in And say hey have you considered A low rank adapter or lower I instead of full parameter fine tuning?

00:10:20: What does actually do to cost?

00:10:21: Elora freezes the original model weights and only trains a tiny matrix.

00:10:26: It can cut compute costs by five times or more before single workload even spins up.

00:10:30: Wow,

00:10:31: and shaping that spin Before it starts is critical now because that brings us to our final theme.

00:10:36: The AI experimentation phase is ending And the token bill has arrived!

00:10:41: It

00:10:41: really has...the flat rate era just slammed into a wall.

00:10:45: Rami Algafaji shared this shocking reality check.

00:10:48: Microsoft, who basically has infinite cloud capacity had to cancel their own internal Claude code licenses because the token billing became untenable.

00:10:56: Yeah and he also mentioned that report about Uber.

00:10:58: Their CTO warned staff they had burned through their entire twenty-twenty six AI budget in just four months.

00:11:04: Four months.

00:11:05: If Uber and Microsoft are hitting the brakes on usage-based pricing, that is a massive structural issue.

00:11:11: It is!

00:11:11: Token Pricing forces you to confront the real microeconomic cost of generative AI—and it's not just heavy model training either.

00:11:18: Venkat Reddy Chintalapudi warned about embedded AI Cost Cree.

00:11:22: Oh right…the thirty dollars per month per user for co-pilot addons

00:11:27: Exactly.

00:11:28: Enterprises are paying to upgrade thousands of employees, but only a fraction actually use the tools enough to justify the

00:11:35: cost.

00:11:35: It's like outfitting your entire company with sports cars When ninety percent of them Only need to drive To the grocery store down The street.

00:11:42: it just kills Your margins.

00:11:44: That is exactly.

00:11:45: But what's fascinating here is it not just financial risk, there are operational landmines.

00:11:50: Elliott Layton Woodruff warned that Microsoft is auto upgrading existing Azure Open AI resources to their new Foundry platform

00:11:57: which sounds like a marketing rebrand but its'nt

00:11:59: No!

00:12:00: Its fundamental shift in the resource provider.

00:12:03: It changes resource IDs and networking configs.

00:12:05: so if you use Terraform or Bicep and don't update your code Your pipeline will detect massive drift

00:12:11: Meaning your automation scripts might interpret that drift as an error and automatically delete your production AI environment.

00:12:17: Yes, it might tear down to recreate the old configuration... That

00:12:21: is terrifying!

00:12:23: Just the idea of consuming AI isn't enough.

00:12:26: you have strictly governed operations?

00:12:28: Which echoes exactly what Kevin Davis in Kashyap-Ai were saying.

00:12:32: The battle isn't about which model is smartest anymore.

00:12:34: It's about operating models

00:12:36: because organizations rushed to public cloud AI, got hit with the token bill and are now pivoting hard.

00:12:43: Yes a massive pivot towards small language models SLMs on private Cloud.

00:12:48: John Yoho cited a Goldman Sachs forecast predicting enterprise token demand will explode twenty four times by twenty thirty.

00:12:56: You literally cannot run twenty-four times the volume on a variable public cloud meter.

00:13:00: It would bankrupt you!

00:13:01: Exactly, that's why we're seeing demand for things like HPEs.

00:13:05: new private Cloud AI stack.

00:13:08: companies want on premise closed loop systems to escape the hyperstalar meter which

00:13:13: perfectly ties back our very first theme digital sovereignty.

00:13:16: Getting off the public meter is just another way of maintaining operational control over your own destiny.

00:13:21: It all connects, and it brings up a really profound point that Greg Powell shared in our sources.

00:13:26: Oh

00:13:26: I love this point!

00:13:27: Yeah he noted that over the last decade The cloud abstracted away the infrastructure layer.

00:13:32: you didn't have to rack servers anymore.

00:13:34: And now AI is abstracting away the execution layer... ...it turns right this block of code into build-this app

00:13:42: Right.

00:13:43: so if Cloud handles the Infer and AI handles the coding execution, what is actually left for the ICT professional over the next decade?

00:13:52: That's a big realization.

00:13:54: The true differentiator won't be pure technical deployment – it will be mastering geopolitical governance and microeconomics of these systems.

00:14:01: So

00:14:02: understanding how a US subpoena hits an European database… or configuring an Azure hybrid tag so you don't double pay for licenses.

00:14:09: Exactly, Or knowing when to suggest a lore I to an engineer before they rack up massive token bill.

00:14:15: Going back that vault analogy it really doesn't matter how thick the walls are.

00:14:19: if You do understand legal jurisdiction of master key and you Don't actively control microeconomic cost of electricity keeping alarms on

00:14:27: roll Then aren't actually in control of the Vault?

00:14:30: Exactly The future belongs people who control operating model and economics

00:14:34: Spot on, that's the real takeaway

00:14:57: here.