RAMmageddon – DigitalWasteland

Using an anonymized, archetypal mid-sized machine builder, Jens shows why AI initiatives in the mid-market rarely fail because of the models, and almost always because of the data architecture. He proposes a lakehouse as an additional layer on top of the grown data warehouse, and gives an honest reality check: federation alone saves no licenses at all. The savings come later, from gradually shifting workloads away from Oracle.

In this episode

• Three typical findings: over eighty percent read load on the Oracle data warehouse, report wait times of several working days, and an AI initiative stuck because the data isn't accessible
• The lakehouse as a federating layer: Apache Iceberg, Parquet, S3-compatible object storage, Dremio or Starburst, dbt, Airflow, a vector index and a locally run open-weight model
• Why "federation alone saves exactly zero licenses", and how cancelling Oracle options, edition downgrades and engine swaps can cut database license costs by thirty to fifty percent over eighteen to twenty-four months
• How the AI use case gets into production within weeks once the data foundation is in place
• Sovereignty as architecture: data residency, open-weight models and portability
• Three consequences for anyone planning a refresh in the second half of 2026

Originally published in German on May 28, 2026: https://klasen.ai/briefing/rammageddon-teil-7-ki-tueroeffner/

English version: https://klasen.ai/en/briefing/rammageddon-teil-7-ki-tueroeffner/

Voices are AI-generated (Jens's voice is a clone of his own). Content and analysis by Jens Klasen.

In many environments, the most expensive part of a database server isn't the hardware, it's the SQL Server license that comes due every year. Jens explains how dozens of small, barely used SQL VMs, wrong editions and per-core minimums add up to a high five- to six-figure bill per year in typical mid-market environments. And why consolidation is a lever that works without buying a single new server.

In this episode

• The typical pattern: dozens of small SQL Server VMs with average CPU load below fifteen percent
• The fine print: per-core licensing in two-core packs, with a minimum of four core licenses per VM
• Classic mistakes: dev and test on paid editions, Enterprise where Standard would do, passive replicas without Software Assurance
• Three levers: consolidate, check editions, use Software Assurance deliberately
• Why license savings are the main point and hardware savings the side effect: "Stop paying for air"

Originally published in German on June 9, 2026: https://klasen.ai/briefing/rammageddon-sql-lizenzkonsolidierung/

English version: https://klasen.ai/en/briefing/rammageddon-sql-lizenzkonsolidierung/

Voices are AI-generated (Jens's voice is a clone of his own). Content and analysis by Jens Klasen.

Modern software protects itself; older software relies on the infrastructure to protect it. In this short episode, Jens explains why that one difference decides what your hardware has to look like, and why it gets skipped in almost every modernization discussion. He describes a node design that serves both worlds on the same hardware foundation.

In this episode

• Two kinds of software with opposite demands: cloud-native applications versus classic single-instance software
• One hardware foundation, two protection models: NVMe-dense nodes as vSAN ESA hosts or as Kubernetes workers with Portworx
• Bringing local NVMe, a FlashBlade object tier and existing Dell EMC systems behind one consistent data layer
• Protection on demand instead of the watering can, and reacting to storage supply shortages without rethinking the stack
• Why you don't have to choose between old and new while running VMware for a while longer

Originally published in German on June 16, 2026: https://klasen.ai/briefing/software-schuetzt-sich-selbst/

English version: https://klasen.ai/en/briefing/software-schuetzt-sich-selbst/

Voices are AI-generated (Jens's voice is a clone of his own). Content and analysis by Jens Klasen.

Disclosure: This briefing mentions Portworx. Jens has worked for Everpure (formerly Pure Storage) and the Portworx team since September 16, 2026 – this briefing was written before he joined.

On July 1, 2026, IBM raised prices on FlashSystem by 40 percent and on Power servers by 17 percent. For Jens, that was the moment the memory crisis moved from a procurement question to the incoming invoices. In this episode he walks through the market update, explains why Apple's price hikes matter for enterprise IT, and ties together three threads from earlier briefings: license costs, self-protecting software and AI-assisted modernization.

In this episode

• Why DRAM contract prices jumped 58 to 63 percent in a single quarter, and why manufacturers are falling back to DDR3 and even DDR2
• How the crisis reached consumers: Apple, Xbox and smartphones, and why that helps your business case with the CFO
• The stock market paradox, and why efficiency is "the only bet without a downside"
• Three threads, one story: license bills, cloud-native efficiency and vibe coding
• The extended five-step formula: extend support, check licenses, rethink architecture, pilot modernization, then order in a targeted way

Originally published in German on July 2, 2026 (no English web version for this part): https://klasen.ai/briefing/rammageddon-teil-8-rechnungseingang/

Voices are AI-generated (Jens's voice is a clone of his own). Content and analysis by Jens Klasen.

Kubrick's black monolith is now on your kitchen table. It's green, nine centimeters tall, and called DDR5. In July 2026, the memory crisis left the data center and showed up as a price tag for everyone who owns a device. Jens explains why this is structural rather than cyclical, why the real issue is control, and what Lenovo's RAMageddon Survival Guide recommends.

In this episode

• DDR5 prices as of July 2026, and the forecasts from Gartner and Jefferies
• The consumerization of visibility: not the technology is democratized, but its scarcity
• Why this isn't a cycle: capacity shifting to AI, Micron's long-term contracts, and a class action against the big three memory makers
• HAL 9000 and the procurement logic of the hyperscalers: are you Dave, or the early humans?
• Five steps from Lenovo's RAMageddon Survival Guide, including moving memory-hungry workloads to GPUs
• Why cloud native competence turns a hardware shortage into a planning task

Originally published in German on July 21, 2026: https://klasen.ai/briefing/rammageddon-teil-9-kuechentisch/

English version: https://klasen.ai/en/briefing/rammageddon-teil-9-kuechentisch/

Voices are AI-generated (Jens's voice is a clone of his own). Content and analysis by Jens Klasen.

In Frank Herbert's Dune, it's not price that decides who gets Spice, but allocation. In August 2026, according to an unconfirmed DigiTimes report, the big memory makers had already fully allocated their DRAM and HBM capacity for 2027. Jens explains the allocation economy, why the private cloud versus hyperscaler debate is the wrong question, and why cleaning up your own DigitalWasteland is now a procurement strategy.

In this episode

• The allocation economy: long-term contracts, prepayments and quotas of sixty to seventy percent of requested volume
• Why hyperscalers sit on the right side of the allocation, and why the Guild doesn't fly for free
• The real question: will you still be able to decide for yourself in 2027?
• The stillsuit principle: not getting more, but losing less, and why the DigitalWasteland is self-made
• Decoupling data from the platform, and five questions to test whether that layer is any good
• Acting from a position of control, not from pressure

Originally published in German on August 13, 2026: https://klasen.ai/briefing/rammageddon-teil-10-spice-ist-verteilt/

English version: https://klasen.ai/en/briefing/rammageddon-teil-10-spice-ist-verteilt/

Voices are AI-generated (Jens's voice is a clone of his own). Content and analysis by Jens Klasen.

Five good fitness apps, five silos, five foreign assumptions about how Jens should train. So he built IDUN, a personal AI coach for training, nutrition, recovery and longevity, with Claude Code and the Claude API, running entirely on his own devices. Jens walks through why software for an audience of one has become economically possible, why the app will never be finished, and how this mirrors the memory crisis from the other end of the stack.

In this episode

• Why none of five quality apps worked, and what IDUN does differently, from morning readiness checks to rep counting with self-soldered sensors
• No backend, no server, no vendor account: privacy as an architecture decision
• The math: around fifteen euros a month in API costs versus around twenty-seven euros in subscriptions it replaces
• Why fixed scope, completion date and depreciation no longer apply, and the hidden costs: scope creep, maintenance and a bus factor of one
• Why judgment, and real doctors, stay essential
• Hardware consolidates upwards, software creation individualizes downwards

Originally published in German on August 27, 2026: https://klasen.ai/briefing/die-zielgruppe-bin-ich/

English version: https://klasen.ai/en/briefing/die-zielgruppe-bin-ich/

Voices are AI-generated (Jens's voice is a clone of his own). Content and analysis by Jens Klasen.

Three weeks after introducing IDUN, his personal AI fitness coach, Jens deleted a working part of it: the Yazio integration, replaced in four days by his own nutrition module. A well-tested training model turned out to be wrong, self-soldered sensors gave way to ready-made hardware, and a large share of the new code replaced things that already worked. Jens explains why that is the real news, and what enterprise IT can learn from the permission to throw things away.

In this episode

• Why tests check whether the code does what you wanted, but not whether you wanted the right thing
• How fifteen recorded sets in the gym exposed a flawed load model for rotating movements
• Why buying the more expensive, ready-made sensors was the strategic choice, and how an unchanged radio protocol made a full hardware switch a side note
• Two independent measurements per rep: wrist sensors for load, the iPad camera with seventeen 3D joint points for execution
• Three lessons for IT: cutting old habits is not failure, a proof of concept is a learning tool, and some investments only pay off later
• Building has become cheap. Judging hasn't.

Originally published in German on September 14, 2026: https://klasen.ai/briefing/was-ich-weggeworfen-habe/

English version: https://klasen.ai/en/briefing/was-ich-weggeworfen-habe/

Voices are AI-generated (Jens's voice is a clone of his own). Content and analysis by Jens Klasen.

Jens Klasen has joined Everpure (formerly Pure Storage) as a Cloud Native Architect, on the team around Portworx – and he explains why. Four open flanks at once – hardware, virtualization, container platform and cloud – mean the key question is no longer which platform is right, but how to keep the option to change your mind later. Season one, "Rewind", starts here and travels back to where RAMmageddon began.

In this episode

• Full disclosure: Jens's new role at Everpure / Portworx – and what it means for the briefings
• Four open flanks: hardware, virtualization, container platform, cloud
• Applications are portable, data is not
• Portworx as a data layer between infrastructure and Kubernetes – including VMs via KubeVirt
• Public cloud: thin provisioning instead of buying block storage in advance
• The catch: one dependency to get rid of four others

Original briefing (September 21, 2026): https://klasen.ai/briefing/vermittler-zwischen-den-welten/

Voices are AI-generated (Jens's voice is a clone of his own). Content and analysis by Jens Klasen.

Memory shortages, platform shake-ups and the IT decisions nobody wants to make. Season one, "Rewind", starts with Jens Klasen's latest briefing and travels back to RAMmageddon Part 1. Voices are AI-generated (Jens's voice is a clone of his own). Content and analysis by Jens Klasen. Original briefings: https://klasen.ai/briefing/