Technical writing
The data warehouse in your own data center
Articles on cost models, data sovereignty, table formats and connecting AI agents. Technical, sourced, without the sales pitch.
What a second server brings under peak load
Our first load test left one question open: what does a second machine bring? Now we have measured it. At 20 concurrent users the platform completes 77 per cent more queries with a second server, and the waiting time halves.
by Daniel Alisch
What a terabyte actually costs
Cloud warehouses bill storage per terabyte per month; your own storage is bought once. The comparison with numbers — colocation, redundancy, and what the media price leaves out.
Daniel Alisch · · 4 min
A stress test where nothing breaks was not a test
What a load test for a data warehouse has to show: not how fast it is, but where the limit sits, what happens at the limit, and how long recovery takes.
Daniel Alisch · · 4 min
What a semantic layer actually contains
Three tables and one rule. What a metric, a dimension and a stored join hold in practice — and why the rule matters more than the data model.
Daniel Alisch · · 4 min
Article 10 of the AI Act asks where your data came from — and the deadline has moved
Obligations for high-risk systems now apply from December 2027. The requirement itself has not changed, and the most demanding part of it is a question for the data warehouse, not the model.
Nils Gregersen · · 3 min
Leaving the cloud: what the EU Data Act changes from January 2027
From 12 January 2027 switching charges are banned, the transition may take no more than 30 days — and the regulation names the move to your own data centre explicitly.
Nils Gregersen · · 3 min
Can you run Snowflake on-premises?
No — in no edition. The more interesting part is what exists instead: dedicated instances, customer-held keys, and an offering called “Bring Your Own Cloud” that does not mean what it sounds like.
Daniel Alisch · · 3 min
Why cloud warehouses get expensive as soon as they succeed
Consumption billing rewards restraint and punishes use. What the figures from Flexera, a16z and 37signals support — and what they do not.
Daniel Alisch · · 3 min
MCP in the data warehouse: agents need a semantic layer, not SQL
Why an MCP server that only offers run_sql is the wrong interface to a warehouse — and how a semantic layer brings permissions, auditing and correct figures together.
Nils Gregersen · · 4 min
Calculating TCO honestly: what an on-premises warehouse really costs
Hardware is the smaller part. What TCO comparisons for on-premises regularly leave out — staff, power, spare capacity — and how to put numbers on it.
Daniel Alisch · · 3 min
Apache Iceberg on your own hardware: what the table format actually solves
Iceberg is often mistaken for a file format. What it really provides — ACID, time travel, schema evolution — and where the catalog becomes the real dependency.
Nils Gregersen · · 3 min
Separating storage and compute without the cloud: what it takes
Separating storage from compute is not a cloud invention. What you need on your own hardware — and where elasticity really ends.
Nils Gregersen · · 3 min
Data sovereignty is not a contract problem, it is a location problem
Why a data center in Frankfurt does not prevent access by US authorities — Article 48 GDPR, the CLOUD Act, and what the Bitkom Cloud Report 2026 shows.
Daniel Alisch · · 3 min
Warehouse migration: what actually takes the time
Moving tables is easy. What costs time is roles, stored logic, time zones and reconciling against the old figures. A realistic look at four weeks.
Nils Gregersen · · 3 min
Why data projects in regulated industries fail at sign-off
It is not technology that keeps modern warehouses out of hospitals, banks and public bodies, but the burden of proof. What auditors actually want to see.
Daniel Alisch · · 3 min
Text-to-SQL does not fail on the model, it fails on the metrics
Why better language models barely improve text-to-SQL accuracy — and which groundwork in the warehouse actually makes the difference.
Nils Gregersen · · 3 min
FinOps is a symptom, not a solution
That companies need a dedicated discipline to manage cloud spend is remarkable. What FinOps achieves — and what it reveals about the billing model.
Daniel Alisch · · 3 min