The independence being decided today is not on a podium. It is in the key inventory.
The question is simple to formulate and uncomfortable to answer: how many of your critical systems depend on a key another company can turn off? The closed AI model that writes your product’s content. The SaaS holding your CRM. The vendor processing your payments. Each of those keys sits in another company’s hand, and your business continuity depends on their commercial policy.
The risk layer nobody inventories
Data sovereignty became standard conversation: where data rests, under which jurisdiction. That layer already has checklists, certifications, and a market.
The layer above still has no owner: operational dependency. A closed model discontinued, repriced overnight, or restricted by an export decision kills a running product, with the data perfectly hosted and the company perfectly compliant. The risk is not the data; it is the brain.
The open alternative stopped being fiction
What changed in recent months is the offer on the other side of the table. Z.AI published the weights of GLM-5.3, a 744-billion-total-parameter (40 billion active) flagship, on Hugging Face, with the line’s Flash under an MIT license. UMD and Meta FAIR’s Huginn-3.5B released weights for a recurrent model. The open frontier stopped being a research project and became a catalog.
At Tech86, the thesis is tested in production: open GLM-line models running on our own cluster, with a managed plan from Z.AI as the middle path for those who want the alternative without becoming a data center operation. We use and recommend both paths, because the right one depends on each workload’s volume and criticality.
The inventory that turns a lecture into a plan
The exercise fits in one afternoon:
- List the five keys whose loss kills the business, and who owns each one.
- Classify the alternatives: open, managed, or nonexistent. "Nonexistent" is the inventory’s red line.
- Test a small migration: one non-critical workload on the open model, with quality, cost, and operations measured.
- Formalize plan B per key: what happens the day it changes price or owner.
The result is the company’s real technical independence map. It is almost always worse than the board imagines and cheaper to fix than the fear suggests.
Conclusion
Technical independence is not nationalism or philosophy: it is operational continuity with method. The difference between those who have it and those who do not is one afternoon of inventory and a well-run pilot.
Open weights changed the board: the alternative to the closed model exists, is frontier-grade, and is in the catalog. The question is no longer "can we be independent?" and became "why is your critical key still someone else’s?".