In every technology wave there is a short period when the foundations are still open to anyone willing to build them. Later, the same foundations take industrial budgets, and most players end up renting from the few who built early.
Artificial intelligence is in that early period now. This article explains what it means to own your AI foundation from the start, why the story of Bitcoin is a useful way to think about it, and what Ravira is building while the window is open.
The Bitcoin lesson: a window that closes
When Bitcoin launched in January 2009, anyone could take part in building the network. Mining ran on an ordinary home computer, and the people who joined in those first years did so with curiosity and a laptop.
The window did not stay open. Around 2010, miners moved to graphics cards. From 2013, machines built only for mining arrived, and the work moved to large, purpose-built facilities. Taking part in the foundation of the network went from something anyone could do to something that needed serious capital.
The lesson is not about price. It is about access: in a new technology, the cost of building the foundation is lowest at the beginning, and it rises quickly once the field matures. Those who understood the technology early, and built while it was still accessible, ended up with a position that later arrivals could not easily recreate.
AI today: renting a model or owning one
Most AI products today rent their intelligence. They send text to a large general model owned by someone else, pay per request, and accept whatever that model does well or badly. Renting is fast to start with, and the price, the rules and the direction of the product stay in someone else's hands.
Owning your foundation means something different. It means having your own building blocks: the way text is broken into pieces the model can read, the knowledge specific to your field, and eventually the model itself, trained for the one job you care about.
Right now, a focused team can still build these pieces. Models built for one job do not need the scale of the largest general systems, and the tools to train them are within reach. As the field matures, the advantage will sit with those who started building their own foundations early, just as it did with Bitcoin.
What Ravira is building
Ravira was built from the start with ownership in mind. Three foundations are already in place or under construction:
- Its own multilingual tokenizer. Before a model can read text, the text has to be split into pieces. Ravira has trained its own tokenizer for close to 100 languages, designed to treat smaller languages fairly and to handle text written with or without diacritics.
- A body of terms. Ravira works with 5.3 million terms across 20 domains and 108 languages. This is the knowledge that turns a fluent translation into a precise one.
- Its own translation model. Ravira is developing a model of its own, trained for translation and terminology rather than general conversation, built on the foundations above.
Together these make Ravira a company that produces its own translation technology, not a layer on top of someone else's.
Why translation, and why now
Translation is one of the clearest places where a dedicated model can stand on its own. The task is well defined, quality can be measured, and the difference between a general answer and the right term is visible in every contract, manual and medical leaflet.
It is also a field where smaller languages are often left behind. General systems are built first for the languages with the most data. A company that owns its foundations can choose to serve Gaelic, Cornish or Estonian with the same care as English or Spanish, and that choice becomes part of what it is.
The tools, the data and the knowledge needed to build dedicated models are available today to a focused team. That is what makes this the early period, and why Ravira is building its foundations now.
Building while the window is open
Bitcoin's early participants did not know exactly how the story would unfold. What they understood was simpler: the foundation was open, and building on it early mattered more than waiting for certainty.
Ravira follows the same idea for translation. Each foundation built today, from the tokenizer to the terms to the model, is one that will not have to be rented tomorrow.
We share the progress here, in this section of the blog. If you work on language technology, translation or AI and would like to talk about what we are building, write to us at humans@ravira.dev.
This article describes Ravira's direction and work in progress. It is not an offer of shares or any other investment.