The company is called River AI, it came out of stealth in June, and it has just raised 1.1 billion dollars (around 950 million euros) in a round formally classed as seed or Series A. Company age at the time of closing: two months. It is led by General Catalyst and AMP PBC, with Nvidia, AMD Ventures, Y Combinator and Singapore's state fund Temasek also joining.

Founder Igor Babuschkin is not a random name - he co-founded xAI, with earlier stops at DeepMind and OpenAI. That is the part of the story that opens doors. A billion dollars is not going towards a product, because there is barely a product - it is going towards a CV and a thesis.

The thesis, unlike many others, is at least specific. Babuschkin argues the whole stack has to be rebuilt from scratch: "the training, the models, the product layer and new hardware that lets personal artificial intelligence live close to you". Instead of agents replacing workers, he wants agents each user trains themselves. In his launch post he describes them as "guardian angels: quietly present, on your side" - and adds that they "will be yours, not somebody else's".

Commercially, River is already selling access through a developer interface, charged per million tokens, with the price depending on the chosen open model. The offer lets developers fine-tune open models instead of trying to coax somebody else's. The company promised that "any enterprise can complete complex training in 15 to 20 minutes without its own infrastructure team, with savings of two to four times compared to closed alternatives".

Those numbers come from the company's own announcement, not from independent verification - an important distinction when the price of the round is being defended with a performance promise. It is also worth noting who is putting up the money: Nvidia and AMD are investing in a company whose product requires their chips. That is not fraud, it is ordinary industrial logic - but it is also not a neutral signal of market confidence.

There is a real basis for the thesis. Companies are slowly realising they do not want to depend on a single closed model, and are moving towards a mix that includes open weights. Knowing how to fine-tune such a model after training is the bottleneck - and that is exactly the gap River is trying to enter.

The old question remains, the kind we know in the Balkans from another sphere entirely: what does it mean when capital arrives before the product? In River's case, the answer has not been written yet. What has been written is the amount - one billion one hundred million dollars in the bank, and two months of track record to justify it.