Big company rents you AI by the token. River AI wants you to own it outright — and that's Babuschkin's first move after leaving xAI.
After departing xAI, he wants everyone to "have their own intelligence."
On June 10, Igor Babuschkin officially unveiled his new company: River AI.
It's one of the highest-profile startups to emerge from the wave of xAI executive departures since February.
Former xAI co-founder, River AI CEO Igor Babuschkin
Babuschkin's resume reads like a who's-who of the past decade's top AI labs: DeepMind, OpenAI, xAI — he even named the AlphaCode project himself.
In a post on River AI's official blog, he laid out the motivation behind founding the company.
https://river.ai/introducing-river-ai.html
Top labs aim to automate all economically valuable human work, Babuschkin said. That made him "a little worried" — it sounded like humans were about to be replaced.
River AI was founded to answer that very question. His bet: AI isn't about "replacing" humans, but "augmenting" them.
It sounds like an empty slogan, but Babuschkin translated it into a more concrete proposition: giving everyone the ability to own AI that truly belongs to them.
AI like a twin brother
"An existence that always understands you"
Among River AI's co-founders are twin brother engineers: Ievgen Soboliev and Dima Soboliev.
River AI co-founders, twin engineers Ievgen Soboliev (left) and Dima Soboliev (right), both former xAI engineers.
Both previously worked as engineers at xAI. Growing up as twins means always having someone beside you — someone who accompanies you, grows with you, and has your best interests at heart.
What if AI could do the same?
That's what River AI is building. Not an assistant you summon when needed, but an "always-present, always-on-your-side" presence — what River AI calls "guardian angels."
But this warm vision is backed by a concrete product strategy centered on "AI that belongs to you."
What does today's AI look like?
A handful of companies hold intelligence and rent it to you by the token. You pay, you call the API, but the model never belongs to you.
River AI argues that rented AI is intelligence you cannot keep, evolve, or truly own.
Its mission, therefore, is to build "personal AI owned by everyone and shaped for everyone." You own the hardware it runs on, the data it's trained on, and the intelligence itself.
Babuschkin frames this as owning and controlling "the means of production of intelligence" — making everyone an "owner" of AI, not a tenant.
The dream starts with v0.1
River's first public product is a model training tool, released as a v0.1 preview.
Their slogan:
Stop wrestling with prompts. Start training your own model.
River API official comparison. Left: "Prompts" can only call general-purpose models with fixed weights. Right: "Train with River" — weights actually update, and the checkpoint is yours.
It helps you "tune" open-source models into what you need, using LoRA fine-tuning and reinforcement learning, covering a range of open-source models from 35B to 1T parameters.
The trained model is yours — you can save it, improve it, deploy it, and integrate it into your own products.
In the official examples, there's a file called rl_loop.py — a minimal training loop.
River API training example. Training is a minimal Python loop: "sample, score, update." One RL training run in the diagram uses about 500 million generated tokens, costing under $1,000 total.
It's clear that River is targeting developers first, not everyday users. It translates "owning the weights" into a product. But Babuschkin admits the API is only the first step — true personalization and continuous learning will come next to turn general-purpose models into something that "really knows you."
An all-star team pulled from xAI
River AI's core team is practically a direct transplant from xAI's inner circle, including former Tesla employees.
According to Bloomberg, River AI's founding team includes Vincent Stark, who was at xAI until December 2025 heading product safety; Lily Lim, xAI's former general counsel; and the Soboliev twins, both former xAI engineers and now co-founders at River overseeing engineering.
After xAI merged with SpaceX, founding team members began to depart one after another. Babuschkin himself left in August 2025.
River AI is a brand-new outfit, officially incorporated only on April 20 this year. Its LinkedIn shows just 11 to 50 employees.
Aligned to you alone
Who's responsible for safety?
River AI's boldest — and most controversial — move is in alignment.
Today's top AI labs align a massive corporate model to all of humanity. River AI takes the opposite approach.
In Babuschkin's own words: Instead of aligning one big corporate model to all of humanity, align your AI directly to you.
In his view, an AI that deeply understands you and acts in your interest is, by definition, highly aligned.
But this raises a question: If AI is aligned only to an individual, who guarantees group safety and societal risk?
River AI has privatized alignment — that's its strength, and the first challenge it will face.
According to Forbes, Babuschkin is in talks to raise up to $1 billion for River at a valuation of up to $5 billion, with Babuschkin personally planning to invest up to $100 million.
The funding round is still ongoing and figures may change; both parties declined to comment.
Is "Personalized AI" the next battleground?
River AI is still at a very early stage, but it doesn't plan to build just another app. It aims to rebuild the entire AI stack from the ground up: local hardware, cloud infrastructure, models and algorithms, personalization and continuous learning, all the way to the end product.
Babuschkin envisions AI eventually "living somewhere very close to you" — a private AI, trained on your own data, that doesn't give you away. In his words:
Intelligence should travel with you.
River AI isn't alone in betting on this path.
Like Babuschkin, engineers who left after the xAI-SpaceX merger are launching a new wave of startups.
Humans& co-founders, from left: Noah Goodman, Andi Peng, Georges Harik, Yuchen He, Eric Zelikman. Yuchen He and Eric Zelikman are from xAI.
Former xAI engineers Eric Zelikman and Yuchen He co-founded AI lab Humans&.
They share Babuschkin's bet on personalized AI: intelligence shouldn't be rented out by a handful of big tech companies — it should belong to everyone.
Big tech is also going all-in on personalization
In fact, top AI labs have long been working on making AI "know you better."
OpenAI launched ChatGPT Memory in 2024, allowing explicitly stated information to carry across conversations. The real turning point came in April 2025: ChatGPT began referencing users' entire chat history to automatically infer preferences, projects, and long-term context.
OpenAI later called this an early form of what would become Dreaming.
On June 4 this year, OpenAI released Dreaming V3, further consolidating users' profession, interests, travel plans, writing preferences, and more into viewable, correctable "memory summaries" — making ChatGPT feel more like an assistant with long-term context.
Anthropic started a bit later.
In August 2025, Claude first gained the ability to search past chats when prompted. In September, Claude Memory launched, starting to remember users' and teams' projects, preferences, and working patterns.
On the surface, the big labs and River are trying to do the same thing: build an AI that knows you better.
The difference? OpenAI and Anthropic still rent you models. No matter how personalized they get, they never truly belong to you.
River wants more than that: personalization shouldn't stop at "remembering you." Both its brain and body should be yours.