River AI Raised $1.1B From Anthropic's Own Investors

Igor Babuschkin wants to end the era of renting intelligence from a handful of labs. The investors writing his checks are also some of the largest shareholders in one of those labs.

River AI, the startup Babuschkin launched two months ago after co-founding xAI, announced on August 11 that it raised $1.1 billion across seed and Series A rounds. General Catalyst and AMP PBC led the deal, with NVIDIA, AMD Ventures, Y Combinator and Temasek also writing checks. Three of those names, General Catalyst, AMP PBC and Temasek, are also listed among the investors who put $30 billion into Anthropic's Series G at a $380 billion valuation in February. NVIDIA is there too, through what Anthropic's own announcement calls "a portion of previously announced investments."

TL;DR

River AI raised $1.1 billion to build tools that let users own and fine-tune their own AI models instead of renting access from labs like Anthropic
General Catalyst, AMP PBC, Temasek and NVIDIA backed River while also holding stakes in Anthropic's $380 billion round
Babuschkin's pitch is explicitly anti-lab; his cap table is diversified across the exact model it argues against
Enterprise customers get fine-tuned checkpoints they can export, but still pay River per token to run them

Babuschkin's framing has been consistent since he left xAI on August 13, 2025 to launch a venture firm before pivoting to building River himself. On River's own site, the pitch reads: "Prompting steers a model you don't own and can't improve. River lets you train open models into ones that are truly yours, and serve them like any other endpoint." The company's first product, the River API, lets customers apply LoRA fine-tuning and reinforcement learning to open-weight models ranging from 35 billion to 1 trillion parameters, including Kimi K2.6 and Qwen 3.6-35B-A3B. River claims complex RL runs finish in 15 to 20 minutes at two to four times the cost of closed alternatives, with pricing starting at $1.00 per million tokens for the smaller Qwen model and climbing to $12.84 per million for Kimi K2.6 at its 262K context length.

River AI's homepage, showing the company's blue-gradient landing page River AI's homepage frames the company around ownership: "Intelligence that flows with you." Source: river.ai

Who Actually Wants What

The round looks less like a coalition behind one idea and more like four separate bets that happen to share a wire transfer.

Stakeholder	Stated position	What they actually need
River AI / Babuschkin	AI should be owned by users, not rented from labs	A wedge against Anthropic, OpenAI and xAI to justify a reported $5 billion valuation
NVIDIA / AMD Ventures	Backing the next generation of AI infrastructure	GPU-hours consumed, regardless of whether a lab or a customer is the one training
General Catalyst / AMP PBC / Temasek	Diversified exposure across the AI stack	Returns on capital, reached by owning positions in River and its stated rivals at once
Enterprise customers	Control over their own models	Cheaper inference than closed APIs, even if it means a new intermediary

A GPU installed inside a desktop tower, cables routed to its power connector Whether AI training happens inside a closed lab or on a customer-owned checkpoint, the hardware bill lands on the same chipmakers. Source: unsplash.com

Who Benefits

NVIDIA and AMD Ventures win regardless of which company in this story succeeds. Every RL run and LoRA job on River's platform still burns GPU cycles, and River's entire cost pitch depends on use, not on displacing hardware demand. General Catalyst, AMP PBC and Temasek benefit from a different mechanism: they get exposure to a fast-growing infrastructure layer without picking a side in the fight between open and closed models, since their Anthropic stakes keep appreciating whether River takes share from it.

Babuschkin benefits from the framing itself. A former xAI co-founder positioning his company as the anti-lab alternative is a stronger pitch to enterprise buyers than "another API reseller," and it likely helped pull in $1.1 billion for a company with no public product history before June.

Who Pays

Enterprise customers pay River per token, the same way they'd pay Anthropic or OpenAI, just for a smaller model they can theoretically export. Whether that counts as ownership depends on what happens if they stop paying. A customer-owned checkpoint is portable in principle; a production deployment serving live traffic through River's endpoints isn't something most teams can migrate overnight. The GPU vendors don't pay anything in this trade. If open-weight fine-tuning grows, they sell more chips. If closed labs stay dominant, they already have several billion dollars in Anthropic and OpenAI exposure to show for it.

The Contradiction Nobody Is Naming

River's marketing draws a hard line between "a model you don't own" and "one that's truly yours." That line gets harder to hold when a third of the round's capital sits on both sides of it. General Catalyst, AMP PBC and Temasek didn't just passively hold Anthropic stock picked up on a secondary market. Anthropic's own release names them as Series G investors, in a round co-led by Abu Dhabi's MGX, which has since put $49 billion behind AI infrastructure bets of its own and watched Anthropic's valuation nearly triple in the Series H that followed three months later. These firms aren't choosing River over Anthropic. They're betting on the category winning no matter which specific company inside it does.

None of River's investors have said the two positions are in tension. NVIDIA, AMD Ventures, General Catalyst and AMP PBC didn't offer additional comment beyond confirming their participation in the round through the funding announcement.

"River is the new standard in personal intelligence."

That's the language on River's own homepage. It's a claim about who owns the model. It says nothing about who owns the company training it, or who else those same backers are betting on to win instead.

What Happens Next

River's actual product, a training service comparable to what fine-tuning platforms like Together AI and Fireworks already offer, still has to prove it's cheaper and more reliable at scale than the labs it's positioned against. A $5 billion reported valuation for a two-month-old company with strategic backing from both major GPU vendors sets a high bar for a company that hasn't published independent benchmarks of its RL claims. The investors who wrote the check aren't betting against Anthropic. They're making sure that whichever way the market moves, they were already there.

Sources: