AI Funding Surge: River AI’s $1.1 Billion Round and the Shift Toward Infrastructure

Crunchbase News' weekly funding tally is once again an AI monoculture, with the week's largest check — $1.1 billion across seed and Series A — going to River AI, the reinforcement learning startup co-founded by an ex-xAI operator and led in the round by AMP PBC and General Catalyst. The thesis is personalized models trained on direct user data, a category still waiting for product-market fit at scale. Below the headline, Vals AI closed a $40 million Series A at a $400 million post-money valuation, with Andreessen Horowitz writing the lead check — a quieter but more strategically telling signal.

The capital stack

Three rounds anchor the week's AI money flow. River AI's combined $1.1 billion is the outlier, the kind of seed-plus-A package that resets category reference multiples and burns through runway in eighteen months at a generous burn rate. Vals AI's $40 million Series A implies a 10x markup for a company selling evaluation infrastructure rather than frontier models. Alloy Robotics collected an $8 million seed led by Square Peg Capital to build debugging agents for robot fleets.

The dispersion is the read. River is buying optionality on personalized inference — capex-heavy, distribution-dependent. Vals is monetizing the audit layer, charging rent on trust in an ecosystem drowning in benchmark theater. Alloy is plumbing the operations stack of physical AI, still pre-PMF at this check size.

Vals AI and the toll booth thesis

The Series A tells you where Andreessen Horowitz sees durable margin. Vals' benchmarks already surface in model cards from OpenAI, Anthropic, Google, Meta and xAI. The company says revenue grew eightfold versus the entirety of 2025, with the customer base doubling and headcount tripling over the past six months. Velocity matters, but the product surface is the real news: Vals Smith turns GitHub repositories into custom coding benchmarks; new frontier-risk tests span cyber, mental health and AI safety; Vals Index 2.0 broadens coverage of the broader economy.

Existing investors 8VC, Pear VC and Bloomberg Beta participated, joined by HRT Ventures and Next Ladder. The cap table signals strategic alignment with the model labs themselves — not just financial upside, but influence over how evaluation gets defined across the stack.

Sober reality check

River AI's $1.1 billion seed-and-A sets a burn-rate bar that demands either rapid commercialization or a fast bridge to a sovereign-grade exit. The personalization thesis competes head-on with the frontier labs, who already own distribution and user data. Vals has the cleaner path: evaluations are infrastructure, not a feature that gets rebuilt every model cycle. Alloy is the longest-dated wager — robot fleet debugging only compounds when fleets actually exist at scale, which they don't, yet.

Capital is still flowing into AI infrastructure, evaluation and operations. The next four quarters will sort the genuine platform plays from the demo-day darlings.