Prime Intellect has raised $130 million in a Series A round at a $1 billion valuation, led by Radical Ventures, with participation from Nvidia Ventures, Intel Capital, Dell Technologies Capital, and Iconiq TechCrunch. The round also drew a notable bench of angel investors: Aravind Srinivas of Perplexity, Aaron Levie of Box, Winston Weinberg of Harvey, Jeff Wang of Cognition, and Brendan Foody of Mercor.

Founded in 2024, Prime Intellect sells compute access, a reinforcement learning framework, and evaluation tooling as a bundled platform for enterprises building their own AI agents rather than relying solely on off-the-shelf frontier models. The company has reached an annualized revenue run rate of $100 million, according to TechCrunch's reporting, and counts Ramp, Zapier, and Flapping Airplanes among its customers.

The jump in valuation is steep by any measure. Prime Intellect's prior disclosed raise was $15 million, announced on its own blog in February 2025, which took cumulative funding past $20 million Prime Intellect. Going from just over $20 million in total funding to a $1 billion valuation in roughly a year and a half places Prime Intellect in the small cohort of infrastructure-layer AI startups whose valuations have scaled ahead of, or at least in step with, disclosed revenue.

Vincent Weisser, Prime Intellect's co-founder and CEO, described the company's pitch to TechCrunch as giving enterprises the tooling to train and fine-tune agents on their own workflows rather than depending entirely on general-purpose models from OpenAI, Anthropic, or Google. That framing puts Prime Intellect in direct proximity to the broader "build versus buy" tension that has defined enterprise AI procurement over the past two years: companies weighing the cost and control benefits of custom, task-specific agents against the convenience and steadily improving capability of frontier APIs.

The Ramp case cited in the funding announcement is the clearest illustration of that value proposition. Ramp, the corporate card and spend-management company co-led by Karim Atiyeh, used Prime Intellect's platform to build an agent that finds answers inside spreadsheets. TechCrunch reports that the resulting agent beat frontier models on accuracy while running faster and at lower cost — a combination that, if it generalizes beyond this one use case, speaks to a genuine gap reinforcement-learning-tuned, narrow agents can exploit against general-purpose models optimized for breadth rather than a single task.

David Katz, the Radical Ventures partner who led the deal, is quoted in the TechCrunch piece, though the specifics of his rationale for the round size and valuation were not detailed beyond the firm's participation as lead investor.

The investor list is worth reading as a signal in its own right. Nvidia Ventures and Intel Capital sitting alongside Dell Technologies Capital in the same round suggests infrastructure vendors see Prime Intellet's compute-and-tooling bundle as complementary to, rather than competitive with, their own hardware and cloud offerings. The angel roster — Srinivas, Levie, Weinberg, Wang, Foody — reads like a cross-section of founders who have each built either an AI-native product (Perplexity, Harvey, Cognition) or a company now under pressure to adopt agentic workflows at scale (Box, Mercor). That mix of enterprise-software veterans and AI-lab founders backing the same seed-stage bet is itself a data point on where informed capital thinks the agent-tooling market is headed.

The broader significance of this round lies less in the dollar figure than in what it says about where enterprises are choosing to spend their AI budgets. Through 2024 and into 2025, much of enterprise AI spending concentrated on API consumption of frontier models — pay-per-token access to GPT-class or Gemini-class systems with minimal customization. Prime Intellect's traction, if the $100 million run-rate figure holds up under scrutiny, suggests a meaningful slice of that spending is migrating toward companies that want to own the training loop: their own reinforcement learning pipelines, their own evaluation harnesses, their own fine-tuned checkpoints running on rented or dedicated compute.

Whether that shift is durable or a product of this particular funding cycle's enthusiasm for anything tagged "agentic" is a fair question, and one TechCrunch's reporting does not fully resolve. Revenue run rate is a snapshot, not a trend line, and a company barely two years old scaling to a ten-figure valuation invites the kind of scrutiny that has, in past cycles, preceded both durable platform businesses and shorter-lived enthusiasm plays. In this author's view, the technical bet underlying Prime Intellect — that narrow, RL-tuned agents can outperform frontier generalists on cost and accuracy for well-defined enterprise tasks — is sound and consistent with what practitioners have observed elsewhere in applied ML for years. Whether Prime Intellect specifically is the vehicle that captures that opportunity at scale is a separate matter that this round's investors are, in effect, betting several hundred million dollars will prove true.

What happens next will likely turn on whether more customers can replicate the Ramp result across different task domains, and whether the compute economics hold as usage scales beyond early adopters into broader enterprise rollout.