Nous Research in Talks to Raise $75M at $1.5B Valuation as Decentralized AI Push Gains Backing
The crypto-native AI lab behind the Hermes open-source agent is seeking new capital led by Robot Ventures, with a valuation that has climbed 50% since April 2025.
Nous Research, an AI lab that builds open-source models and decentralized training infrastructure on Solana, is in active negotiations to raise at least $75 million in a Series B round at a $1.5 billion valuation, according to a report published by TechCrunch on July 13, 2026. Robot Ventures is leading the round, with Union Square Ventures among the co-investors. The deal has not closed and terms remain subject to change.
The fundraise follows a $50 million Series A closed in April 2025 at a $1 billion token valuation, led by Paradigm. If the current terms hold, the valuation would have grown 50% in roughly 15 months, reflecting accelerating investor appetite for projects that combine open-source AI infrastructure with blockchain coordination.
The company has raised approximately $70 million in total across seed and Series A capital before this round, with early backers including Distributed Global, North Island Ventures, and Delphi Digital.
Robot Ventures, the lead investor in the Series B talks, is a crypto-focused early-stage fund co-led by Robert Leshner, who founded the DeFi lending protocol Compound, and Tarun Chitra, who founded the on-chain risk modeling firm Gauntlet Network. The fund is currently deploying its fourth vehicle, a $75 million fund raised in August 2024, and has backed more than 230 projects including Polymarket, Monad, LayerZero, Jito, and Lido. Its most recent prior investment before this deal was Squads, a multisig infrastructure project built on Solana. The Solana connection is notable: Nous Research's Psyche network uses Solana smart contracts to coordinate its decentralized training infrastructure, which aligns with Robot Ventures' prior investments in Solana-native projects such as Jito and Squads.
The technical centerpiece of Nous Research is Psyche, a distributed AI model training network that allows anyone with a GPU to contribute compute toward training large language models. Rather than relying on centralized data centers, Psyche pools resources from participants worldwide and coordinates them via Solana. To make this viable over ordinary broadband connections rather than high-capacity data center fiber, the system uses an algorithm called DisTrO (Decoupled Momentum Optimisation), which compresses gradient data by transmitting only the sign of frequency-transformed momentum components, achieving more than 3x compression. The network's peer-to-peer layer is built on the Iroh P2P network using QUIC-based direct connections, with nodes identified by Ed25519 public keys rather than IP addresses.
The first Psyche training run, called Consilience, targets a 40 billion parameter model trained over 20 trillion tokens and is described by the team as the largest pretraining run conducted over the open internet to date.
Paradigm partner Arjun Balaji called this approach "a powerful contrast to the closed, centralized efforts from incumbent labs" when the Series A was announced.
Karan Malhotra's public statements consistently frame the company's use of blockchain in practical rather than ideological terms. "We think of the incentive mechanism behind crypto to push people to actually utilize their idle compute," the Nous Research co-founder said in April 2025.
Industry estimates suggest more than 40% of global GPU capacity sits idle at any given time, a figure that forms the core market assumption underpinning Psyche's design.
For context, the broader decentralized compute sector is growing: Akash Network recorded $5 million in compute spend in Q1 2026, and io.net now aggregates more than 300,000 GPUs across more than 55 countries, with monthly active GPU providers growing roughly fivefold between Q1 2025 and Q1 2026.
For developers in India and Nigeria, among the largest non-US Web3 developer populations in the world, Nous Research's architecture carries practical significance.
India now counts approximately 119 million active crypto users and accounts for an estimated 20 to 30 percent of global Web3 developers. Nigeria leads Africa with 4% of all global Web3 developers, a figure that grew 36% year-over-year in 2025, and its Web3 startups raised $43 million in 2025, more than double the prior year.
Both communities face structural cost barriers when accessing frontier AI tools priced in US dollars. Hermes Agent, the company's open-source autonomous agent stack launched in February 2026, is available in 3B, 8B, 70B, and 405B parameter variants with fully open weights on Hugging Face. The product also integrates MiniMax AI's M2.7 model, expanding its third-party model support. Hosted tiers are priced between $20 and $200 per month, and all models support local deployment via Ollama, llama.cpp, and LM Studio. The Hermes Agent repository has accumulated more than 32,000 GitHub stars since launch.
Psyche's broadband-compatible design also means developers in cities with improving fiber or fixed wireless coverage could potentially contribute compute and earn participation rewards, though the precise form of those rewards has not been publicly confirmed. Outside major urban centers, however, practical barriers remain significant: consumer GPU penetration is low in many markets, and rural broadband quality is inconsistent across much of Africa and other developing regions.
One critical caveat for retail investors in both regions: Nous Research has not issued an official NOUS token. Unofficial tokens using the Nous name have appeared on Solana-based decentralized exchanges and are not endorsed by the company. The reward mechanism for Psyche contributors, whether a proprietary token or native SOL, has not been publicly decided. If the Series B closes at the reported terms, the headline valuation will likely generate renewed speculative activity around unofficial tokens, and readers should treat any such offerings with significant caution.
Looking ahead, the round's outcome will serve as a signal for how much institutional capital is prepared to commit to decentralized AI infrastructure at scale, as opposed to closed commercial AI systems. Nous Research has spent four years building credibility through open model releases before making its blockchain infrastructure the headline product. Whether Psyche's first major training run can demonstrate the technical case will determine whether the current valuation reflects durable fundamentals or early-cycle momentum.