Nous Research raises $50 million for distributed AI training led by paradigm

Decentralized artificial intelligence startup Nous Research said today it raised $50 million in an early stage round, almost fully funded by crypto-focused investment company Paradigm.

According to a Fortune report, the Series A round follows a total of $20 million seed rounds from investors, including Decentralized Global, North Island Ventures and Delphi Digital. The funding will bring the company’s crypto token valuation to $1 billion, according to the report.

Novus uses the Solana Blockchain network to enable distributed training of AI models by allowing users to commit idle calculations from both consumer and industrial graphics processing units.

The company is trying to disrupt the current paradigm of centralized AI development for giants such as Deepseek and Openai by allowing large numbers of participants to join the network and commit resources.

Nous didn’t start with the idea of ​​adopting blockchain technology, but ultimately sought it because it could be built into promoting distributed model training. It can also be used to reward participants in unused GPU computing who contributed to the system with blockchain-based cryptocurrency.

“We’re thinking about the incentive mechanism behind Crypto, and we’re urging people to actually use their idol computing as a donation, but as a deal,” Nous co-founder Karan Malhotra told Fortune.

Additionally, blockchain technology includes mechanisms for tracking transactions that allow companies to build a distributed training process and create a complete, transparent historical audit trajectory of all data committed to the system.

Nous hopes that this ability will allow you to detect and avoid data addiction, also known as AI addiction. This is a type of tiebar attack in which malicious actors manipulate or corrupt training data used to build machine learning models. The goal is to influence the model’s behavior and output outcomes, leading to biased or harmful outcomes by injecting false or misleading data, deleting data, or changing labels.

Malhotra added that there is currently a lot of skepticism about blockchain technology and cryptocurrency, particularly regarding fraud and hacking, pointing out that there is “a lot of room for glyft.” However, he said that many research teams are well-versed in the crypto business and initially did not support adding blockchain elements until the recent research breakthrough in decentralized training.

“We don’t want to be kind of upset by the traditional views of how cryptography operates when we are very serious labs and academic labs,” Malhotra said.

Nous was recently named after the Goddess of the Soul in Greek mythology, a network of Solana blockchains that allows for globally distributed training of fault resistance in AI models. The company said it successfully tested the 15 billion parameter model by expanding its network at 11,000 levels in December 2024.

At the heart of Psyche is a training optimization component called Nous Distro, which trains each node training independently rather than adjusting it at every step. The company said this would significantly reduce the bandwidth required to achieve results comparable to intensive training.

Image: Siliconangle/Microsoft Designer