Smart Money: The Role of AI in Shaping the Cryptocurrency Landscape

An investment thesis for $NMT, written for Oregon Blockchain Group’s Venture Capital style investment fund. This is written by students for educational purposes, not financial advice.

Bryson Ballew

Published in

Oregon Blockchain Group

13 min readApr 19, 2024

Analysts: Jordan Brewer & Bryson Ballew

General Background on Protocol

NetMind Power ($NMT), developed by NetMind.AI is based in London, UK, and Washington DC, USA. It is a decentralized platform that focuses on AI training and machine learning by providing global large-scale computing through its own Layer 1 blockchain. By utilizing idle GPUs of users worldwide, people can rent out their computing power remotely to get paid in NMT (NetMind Token), the native token on the blockchain. By leveraging this audience they can capture an untapped area of computing power that works as an alternative to huge data centers that are both expensive to create and to operate. Democratizing access to computing power for businesses and research institutions offers an easier and more affordable method for running AI models. Their blockchain uses a PoA (Proof of Authority) consensus mechanism which is similar to PoS (Proof of Stake), which can be seen with Ethereum 2.0. Although NetMind Power operates on its blockchain, one of its greatest strengths is its interoperability. Compatible with a wide range of AI models and frameworks allowing for users the flexibility to use what they most prefer. By offering such a degree of flexibility users can reap a plethora of rewards across the board making it an extremely attractive and unique protocol with more growth to come soon.

Macro Factors Impacting Protocol

AI is top of mind for investors and builders alike. With the promise of improved productivity and efficiency, companies have taken notice with 179 companies in the S&P 500 using the term “AI” in their Q4 2023 earnings call. This is well above the 5-year average of 73. This is emphasized by Bloomberg’s expectation that the generative AI industry will grow at 42% per year through 2032. In addition to the increased popularity of the idea of AI, AI companies themselves have seen dramatic inflows of capital. Sam Altman was recently seeking $7 trillion and Nvidia has seen its market capitalization increase from $687.5 billion to $2.2 trillion in the last year, adding $1.5 trillion in value. There is a demand to invest in AI and AI adjacent stocks, however, there are not many companies that are directly building AI that people can invest in. AI-related crypto projects serve as ways for individuals to directly get exposure to AI that they would not otherwise have. Furthermore, blockchain infrastructure creates an open-source environment that siloed models will struggle to emulate. The need for computational resources and an environment to train and run an AI model is a need that is not being met by legacy infrastructure. Free and open development of AI will allow it to grow and improve society beyond what only a few entities could achieve. Freedom to choose the model, data, and training method is important to open the possibilities of AI, which cannot be done in a reliable and trustless way in the current centralized architecture.

Who’s the Team Building the Protocol

Kai Zou is the Founder and CEO of NetMind.AI. A notable career within academia, Kai received his Bachelor’s degree in Mathematics & Physics from Tsinghua University as well as a Master’s in Mathematics & Statistics from Georgetown University. Having authored 11 academic papers, he is known for his entrepreneurial background and extensive experience within the artificial intelligence space. Alongside him is Huanzhou Huang the CTO with previous experience at both Coinbase, Microsoft, and Microstrategy. As COO, Yuan He brings his 20 years of experience to the table including industries like IT and fintech as well as real estate. Overall, the entire team is doxxed and has a proven track record. As Marketing Lead, Joshua Chen holds a master’s degree in Marketing from Northwestern University as well as has prior experience as a director at a major US marketing agency in addition to working in global marketing and sales for a Chinese tech unicorn. As Product Lead, Mengni Shen holds a Ph.D. in Medical Science from the University of Oxford while also hosting the #2 Emerging Podcast in the Chinese wellness/ health optimization sphere. As Product Lead, Hongyu Ren holds a bachelor’s degree in Mathematics and Physics from Tsinghua University as a PhD in Physics from Tsinghua University. Currently, there are 2 research advisors (Zheng Yuan & Yu Chen), both with impressive academic and industry experience.

General Auditing Background for Protocol

Currently no audits.

Specific on What Protocol Does

NetMind Power have created what they call The Volunteer Computing Network which remains at the core of their platform. By acting as a utilitarian host platform that enables people to join as “volunteers” by contributing their GPU power to the platform, a sense of community can be built where both the buyer and the seller can be satisfied. On the seller side (volunteers who contribute their GPU to the platform), providers can receive passive income as a result of sharing their computing power. On the buyer side, not only can individuals/organizations outsource varying levels of computer power but also leverage an untapped market which results in it being the cheaper alternative when compared the the status quo of large data centers.

Additionally, their Training Platform utilizes a dynamic decentralized architecture making it both scalable and secure for all users. Through the use of data and model parallelism, proper resource allocation can maximize efficiency based on the demand of computing power needed for that particular model. To help assist with this process, data partitioning, and model aggregation are employed to enhance ease of deployment. Once completed, the platform aggregates the results from each device to combine them into its final form. With advanced encrypted and secure MPC (multi-party computation) and differential privacy techniques, users can rest assured that their data is protected.

Operation on the NetMind Chain will be facilitated via the use of the native utility token, NMT (NetMind Token). Pricing mechanisms are decentralized, allowing for varying providers to set and set their own rates. Open-source models deployed on the blockchain benefit from the community by allowing users to find pre-trained AI models to adjust their parameters and train with specific datasets. Additionally, onchain users have the option of staking their NMT against Mind Nodes which allows them to earn additional yield. The 21 nodes with the highest amount staked become Master Nodes and will receive 20% additional total daily staking rewards which are split equally amongst all of the master nodes.

The Inference Platform is designed to work in tandem with the training platform, allowing for increased utility on the platform by letting users deploy and run inference on their own models, models created by others, and open-source models as well. Through containerization mechanisms and the use of APIs, users can deploy their trained AI models on the inference platform thus making them accessible to other users on the platform as well.

Holders of NMT are also able to gain exclusive access to all of their newer products moving forward. The future is looking bright with future upgrades posing to include the likes of a Google collab integration, no code solution for fine-tuning models, and the ability for users to deploy and inference their model whilst in production. Furthermore, offering external services within the NetMind ecosystem creates a more cohesive and easy-to-use user interface. Services such as Avagi.ai which was launched in early 2023, as well as NetMind Life and NetMind Chat. The NetMind Chat service allows for a multitude of ways for people to find value within the ecosystem. One example is enabling the forking and fine-tuning of the open-source model through the NetMind platform. By offering an AI chat model like such, it is also possible to integrate models into dApps with seamless deployment. NetMind claims their distinctive difference lies within the GAIA which is based on the Generative Pre-trained Transformer family. Avagi.ai offers a dual chat/video AI chatbot service with a virtual face that speaks back to you with prompts using lip-syncing technology which is surprisingly accurate. NetMind Life on the other hand aims to help predict overall wellness by prompting you to answer a series of questions and developing a digital analysis of your health level. These are just a handful of the services that the NetMind.ai team is building.

Why the Protocol Offering Matters to Consumers

The growing demand for AI has led to growing demand for high-performance computing. NetMind offers a solution through its access to a decentralized computing network. While a decentralized computing network on its own is not a valuable enough offering in a saturated market, its partnerships with universities to drive forward AI research and innovation are incredibly valuable. NetMind’s optimized infrastructure allows for easier model creation and use. Its training platform uses data parallelism and model parallelism to enhance training efficiency, and its inference platform allows anyone to use a variety of models via API and also supports containerization for easy model deployment on the NetMind Chain. Freedom to use different models is important for users, the same way it is important to have the ability to use both GPT 3.5 and Gemini. By ensuring that models are trained safely via secure enclaves and MPC, developers can be sure their model is being trained how they want with the data they want, encouraging a wide variety of model development. Additionally, NetMind’s collaboration with established AI companies. Working with Nvidia and the University of Edinburgh are several examples of NetMind’s dedication to innovation.

Qualitative Competition Landscape

Protocol Go To Market Strategy Versus Competitors

NetMind’s closest competitors are Bittensor and SingularityNET (AGIX). Compared to Bittensor, NetMind’s partnerships with universities and focus on growing its model ecosystem puts it in a strong position. Bittensor subnets and access to many models is a strength, however it does not as actively seek out development of models the way NetMind does. NetMind’s dedication to innovation can be seen in its continuous partnerships and academic research. NetMind Chat, Avagi, and Life are all examples of how NetMind’s focus on AI models impact the products and services it offers.

SingularityNET has strong partnerships such as its recent alliance with Ocean Protocol and Fetch AI, however, we believe NetMind’s partnerships with universities are more important as they push the frontier of AI. NetMind’s recent partnerships with ZK Hive and Haiper are similarly important collaborations.

NetMind’s partnerships solidify it as a strong competitor in the AI industry. Its partnerships with universities, participation at the Nvidia GTC conference, and even hosting Nvidia in it’s London office are all indicative of NetMind’s potential with AI and the evolution of machine learning. Rather than competitors that only focus on the decentralized compute or only focus on the model infrastructure, NetMind has a focus on the models themselves. Combining all of this with its strong infrastructure such as NetMind Power and NetMind Chain make it the perfect candidate for AI development and innovation.

How Token Extracts Value

NMT can be used to elect one of the 21 Master Nodes through staking, allowing participation in governance. Additionally, NMT is the medium of exchange used to pay for various fees on the NetMind platform and chain, in addition to serving as a reward for network participation.

Network Connection Reward

Volunteering computing resources is one way providers can earn NMT. If the machine is connected to the network for more than 5 non-consecutive hours in one 24 hour period. The reward is determined by the GPU type, the network bandwidth speed, and the amount of connected machines on any given day. In addition to the connection reward, the volunteers receive the compute fee associated with training and fine-tuning jobs.

Fees

Training and fine-tuning models on the NetMind Power platform do so by paying with NMT. Training synchronously gives 50% of the compute power fee to the volunteer provider, the remaining 50% will be burned. Training asynchronously gives 20% of the utilization fee to the volunteer provider and 80% will be burned. If a volunteer disconnects mid-project, causing the job to fail, they will not receive the compute fee but will still receive the connection reward.

Staking Rewards

Users can stake NMT with any of the 21 main nodes and receive NMT rewards.

Mining Rewards

Users can join as node operators to receive rewards for node operation and Gas fee rewards from users training models or using services on the platform.

Tokenomics/Vesting Schedule

With the total supply of NMT being 10 billion tokens, NetMind Power has stated in their whitepaper the following distribution plan:

40% of the total supply (4 billion NMT) goes toward Volunteer Computing Providers
30% of the total supply is reserved for Staking rewards
20% of the total supply (2 billion NMT) will go to the DAO community
10% of the total supply (1 billion NMT) will go toward the Technical Team which will be unlocked at a rate of 0.1% over 100 years

The power utilization fee illustrates the fees associated with the computational transaction. The two types of rates are “Sync Mode” & “Async Mode.” In sync mode, 50% of the computational power utilization fee will go towards the power supplier while the remaining 50% will be burned. In async mode, 20% of the total computational power will go toward the power supplier while the remaining 80% will be burned.

Quantitative Comparable Analysis

AI projects do not have easily observable KPIs such as volume or TVL. While fees generated and TVL are good KPIs for blockchains such as NetMind Chain, being in such an early stage, this data is not retrievable and gives little insight into whether or not the project is undervalued or overvalued. Instead of looking at KPIs, the compound annual growth rate (CAGR) of the generative AI industry and the semiconductor industry serve as good proxies for the amount of capital expected to flow into the space in the coming years. Compared to all other competitors, NetMind has the lowest valuation. While it has the highest FDV, we do not see this as a concern as a majority of this locked supply is allocated to staking rewards, which will be emitted slowly over time, and the DAO. While low valuation alone isn’t enough to suggest NMT is a good investment, we believe that its strong fundamentals and partnerships will generate outperformance in the industry in the coming years.

Road Map

Platform Enhancements

The future objectives for NetMind Power and the potential impact the platform could have on the AI landscape are apparent, to say the least. The team plans to support more machine learning frameworks and enhance security & privacy mechanisms, and overall performance throughout the platform. The goal is to increase market share and remain attractive to a wide range of users both on the buyer and seller side of the platform.

Community

Community is another core aspect of the protocol as it is the backbone of the entire operation. The platform itself acts as its community in hopes of fostering a global and interconnected atmosphere, beneficial for all. Creating a thriving ecosystem could be achieved by providing the means for users to train models on the platform and publish them publicly. Since the platform doubles as both a transactional computing marketplace and a forum-like space, staying connected and communicative has never been easier. Between the social/interconnected nature of the internet with the transparency that blockchain offers, all users on the platform can rest easy knowing that everybody is on the same page, and therefore a sense of trust can be created as a result.

Decentralized Pricing

According to Deloitte, “79% of respondents expect generative AI to transform their organizations within three years” proving how monumental artificial intelligence adoption could be within the world. Regardless of platform participation, $NMT as a token itself will undoubtedly gain momentum as the platform continues to grow. With the current rate of growth within the sector, NetMind Power possesses a skeleton for a successful business model down the line and it can be said that decentralized protocols such as $NMT may also ride the wave of success from other more traditional blue chip companies like NVIDIA and others alike.

Investment Thesis

NetMind’s infrastructure to facilitate the development of AI models combined with its focus on model development set it up as a promising platform in the AI space. Clearly the team has put thought into NetMind Chain and NetMind Power, enabling model development. This alone is not enough to necessitate an investment, but the team has also been making big moves towards model development.

Companies developing on the NetMind platform:

ProtagoLabs: Utilize machine learning research, technology, and big sata to establish human-like machine perception systems and achieve safe AGI.
Hairper: Setting a new standard for graphics technology through AI.
Autoedge: End-to-end data science platform that uses AI to find insights.
Qdot Technology: Hybrid-electric propulsion system development to improve aircraft safety, efficiency, and performance with no carbon emissions.
Orbit: Mine news feeds and unstructured data at speed and scale to identify risk based on set search parameters.

Collaborating research institutions:

University of Edinburgh: Research in computing systems and ML.
Tsinghua University: Using Graph Neural Networks for anomaly detection.
Rice University: Use neural architecture search to automatically tune deep NNs.
University of Cambridge: Use NLP for education and commercial needs.
Huazhong University: Working on data/model privacy, security, and learning.
Shanghai Jiao Tong University: Research NLP systems to better understand data.
University of Wisconsin-Madison: Research on multilingual NLP
Fudan University: Predict RNA-protein interactions to facilitate therapy development.

Furthermore, the team is incredibly qualified with experience in AI, web3, and fintech. The team has also been communicating with large firms in the industry, as evident by NetMind’s presence at the Nvidia GTC conference and inviting Nvidia to their London office. Considering the partnerships, team, and collaboration that NetMind has been undertaking, we believe this project will be able to capture attention as increased capital flows into the generative AI industry.