Artificial Intelligence, Blockchain and Crypto: A Confluence

• Funds raised by companies in the artificial intelligence (AI) sector since 2019 stands at over USD 220 billion. Demand for AI tools and high-end hardware rises. 
• To meet surging demand, decentralised solutions emerged, and crypto markets saw a new narrative: the confluence of AI and blockchain technology. 
• AI-powered onchain agents and decentralised physical infrastructure networks (DePINs) are the two main AI-themed sectors in crypto. 
• Bittensor, Fetch.ai, Render Network and Worldcoin are the top AI-themed crypto projects in terms of market capitalisation and popularity. 
• Zero knowledge (ZK) machine learning (ML) shows promise as a possible solution to balance the data-intensive nature of AI and the cypherpunk privacy-first ethos of crypto.  
• The sector is promising, however, multiple challenges currently exist which include service provider reliability and lagging demand. 
• While the possibilities for AI and crypto are exciting, it would be wise to not have unrealistic expectations of what could be possible. 

Much recently, cryptocurrencies themed around artificial intelligence (AI) were all the rage in crypto. Experts called for an AI revolution in the cryptocurrency markets. Currently, according to CoinGecko, the combined market capitalisation of digital assets in this narrative stands at over USD 44 billion.

At one point in time, any project with the term “AI” in its pitch automatically became a crypto-Twitter favourite and was being shilled by self-proclaimed crypto experts on Twitter. However, this narrative soon changed for the better.

Some AI crypto projects were unable to deliver on their promises. How these AI crypto projects branded themselves vs what they became was a topic of open discussion. This is when users began paying closer attention to detail and this pushed projects to be practical when setting expectations for its users.

So how did this AI hype and market narrative come about? What are some top projects building in this space? How does the future for this sector look? These will be addressed in this article. Let’s dive into it.

Free, open markets have always incentivized leaps in human productivity. The industrial revolution was the biggest jump we witnessed in this regard – something that forever changed the way markets function. From transportation to manufacturing, it offered a better lifestyle and brought about a structural change in the job market.

Fast forward, the dot com boom too radically changed our daily life. It brought more investments into technology companies and led to the displacement of less efficient ones. Companies that leveraged tech took the lead. This lead is now widening with the advent of artificial intelligence.

AI companies have brought about a revolution in increasing human efficiency and the sector has attracted investments. According to Statista, since 2011, total funding into this sector has been over USD 284 billion while in 2023 alone, total funding of companies offering AI services exceeded USD 40 billion.

The breakthrough in AI was brought about by Open AI’s large language model (LLM) ChatGPT which put the technology into the hands of the common man through a simple chatbot interface. Most AI solutions in the market were based on generative AI (Gen AI).

For context, generative AI models are those that can generate text, images, videos or other data in response to prompts (requests). This includes ChatGPT which reached a million in user base in merely 5 days. It currently has over 180 million monthly active users.

This growth inspired further research and developments in the AI chatbots sector. For example, Google launched Bard, then ended the year with another one, Gemini. Baidu launched Ernie. Meta put out various AI tools. Amazon launched Amazon Q, although only in preview. Image creating tools also proliferated and improved. Midjourney led the way in terms of providing the richest and most complex output but OpenAI’s DALL-E too advanced rapidly.

Meanwhile, even open-source efforts emerged. HuggingFace has now become a leader in open-source generative AI. This even led to structural changes in the job market. For example, companies increased job postings for machine learning (ML) engineers to build AI systems.

For the uninitiated, ML is a branch of AI that focuses on enabling a machine to imitate intelligent human behaviour. It involves training AI algorithms on various data sets to ensure it provides an optimal output for a specific input.

AI models are computationally intensive to run. The hardware resources required to run an AI program can vary depending on several factors like the complexity of the program, size of the datasets required to train the model and the desired performance level.

While computer central processing units (CPUs) were key in running these complex programs, graphic processing units (GPUs) were the game changer. CPUs are versatile and can handle a variety of input types. However, GPUs enable parallel processing in computer systems.

This helps process data several orders of magnitude faster than a CPU. With most AI models built around graphics rendering (DALL-E and Midjourney, for example) and natural language processing (ChatGPT is an example), the demand for GPUs went up.

Every tech enthusiast soon wanted to customise open-source AI models according to their requirements and on their own machines. Developers and tech companies therefore began buying GPUs in bulk. This led to a demand explosion and a spike in prices across the globe for GPUs.

Once it became difficult to get their hands on such hardware to run models, developers looked to non-traditional rails to do the same – the decentralised web. Thus came about a new wave: the crossover of AI, decentralised computing and cryptocurrencies.

AI-themed cryptocurrency projects and tokens are ones that either provide AI solutions or remotely provide resources to run AI tech. They mark a key narrative in the blockchain industry. By combining AI and blockchain technology, AI-based crypto projects aim to solve issues such as GPU scarcity and help users who want to leverage AI but lack the technical know-how.

Implementing AI on blockchains will also allow smart contracts to securely query machine learning models. This will create a new area of blockchain technology known as zero-knowledge machine learning (ZKML). We will explore this in a section further below.

For now, let’s explore the two largest AI-based sectors we have in crypto at the moment – decentralised physical infrastructure and onchain AI agents.

As discussed earlier, a surge in popularity of AI chatbots like ChatGPT and Bard sparked an increased need for graphics processing units (GPUs) which are crucial for running AI. This surge caused a global GPU shortage, prompting cloud giants like Amazon Web Services (AWS), Microsoft, Google and Oracle to limit GPU access for customers.

Wait times reportedly stretched for months. This is when developers began considering decentralised compute networks, also called decentralised physical infrastructure networks (DePINs) as a viable alternative.

DePINs connect people who need computing power with unused CPU or GPU resources. These networks connect to unused GPUs around the world, which can help balance supply and demand.

They also tend to offer lower prices than centralized providers because individuals and organizations do not have to pay as much in additional costs to provide their unused resources. Let’s have a look at some projects building such solutions.

• Render Network, established in 2020, is a decentralized network for cloud-based 3D rendering. It bridges the gap between node operators (with GPU resources) and users who need content rendering services. This content can be a simple image, animation or motion graphics. Users can submit rendering requests which are then fulfilled by node operators using their GPU resources. The finished renders are returned and node operators receive RNDR tokens as compensation for their services.
• Io.net is another open-source network of GPUs dedicated to machine learning (ML) applications. It has secured USD 30 million in funding and is currently valued at USD 1 billion. It has also partnered with Render and the data storage network Filecoin. It also allows users of Apple devices to contribute to the network through support for Apple silicon chips, promoting retail participation. The project has plans to launch its own token.
• Akash Network is a decentralised open-source cloud computing platform. The network connects server owners that need computing power to host applications with cloud computing resources (GPUs) through its marketplace. The network runs on a Tendermint-based blockchain built using the Cosmos SDK. At the time of writing, it has a computation resource pool of 74 GPUs and close to 5.6K CPUs. A USD 3.5 million incentive program is underway to attract more GPU providers to allocate Nvidia chips to the network for a minimum duration of one year.
• Grass, previously known for providing residential proxies and unused internet bandwidth monetization, has pivoted to providing data for machine learning. It offers web scraping services to gather data from websites and processes it to meet the needs of AI applications. To improve its data quality, Grass engages users in data labeling and is looking to create models that automate this task. Their goal is to establish a data infrastructure which can be leveraged by any AI model deployed onchain.

Another use-case for AI in crypto are onchain agents. AI has opened new ways to design smart contracts for decentralised finance (DeFi) dapps. Developers can now make tools and software that make on-chain transactions easier for end users, reducing the need for manual execution.

AI agents in smart contracts can also get more complex. For instance, in on-chain games or AI-based marketplaces, AI characters could interact with each other on their own, carrying out actions based on what the user wants them to do.

An advantage is that infrastructure designed without central control (as is the case with DeFi) suits AI agents. This is because there are fewer barriers than in traditional financial systems.

For example, crypto marketplaces do not require personal identity verification (KYC) which is standard in traditional financial services AI agents make it easier for users who may not understand blockchains well enough to interact with them by themselves. They connect directly with the underlying smart contracts, letting them do tasks that would be hard for regular users who use interfaces that are meant to be easy to use.

Some use cases of onchain AI agents might be as follows.

• Automation in DeFi: Protocols in DeFi could be greatly automated through AI. For instance, lending platforms could employ machine learning (ML) models to dynamically adjust parameters in real time. Currently, lending protocols rely heavily on off-chain models managed by their respective decentralised autonomous organisations (DAOs) or sometimes even private organizations for determining parameters like collateral factor, loan-to-value (LTV), liquidation thresholds and other risk factors. A possible approach would be to use open-source models trained by the community, which can be publicly monitored and validated.
• On-chain credit assessment: A machine learning (ML) model can use data from a user’s wallet activity, past transactions and behaviour onchain to create a clear and trustworthy way to evaluate their creditworthiness on the network. If such models can be open-sourced, it would be a stepping-stone for permissionless collateral-free lending in DeFi to gain adoption.
• Gaming: As previously stated, blockchain-based gaming environments can leverage artificial intelligence (AI) to enhance their offerings. For example, AI development can assist in the creation of visual game content including characters, objects and environments. Additionally, non-player characters (NPCs) can leverage AI to adapt and customize themselves based on a player’s profile and in-game surroundings, thus elevating the immersive experience.

These concepts may sound futuristic. However, groundwork is being carried out by many protocols. Protocols currently are very basic in terms of decentralising AI services. They are either done manually by blockchain network securing entities called nodes or provide tools to developers to launch AI-powered agents onchain.

Let us have a look at two of the most popular ones in this space – Bittensor and Fetch.

• Bittensor is a platform where users can rent out their AI models to be used in various applications. It has multiple subnets, each specializing in a particular machine learning task or providing specific resources. These subnets offer a variety of services such as AI chatbots, financial prediction tools, pre-trained models, data storage and more. To access these services, users must go through subnet validators, who serve as the gatekeepers for each subnet. According to Tao Stats, 32 subnets and 1001 active miners exist at the time of writing.
• Fetch.ai is a platform that provides tools and software for developers to create AI agents. Recently, Fetch launched Fetch Compute – a program investing USD 100 million to purchase Nvidia H100s and A100s. These GPUs will be used by developers and users for computing tasks on the Fetch platform. By contributing to the network, users who stake the project’s native FET token can earn Fetch Compute Credits for access to GPU use.

The integration of AI and cryptocurrency requires balance. While AI systems require extensive data, the crypto’s cypherpunk culture emphasizes privacy and anonymity. This poses a challenge in improving the accuracy of AI models without compromising data integrity. This is where zero knowledge proofs (ZKPs) come in. 

ZKPs offer a potential solution by allowing verification without exposing sensitive information. They allow a person (prover) to show that something is true without sharing additional details. When combined with machine learning (ML), this technology enables the generation of results without revealing the sensitive data used for training.  

This technology guarantees both computational accuracy and data privacy. This allows AI models to demonstrate their effectiveness publicly while maintaining secrecy. This convergence of ZKPs and ML is known as ZKML – zero knowledge machine learning. Interested readers can dive deeper into ZKP technology through our blog on the topic here.  

• Worldcoin is a notable project that utilizes ZKML technology. Its ZK-based iris recognition system verifies the uniqueness of individuals without disclosing specific iris features or model parameters. ZKML has numerous potential applications. For instance, decentralized social media platforms (like Lens Protocol or Farcaster) could incorporate ZKML to develop tailored social feed recommendations based on each user’s blockchain activities and interactions.

AI relies heavily on computational power and data availability. While the past year has seen significant traction for GPU networks leading to a crowded market, data-focused projects have received less attention.

However, this trend is changing, with growing emphasis on data-centric initiatives. The potential for incentivized, crowdsourced data networks remains largely unexplored and holds promising prospects. 

There is market optimism about the potential of combining cryptocurrency and artificial intelligence. AI-related cryptocurrencies continue to be highly popular. This year, the global AI market is projected to reach USD 305 billion and have a CAGR of over 15% until 2030. Parallelly, the GPU-as-a-service market is expected to expand significantly, reaching USD 28.7 billion by 2030. However, despite these promising figures, there are still challenges to overcome in this field. 

Growth in GPU networks depends on balancing supply and demand. To expand, networks must increase demand from paying customers and attract and fulfill the supply of computational resources. However, a challenge exists in that suppliers require income to participate while customers need enough suppliers to meet their needs.  

This is because even though centralised computing services face a supply crunch, the supply of  decentralised computing resources far exceeds demand. For instance, Akash Network’s utilization rate has dropped from 40% in January to around 30% currently.

Cheaper decentralized compute networks face challenges in attracting users who can get their needs met by traditional GPU providers, especially since the cost savings are not as substantial as advertised. Moreover, the demand for these networks is further complicated by the different utilization rates of GPUs: low-end GPUs are often underutilised while high-end ones like the Nvidia A100s and H100s are consistently operating at maximum capacity, leading to limited availability and undermining potential supply. 

When AI companies and users compare cloud computing services from web2 and web3 platforms, they prioritize both cost and reliability. While web3 platforms like DePIN offer lower costs, they often face issues with reliability and reputation.

Therefore, improving on reliability will strengthen demand. At the same time, offering token rewards and sharing in the network’s growth incentives will attract suppliers to join the network. This approach provides economic incentives for new suppliers to join and contribute to the network’s growth.

Many cryptocurrencies labeled as AI are currently riding a wave of hype, leading to inflated prices for these projects. This excitement around AI resembles the initial hype around crypto and blockchain.

While crypto and blockchain have had real impacts and will continue to evolve, they have not lived up to all the exaggerated claims made at the beginning. The same could happen with AI: the hype might overshadow its potential, causing disappointment.  

Many use cases are still in their infancy and most projects may not endure beyond this market cycle, although a select few might. Investors need to be prudent to meticulously select their investments based on fundamentals rather than solely riding the current hype wave.

Artificial intelligence, blockchain and cryptocurrencies have enormous potential for collaboration, offering many advantages. The possibilities are vast and exciting, but we must avoid having unrealistic expectations in terms of what could be possible.