TL;DR

• ARC is using AI to solve the severe player liquidity problem in indie and web3 games. It’s a big market.

• With the launch of ARC and ARC RL, the project is no longer just a single-title game studio (AI Arena)—it’s now positioning itself as a platform and AI play. This shift should lead to a re-rating of the NRN token.

• ARC B2B is an AI-powered gaming SDK that can be easily integrated into any game. ARC partners directly with game studios to deliver AI-powered gaming experiences.

• ARC Reinforcement Learning (RL) uses crowdsourced human gameplay data to train “superintelligent” game agents.

• ARC’s business model ties its success to the game studios it integrates with. Revenue flows are based on token allocations (in Web3 games) and royalty-based payments from the games.

• ARC can generalize across game genres. Over time, this enables them to aggregate a one-of-a-kind reservoir of gameplay-specific data, which can be leveraged to train a generalizable AI model for game development.

Back in 2021, I was an Axie Infinity player and ran a small scholarship guild. 

And if you weren’t around in those times, let me tell you — they were absolutely wild.

Axie Infinity was the game that made people realize crypto and gaming could actually be a thing. At its core, it was a simple, Pokémon-style turn-based strategy game where you’d assemble a team of three Axies (adorably ferocious fighters), each with unique abilities. You’d take your squad, battle other teams, and earn SLP tokens for participating and winning.

But what really got non-gamers excited was the potential to profit from playing. Axie had two standout mechanics that drove its meteoric rise:

The first was Breeding Axies. Take two Axies, breed them using SLP tokens, and voilà—a new Axie with a unique mix of its parents’ abilities. Rare, overpowered Axies (OP Axies, for the gamers) became a hot commodity, and a bustling breeding market emerged.

Secondly, scholarship programs. Entrepreneurial players from around the world began lending out Axies to “scholars”. These were players, often from developing countries like the Philippines or Argentina, who couldn’t afford the $1,000+ upfront cost of owning three Axie NFTs required to play. Scholars would play daily, earn tokens, and split the profits with their guilds, who typically take a 30–50% cut.

At its peak, Axie had a significant impact on the local economy of developing countries, especially during the COVID-19 pandemic. Many players in the Philippines, where ~40% of Axie Infinity’s user base was located, could earn incomes significantly higher than minimum wage. Guilds profited handsomely.

These programs solved a key problem for game devs: player liquidity. By incentivizing players to be actively playing for hours a day, Axie ensured every player always had an opponent waiting, making the player experience more engaging.

But there was a tradeoff.

To solve the player liquidity problem, Axie gave away a massive amount of tokens to incentivize participation. And here’s where things unravelled. With no cap on SLP, the token inflated like crazy, prices tanked, and the ecosystem collapsed. When the token lost value, players left. Axie went from the poster child of play-to-earn to a cautionary tale almost overnight.

But what if there were a way to solve the player liquidity problem without unsustainable tokenomics?

That’s exactly what ARC / AI Arena has been quietly working on for the past three years. And now, it’s starting to bear fruit.

(Note: The team behind Axie, Sky Mavis, has since evolved the game into something different and remains a leading Web3 gaming studio today)

Player Liquidity = Lifeblood

Player liquidity is the lifeblood of multiplayer games and the key to long-term success.

Many Web3 and indie games struggle with the “cold start” problem—too few players for quick matchmaking or thriving communities. They don’t have the marketing budgets or natural IP awareness that large game studios have. This results in long wait times, mismatched opponents, and high churn.

These games often end up in a slow, painful death. RIP.

Hence game developers must prioritize player liquidity from the outset. Games require varying levels of activity to stay fun—chess needs two players, while large-scale battles need thousands. Skill-based matchmaking further raises the bar, demanding a larger player pool to keep games fair and engaging.

For Web3 games, the stakes are higher. According to Delphi Digital’s yearly gaming review, user acquisition costs for Web3 games are 77% higher than traditional mobile games, making player retention critical.

A strong player base ensures fair matchmaking, vibrant in-game economies (more buying and selling of items), and more active social interactions, which makes games more enjoyable.

ARC — Pioneering AI-Driven Gameplay

ARC, by ArenaX Labs, is pioneering the AI-driven future of online gaming experiences.

In a nutshell, they use AI to solve the player liquidity problem that plagues newer games.

The problem with most AI bots in games today is that they’re terrible. Once you’ve spent a few hours learning the ropes, these bots become laughably easy to beat. They’re designed to help new players but don’t provide much challenge or engagement for experienced ones.

Imagine AI players with skills rivalling those of top human players. Imagine playing against them anytime, anywhere, without waiting for matchmaking. Imagine training your AI player to mimic your playstyle, owning it, and earning rewards from its performance.

This is a win-win for both players and studios.

Game studios use human-like AI bots to populate their games, boosting player liquidity, improving user experiences, and increasing retention—key factors for new titles trying to survive in a competitive market.

Players gain a new way to participate in the game, building a stronger sense of ownership as they train and compete with their AI.

Let’s look at how they do this.

Product & Architecture

ArenaX Labs is the parent company building a suite of products to tackle the player liquidity problem.

1. Existing: AI Arena, an AI fighting game

2. New: ARC B2B, an AI-powered gaming SDK that can be easily integrated into any game

3. New: ARC Reinforcement Learning (RL)

#1. AI Arena: The Game

AI Arena is a brawler-style fighting game reminiscent of Nintendo’s Super Smash Bros. It features quirky, cartoonish characters battling it out in an arena.

But in AI Arena, every character is controlled by AI—you don’t play as a fighter but as their coach. Your job is to train your AI fighter using your strategy and expertise.

Training your fighter is like preparing a student for battle. In training mode, you turn on data collection and create combat scenarios to fine-tune their moves. For example, if your fighter is close to their opponent, you might teach them to block with your shield and follow up with a punch combo. At a distance? Train them to launch a ranged attack to close the gap.

You control what data is collected, ensuring only the best moves are recorded for training. After practice, you can refine hyperparameters like learning rates and batch sizes for a more technical edge, or simply use the beginner-friendly default settings. Once training is complete, your AI fighter is ready to compete.

Getting started isn’t easy—training an effective model requires time and experimentation. My first fighter repeatedly fell off the platform without being hit by the opponent. But over several iterations, I managed to create a model that could hold its own. It’s humbling but deeply satisfying to see your training pay off.

AI Arena introduces additional depth through NFT-based fighters. Each NFT character has unique cosmetic traits and combat attributes—like elemental effects—that influence gameplay. This adds another strategic layer (more details in the game documents)

Currently, AI Arena is available on Arbitrum mainnet and is accessible only to those with AI Arena NFT, keeping the community exclusive while gameplay is refined. Players can join Guilds, pooling champion NFTs and NRNs for ranked on-chain battles with rewards and guild multipliers. This is done to attract dedicated players and fuel a competitive scene.

Ultimately, AI Arena is a showcase for ARC’s AI training technology. While it’s the entry point into their ecosystem, the real vision extends far beyond this single game.

Which brings us to…

#2. ARC: The Infra

ARC is a purpose-built AI infrastructure solution designed specifically for gaming.

The ArenaX team started from the ground up, even developing their own game infrastructure because existing solutions like Unity and Unreal couldn’t match the scope of their vision.

Over three years, they crafted a robust tech stack capable of handling data aggregation, model training, and model inspection for imitation and reinforcement learning. This infrastructure is the backbone of AI Arena, but its potential is much greater.

As the team refined their technology, third-party studios approached ARC, eager to license or white-label the platform. Recognizing this demand, they formalized ARC’s infrastructure as a B2B product.

Today, ARC partners directly with game studios to deliver AI-powered gaming experiences. The value propositions are:

1. Permanent Player liquidity as a service

2. Bringing AI gameplay as a simple integration

Permanent Player Liquidity as a Service

ARC focuses on human behaviour cloning—training specialised AI models to mimic human actions. This differs from the dominant use of AI in gaming today, which uses generative models to create game assets and LLMs to power dialogue.

With the ARC SDK, developers can create human-like AI agents and scale them to fit their game’s needs. The SDK simplifies the heavy lifting. Game studios can bring in AI without dealing with the intricacies of machine learning.

After integration, deploying an AI model requires just one line of code, with ARC handling infrastructure, data processing, training, and deployment on the backend.

ARC takes a collaborative approach with game studios, helping them:

1. Capture raw gameplay data and convert it into meaningful datasets for AI training.

2. Identify key gameplay variables and decision points tied to game mechanics.

3. Map AI model outputs to in-game actions, ensuring smooth functionality—for example, linking an AI’s “punch right” output to a specific game control.

How the AI works

ARC uses four types of models tailored to game interactions:

1. Feedforward Neural Networks: Ideal for continuous environments with numerical features like speed or position.

2. Tabular Agents: Ideal for games with finite, discrete scenarios.

3. Hierarchical & Convolutional Neural Networks are in development.

There are two interacting spaces relating to ARC’s AI model:

The state space defines what the agent knows about the game at any given moment. For feedforward networks, this is a combination of input features (like a player’s speed or position). For tabular agents, it’s discrete scenarios the agent might encounter in the game.

The action space describes what the agent can do in the game, from discrete inputs (e.g., pressing buttons) to continuous controls (e.g., joystick movements). This is mapped to game inputs.

The state space provides inputs to ARC’s AI model, which processes them and generates outputs. These outputs are then translated into game actions through the action space.

ARC collaborates closely with game devs to identify the most critical features and design the state space accordingly. They also test various model configurations and sizes to balance intelligence and speed, ensuring smooth and engaging gameplay.

According to the team, demand for their player liquidity service is particularly high among Web3 studios. Studios pay for access to better player liquidity, and ARC will reinvest a significant portion of that revenue into NRN token buybacks.

Bringing AI Gameplay to Players: The Trainer Platform

The ARC SDK also enables studios to access a Trainer Platform for their game, allowing players to train and submit agents.

Like in AI Arena, players can set up simulations, capture gameplay data, and train blank AI models. These models evolve over time, retaining previous knowledge while incorporating new gameplay data, eliminating the need to start from scratch with every update.

This opens exciting possibilities: players could sell their custom-trained AI agents in a marketplace, creating a new layer of in-game economy. In AI Arena, skilled trainers form guilds, and they can offer their training expertise to other studios.

For studios that fully integrate agent capabilities, the concept of Parallel Play also comes to life. AI agents, available 24/7, can participate in multiple matches, tournaments, or game instances simultaneously. This resolves player liquidity issues and opens new opportunities for engagement and revenue generation.

But….that’s not all…

#3 ARC RL: From one-to-one to many-to-one

If AI Arena and ARC Trainer Platform feel like single-player modes—where you train your personal AI model—ARC RL is akin to the multiplayer mode.

Picture this: an entire gaming DAO pooling its gameplay data to train a shared AI model that everyone co-owns and benefits from. These “master agents” represent the combined intelligence of all the players, transforming esports by introducing competition fueled by collective effort and strategic collaboration.

ARC RL uses reinforcement learning (that’s the “RL”) and crowdsourced human gameplay data to train these “superintelligent” agents.

Reinforcement learning works by rewarding agents for optimal actions. It works especially well in games because the reward functions are clear and objective, like damage dealt, gold earned, or victories. 

There are precedents for this:

AlphaGo by DeepMind defeated professional human players in Go by playing millions of self-generated matches, refining its strategies with each iteration.

I hadn’t realized it before, but OpenAI was already well-known in gaming circles long before chatGPT was created.

OpenAI Five used RL to dominate top human players in Dota 2, defeating the world champions in 2019. It mastered teamwork and advanced strategies through accelerated simulations and massive computational resources.

OpenAI Five trained by running millions of games daily—equivalent to 250 years of simulated gameplay per day—on a powerful setup of 256 GPUs and 128,000 CPU cores. By skipping graphics rendering, it accelerated learning dramatically.

Initially, the AI displayed erratic behaviour, like wandering aimlessly, but quickly improved. It mastered basic strategies like farming creeps in lanes and stealing resources, eventually progressing to complex manoeuvres such as ambushes and coordinated tower pushes.

The key idea in RL is that the AI agent learns how to succeed through experience rather than being directly told what to do.

ARC RL differentiates itself by using offline reinforcement learning. Instead of the agent learning from its own trial and error, it learns from the experiences of others. This is like the student watching videos of others riding a bike, observing their successes and failures, and using that knowledge to avoid falling and improve faster.

This approach provides the opportunity for an added twist: the collaborative training and co-ownership of models. This not only democratizes access to powerful AI agents but also aligns incentives for gamers, guilds, and developers.

There are two key roles in building a “superintelligent” game agent:

1. Sponsors: Guild-like leaders who stake significant NRN tokens to kickstart and manage the RL agent. Sponsors can be any entity but will likely be gaming guilds, DAOs, web3 communities, and even popular on-chain personality agents like Luna.

2. Players: Individuals who stake smaller amounts of NRN tokens to contribute their gameplay data to train the agent.

Sponsors coordinate and guide their team of players, ensuring high-quality training data that gives their AI agent a competitive edge in agent-based competitions.

Rewards are distributed based on the super agents’ performance in competitions. 70% of rewards go to the players, 10% to the Sponsor, and the remaining 20% is kept in the NRN treasury. This structure aligns incentives for everyone involved.

Data contributions

How do you get players excited about contributing their gameplay data? Not easy.

ARC makes contributing gameplay data simple and rewarding. Players don’t need expertise—just play the game. After a session (e.g., Mario Kart), they’re prompted to submit data to train a specific agent. A dashboard tracks their contributions and supported agents.

ARC’s attribution algorithm ensures quality by evaluating contributions and rewarding high-quality, impactful data.

Interestingly, your data can be useful even if you’re a lousy player (like me). Poor gameplay helps agents learn what not to do, while skilled gameplay teaches optimal strategies. Redundant data, like repetitive farming, is filtered out to maintain quality.

In short, ARC RL is designed as a low-friction, mass-market product centred on the co-ownership of agents that surpass human capabilities.

Market Size

ARC’s technology platform is versatile and designed to operate across multiple genres such as shooters, fighting games, social casinos, racing, trading card games, and RPGs. It is tailor-made for games that need to keep players engaged.

There are two natural markets that ARC targets with its products:

ARC primarily focuses on indie developers and studios rather than large, established ones. These smaller studios often struggle to attract players early on due to limited branding and distribution resources.

ARC’s AI agents solve this problem by creating a vibrant in-game environment from the get-go, ensuring dynamic gameplay even during a game’s initial stages.

It may be surprising to many, but the indie game sector is a major force in the gaming market:

• 99% of games released on Steam are indie titles (Source)

• Indie games generated 48% of total revenue on Steam in 2024.

Another target market is Web3 Games. Most Web3 games are developed by new studios, which also face unique challenges like wallet onboarding, crypto scepticism, and high customer acquisition costs. These games often suffer from player liquidity issues, where AI-driven agents could fill gaps in matches and keep gameplay engaging.

While Web3 gaming has struggled recently due to a lack of compelling experiences, signs of revival are emerging.

For example, “Off the Grid”—one of the first AAA Web3 games—achieved early mainstream success recently, with 9 million wallets created and 100 million transactions in its first month. This paves the way for broader success in the sector, creating opportunities for ARC to support this resurgence.

The ARC Team

The founding team behind ArenaX Labs has a wealth of machine learning and investment management expertise.

Brandon Da Silva, CEO and CTO, previously led ML research at a Canadian investment firm specializing in reinforcement learning, Bayesian deep learning, and model adaptability. He spearheaded the development of a $1 billion quant trading strategy centred on risk parity and multi-asset portfolio management.

Wei Xie, COO, managed a $7 billion liquid strategies portfolio at the same firm and chaired its innovation investments program, focusing on emerging fields like AI, machine learning, and Web3 technologies.

ArenaX labs raised a $5M seed round in 2021 led by Paradigm and with participation by Framework ventures. It raised a follow-on round of $6M in January 2024, led by Framework with SevenX Ventures, FunPlus / Xterio and Moore Strategic Ventures participating.

NRN Token-economics — A Healthy Revamp

ARC/AI Arena has a live token, NRN. Let’s first take stock of where we are today.

Examining the supply-side and demand-side dynamics will give us a clearer picture of where this might be heading.

Supply Side

The total supply of NRN is 1B, of which ~409M (40.9%) is in circulation today.

At the time of writing, the token price is $0.72, implying a market cap of $29M and a fully diluted valuation of $71M.

NRN was launched on 24 June 2024, and the 40.9% circulating supply comes from

• Community airdrop (8% of total)

• Foundation treasury (2.9% unlocked out of 10.9% total, 36 month linear unlock)

• Community ecosystem rewards (30%)

Most of the circulating supply (30% out of the 40.9%) consists of community ecosystem rewards, which the project manages and strategically allocates for staking incentives, in-game rewards, ecosystem growth initiatives, and community-driven programs.

The unlock schedule is reassuring, with no major events in the near term:

• The next unlock is from the foundation’s OTC sale (1.1%), starting in December 2024 and unlocking linearly over 12 months. This adds just ~0.09% to monthly inflation and is unlikely to cause significant concern.

• Investor and contributor allocations (50% of total supply) won’t begin unlocking until June 2025, and even then, it will be released linearly over 24 months.

For now, sell pressure is expected to remain quite manageable, primarily stemming from ecosystem rewards. The key will be trust in the team’s ability to deploy these funds strategically to drive protocol growth.

Demand Side

NRN v1 — Player-based Economy

Initially, NRN was designed as a strategic resource tied exclusively to the AI Arena game economy.

Players stake NRN on their AI players, earning rewards from a pool if they win and losing part of their stake if they lose. This creates a “skin-in-the-game” dynamic, turning it into a competitive sport with financial incentives for skilled players.

Rewards are distributed using an ELO-based system, ensuring balanced payouts based on skill. Other revenue streams include in-game item purchases, cosmetic upgrades, and tournament entry fees.

This initial token model relies entirely on the game’s success and a steady influx of new players willing to buy NRN and NFTs to participate.

Which brings us to why we’re so excited…

NRN v2 — Player & Platform-based Economy

NRN’s revamped v2 tokenomics introduces powerful new demand drivers by expanding the token’s utility beyond AI Arena to the broader ARC platform. This evolution turns NRN from a game-specific token into a platform token. This is hugely positive, in my opinion.

The 3 new demand drivers for NRN include:

1. Revenue from ARC integrations. Game studios integrating ARC will generate revenue for the treasury through integration fees and ongoing royalties tied to game performance. Treasury funds can fuel NRN buybacks, grow the ecosystem, and incentivize players on the Trainer platform.

2. Trainer Marketplace fees. NRN captures value from fees in the Trainer Marketplace, where players can trade AI models and gameplay data.

3. Staking for ARC RL participation: Both sponsors and players must stake NRN to join ARC RL. As more players dive into ARC RL, the demand for NRN scales accordingly.

What’s especially exciting is the inclusion of revenue from game studios. This marks a shift from a purely B2C model to a hybrid B2C and B2B model, creating consistent external capital inflows into the NRN economy. With ARC’s broader addressable market, this revenue stream will eclipse what AI Arena alone can generate.

Trainer Marketplace fees, while promising, depend on the ecosystem achieving critical mass—enough games, trainers, and players to sustain vibrant trading activity. It’s a long-term play.

In the near term, staking for ARC RL is likely the most immediate and reflexive demand driver. A well-funded initial rewards pool and the excitement of a new product launch could spark early adoption, driving up token prices and drawing in participants. This creates a feedback loop of rising demand and growth. However, the reverse is also possible—if ARC RL struggles to keep users engaged, demand could fade just as quickly.

The potential for network effects is enormous: more games → more players → more games join → even more players. This virtuous cycle could position NRN as a central token in the Crypto AI gaming ecosystem.

The Mother of Gaming AI Models

What’s the endgame? ARC’s strength lies in its ability to generalize across game genres. Over time, this enables them to aggregate a one-of-a-kind reservoir of gameplay-specific data. As ARC integrates with more games, it can continually feed this data back into its ecosystem, creating a virtuous cycle of growth and refinement.

Once this cross-sectional game dataset reaches critical mass, it will become an immensely valuable resource. Imagine leveraging it to train a generalizable AI model for game development—unlocking new possibilities for designing, testing, and optimizing games at scale.

It’s still early days but in the AI era where data is the new oil, the potential here is limitless.

Our thoughts

1. NRN Evolves Into a Platform Play — Token Repricing

With the launch of ARC and ARC RL, the project is no longer just a single-title game studio—it’s now positioning itself as a platform and AI play. This shift should lead to a re-rating of the NRN token, which was previously limited to AI Arena’s success. Introducing new token sinks through ARC RL, combined with external demand from revenue-sharing agreements with game studios and trainer transaction fees, creates a broader and more diversified foundation for NRN’s utility and value.

2. Success Tied Closely to Partner Game Studios

ARC’s business model ties its success to the studios it integrates with, as revenue flows are based on token allocations (in Web3 games) and royalty-based payments from the games. It is worth watching the games it integrates with closely.

If ARC-enabled games achieve mega-success, the resulting value will flow back to NRN holders. Conversely, if partner games struggle, value flows will be limited. This structure naturally aligns the incentives between ARC and the game studios.

3. Expecting more integrations with Web3 Games

The ARC platform is a natural fit for Web3 games, where incentivized competitive gameplay aligns perfectly with existing token-based economies.

By integrating ARC, Web3 games can immediately tap into the “AI Agent” narrative. ARC RL brings communities together and motivates them towards shared goals. It also opens up new opportunities for innovative mechanics, like making play-to-airdrop campaigns more engaging for players. By merging AI and token incentives, ARC adds layers of depth and excitement that traditional gaming can’t replicate.

4. AI gameplay has a learning curve

AI gameplay introduces a steep learning curve, which can create friction for new players. It took me an hour just to figure out how to properly train my player in AI Arena.

However, the player experience in ARC RL is lower friction, as the AI training is taken care of in the backend while players play games and submit their data. Another open question is how players will feel about competing against others, knowing that their opponent is an AI. Will it matter to them? Will it enhance or detract from the experience? Only time will tell.

A Bright Future

AI is set to unlock new groundbreaking experiences in the gaming world.

Teams like Parallel Colony and Virtuals are pushing boundaries with autonomous AI agents, while ARC carves out its niche by focusing on human behaviour cloning—offering an innovative approach to solving player liquidity challenges without unsustainable tokenomics.

The shift from a game to a full-fledged platform is a huge leap for ARC. This not only opens up bigger opportunities with game studios but also reimagines how AI integrates with games. 

With its revamped tokenomics and the potential for powerful network effects, ARC seems to be just getting started.

Thanks for reading,

Teng Yan

This research deep dive was sponsored by ARC, with Chain of Thought receiving funding for this initiative. All insights and analysis are our own. We uphold strict standards of objectivity in all our viewpoints.

To learn more about our approach to sponsored Deep Dives, please see our note here.

This report is intended solely for educational purposes and does not constitute financial advice. It is not an endorsement to buy or sell assets or make financial decisions. Always conduct your own research and exercise caution when making investment choices.