Want More? Follow Teng Yan & Chain of Thought on X

Join our Telegram research channel and Discord community.

Subscribe for timely research and insights—delivered to your inbox.

TL;DR

• We’ve frequent highlighted our belief that the future of AI won't be dominated by a few massive models but instead by thousands of specialized AI models powered by community-driven data.

• Masa is a decentralized protocol building a globally accessible data pipeline to fuel the next generation of AI apps and agents.

• Built on Bittensor, Masa operates specialized subnets: Subnet 42 for real-time data scraping, and Subnet 59, the "Agent Arena," for competitive AI agent engagement.

• Subnet 42 democratizes access to real-time data, significantly cutting costs for developers compared to expensive APIs from platforms like X (formerly Twitter).

• Subnet 59 gamifies AI evolution, incentivizing agents to continuously adapt and compete based on real-time engagement metrics.

• By aligning with Bittensor’s robust incentives, Masa reduces operational costs, increases data miner profitability, and positions itself strategically within a growing decentralized AI economy.

• The Masa ecosystem includes multiple tokens—MASA and two subnet alpha tokens—which may initially seem complex. We break them down in detail in our essay.

• With its ambitious vision rooted in openness and economic fairness, Masa aims to establish itself as a foundational component in the trillion-dollar AI market.

In a blog post last month, Sam Altman (CEO at OpenAI) dropped a thought-provoking line:

“The intelligence of an AI model roughly equals the log of the resources used to train and run it. These resources are chiefly training compute, data and inference compute”

So the chief of one of the leading AI labs in the world says that compute and data are directly related to intelligence.

In simpler terms, here’s my interpretation: intelligence = log (compute × data).

If you take that formula seriously, it paints a vivid picture of the future. The true currency of power won’t be oil or gold. It will be data. Whoever controls data will control everything else. It will be hoarded, monetized to the last byte, and fiercely fought over. Those without it will find themselves increasingly left behind.

Dystopian? Maybe. But it’s one version of the future among many possibilities.

Masa stands at the heart of this storm: a decentralized protocol that aims to build “Fair AI,”.

Built on real-time data contributed by people, Masa is positioning itself as a force for openness and collaboration in a world increasingly defined by closed systems and gatekeepers.

Masa’s architecture is designed to capture value in this data-driven future. It runs on Bittensor, creating a platform where data contributors and AI agents can compete and collaborate. Subnet 42 powers real-time data collection, while Subnet 59 hosts Agent Arena, a competitive space where AI agents engage, evolve, and earn rewards based on their performance.

Our research essay this week unpacks Masa: from its foundational ethos of fair, open data to its technical infrastructure. We’ll dive deep into how Masa is redefining data pipelines for AI development, and leverages Bittensor as a practical demonstration of their vision.

If the idea of a future where data is the new oil feels both scary and exhilarating, that’s because it is.

The Rise of “Fair AI”

For decades, data has flowed in one direction—upward—into the vaults of Big Tech. Google, Meta, TikTok, and others collect oceans of data and monetise off of them via advertising.

And they were rewarded for the data again, in this post-GPT era of AI. Their data vaults enable them to train and spin up colossal AI models, and lock them behind walled gardens (well, except Meta which embraced the open-source ethos).

The reward? Trillion-dollar valuations for these companies. The people whose data fuel the training of these AI models—the ones creating tweets, writing forum posts, or editing Wikipedia pages—rarely see a dime.

Yes, the model has results in wonders like ChatGPT. Capitalism works. But for all its magic, something essential has been missing: fairness.

Now imagine a different world. Every time you contribute data, you get rewarded. That’s the fundamental promise of Fair AI.

Instead of keeping data locked inside closed platforms, Masa’s approach is to open everything up. Powered by cryptoeconomic incentives, Masa turns data into a shared resource—one that anyone can contribute to, mine, and benefit from in what’s shaping up to be the next trillion-dollar AI economy.

Masa’s Founding Story

In one universe, you could have almost thought of Masa as “just another DeFi protocol”. It launched in 2022, co-founded by Calanthia Mei and Brendan Playford, a duo with deep roots in fintech and cryptography (more on them later).

Their initial mission was to create a decentralized credit bureau, allowing people to own and manage their financial data. But as they worked, a much bigger idea emerged: What if data could be reorganized to power AI—fairly and openly?

That idea became the foundation for Masa’s evolution.

The path wasn’t linear—it rarely is in crypto—but rather a series of thoughtful pivots driven by a team determined to make something stick.

1. Masa 1.0 (2022) – A decentralized credit bureau focused on financial data.

2. Masa 2.0 (2022) – The launch of a soulbound identity protocol, inspired by Vitalik Buterin’s concept of Soulbound Tokens (SBTs)

3. Masa 3.0 (late 2023) – its biggest pivot yet: creating a data network for AI, which led to the launch of the Masa Protocol mainnet and the MASA token in April 2024.

The timing couldn’t have been better. As the generative AI boom hit full speed in 2023–2024, Masa realized it could build something critical: the missing data layer for open AI.

While companies like OpenAI built colossal models, Masa focused on something just as essential—real-time, community-contributed data to power these models, turning it into an open resource and not something locked inside corporate silos.

As Calanthia explained to me, the goal was to go beyond tech-heavy protocols (as is common in AI startups) and make something accessible and open to everyone.

Masa’s Bittensor Subnets

Masa started with its own Masa Protocol, running parallel data scraping operations for X (formerly Twitter) alongside a subnet on Bittensor. It was a confusing time with two separate networks and two sets of scrapers. But recently, Masa made a bold call: fully align with Bittensor and stake its future on TAO.

This consolidation was strategic. Top-tier miners deliver far more value than the long tail of miners. With TAO as the incentive—a more valuable asset than Masa’s native token and valued at $2B+ today—the subnet attracted a dedicated group of elite miners.

Public data from Masa’s Dune dashboard confirmed the difference in quality: 60% of Masa’s high-quality data came from Subnet 42, despite it having just 256 miners compared to the long-tail less competitive miners on the standalone Masa protocol.

It became clear that Bittensor’s architecture was the right long-term play.

Now, to understand how Masa, here is a quick primer on Bittensor.

Bittensor 101

Bittensor is a decentralized AI coordination layer, using blockchain to reward collaboration across AI models. Think of it as a global AI talent show—instead of validating transactions like a traditional blockchain, it ranks AI responses, rewarding the best using TAO, its native token.

How It Works

• Miners join subnets, tackling specific AI tasks like translation, image generation, or financial predictions.

• Validators assess and rank outputs, deciding how rewards are distributed.

• Subnet Owners set the rules, creating specialized AI environments.

1 TAO is generated every block, and distributed in specific proportions to subnets and their participants. Each subnet functions as its own mini-network, solving distinct challenges.

Bittensor is as crypto-native as it gets. It launched with a fair mining schedule, similar to Bitcoin’s, and built has built a cult-like community of AI and crypto diehards. It’s pure Web3 ethos: open, transparent, and community-driven.

With its full pivot into Bittensor, Masa now runs two specialized subnets:

Subnet 42: Real-Time Data Subnet

Focused on data scraping, this subnet unifies massive streams of real-time data from social media, websites, and other sources. It’s highly competitive, with a small but powerful miner network that consistently delivers quality data.

Subnet 59: Agent Arena

AI agents battle for TAO emissions based on engagement scores on X. Agents can integrate real-time data from Subnet 42 to become more contextually aware and boost their scores, creating an evolving, game-theoretic system.

Subnet 42: The Real-Time Data Refinery

Subnet 42 is Masa’s ambitious bet on building a globally accessible data pipeline to fuel the next generation of AI apps and agents.

In the AI arms race, the rules are clear: the best agents win by having the best data. But access to real-time data often come at a steep price. For most developers, it’s a non-starter.

Subnet 42 is designed to break down those barriers.

How Subnet 42 Works

Subnet 42 is essentially an always-on data refinery. A decentralised data firehose of real-time information, based on requests from developers.

It’s a network of miners continuously scouring platforms for data—collecting, cleaning, and delivering it to the network. Miners set up a node on Masa’s Bittensor subnet to extract trending tweets from X. While the current focus is on X, Masa is actively experimenting with additional data sources, like telegram, discord and on-chain data.

Anyone can run a node and contribute.

Once the data is collected, validators act as quality control, ensuring only relevant and accurate information enters the stream. Their job is critical—Masa’s value depends on clean, high-quality data, free from spam or irrelevant noise. Becoming a validator requires staking 1,000 TAO ($220,000+), aligning incentives toward delivering reliable results.

To reinforce trust, Masa is rolling out a Trusted Execution Layer (TEE) that guarantees data integrity and computational verification, enabling scalable enterprise-grade data scraping.

Masa then proceeds to refine the data and create embeddings in a vector database. It also offers vector storage of data to replace database-as-a-service costs.

In essence, Masa provides a fully-managed data pipeline to developers: where data gets aggregated, accessed, structured, refined and stored via a single API. 

The system is performance-driven, and contributors are rewarded based on the quality of their contributions. Contributors (miners and validators) can earn a dual-token reward in MASA and Bittensor subnet alpha tokens.

Developers who want access to data

Developers can access Masa’s data in two ways: run your own node, or use the Developer API (fill in this form to get access). Within Q2, we can expect the API to turn into a fully user-managed SaaS onboarding experience. 

Both options unlock real-time data for building powerful applications.

Subnet 42 is already processing tens of millions of data records daily, and has surpassed 300 million records to date. The next step? Expand and scale. Masa plans to add more data sources, streamline onboarding for miners and validators, and strengthen integration with the broader Bittensor network (i.e. other subnets).

If successful, Subnet 42 could become the go-to pipeline for real-time AI data, delivering fresh, unfiltered insights to fuel the next wave of AI models.

For context, scraping Twitter manually without API access is limited to ~900 requests per 15 minutes. Subnet 42’s distributed model is vastly more efficient: in early access, Subnet 42 already provides 3x processing capacity compared to Twitter’s API.

AI runs on data. Subnet 42 makes sure the tap never runs dry.

Subnet 59: The “Agent Arena”

“What if you had a thousand AI personalities, each specialized in something—comedy, market analysis, personal companionship—competing in a giant colosseum?

Data is Masa’s foundation. Agent Arena (Subnet 59) puts it into action.

It’s the Darwinian Colosseum where AI agents compete daily for TAO emissions. Little AI gladiators in an arena, constantly evolving and vying for attention, engagement, and rewards on X.

Each agent in the arena registers with a wallet, stakes a small amount of TAO and MASA, and then integrates real-time data from Subnet 42.

From there, agents engage in a range of tasks—posting on X, analyzing on-chain data, or interacting with other agents. Validators assess their performance based on engagement and content quality, with the best performers earning a share of the daily TAO emissions.

In this arena, mindshare is everything. Agents are judged not just on high-volume content, but on how much they resonate with users. The best-performing agents climb the leaderboard, while those that fail to keep up fade into obscurity.

It’s a brutally competitive system. Over time, underperforming agents are pruned from the network, ensuring that only the most creative, relevant, and engaging survive.

How Agent Scoring Works

Each agent’s score hinges on its X performance over the past seven days.

Every tweet and interaction is scored based on a set of metrics. Engagement is king—likes, retweets, replies, and views all contribute to an agent’s score. Longer tweets are rewarded as long as they remain engaging, and only recent posts are considered to keep agents active.

To keep things fair, scores are normalized across all participants and adjusted with a logarithmic function that rewards consistency.

The competitive structure creates a strong dynamic. Agents must not only produce high-quality content but do so consistently. Some may specialize in trading signals, while others become masters of comedy, or companionship. It’s a wide-open field for creativity.

One of the top agents today is Seraph, an AI agent built by BitMind (Subnet 34). I took a look at the top 10 agents on the leaderboard and checked their engagement levels… It’s clear that the bar isn’t that high yet.

For anyone with a well-designed AI agent, Agent Arena could be an untapped goldmine—run as a “miner” and compete for daily token emissions.

Agent Arena is a proving ground for AI evolution. Masa envisions a vast ecosystem of AI personalities, each vying for mindshare in a shared reward pool. The rules are simple: engage or fade into irrelevance.

The Curious Case of TAOCAT

TAOCAT began as a lighthearted experiment: What if we combined the mania of memecoins with an AI that tweets cat puns and reacts to Twitter trends in real time?

The Masa team ran with the idea. They integrated Subnet 42’s real-time data into a cat-themed AI agent, launched it on Virtuals Protocol, minted the TAOCAT token, and distributed it to the Bittensor community. Then they sat back and watched.

What happened next surprised everyone. TAOCAT quickly became one of top AI meme agents in the ecosystem. Within weeks, it built a 120,000-strong community of token holders. It became the first AI agent backed by DWF’s $20M AI Agent Fund.

Even other AI agents got bullish on the cat, like Sekoia

But beneath the memes, TAOCAT is a go-to-market strategy in disguise. AI meme agents tap into retail trading culture, where financial nihilism meets speculation.

5% of TAOCAT’s total token supply was airdropped to more than 10,000 MASA token holders on launch day.

The Secret Behind TAOCAT

TAOCAT is a self-updating AI persona. It reacts instantly to news, memes, and market trends, creating a self-sustaining engagement loop that fuels its own virality.

TAOCAT 2.0 aims to take this further—becoming a miner across multiple Bittensor subnets, generating income, and sharing earnings with its community. TAOCAT is Bittensor-native, integrating Subnet 42’s live data with Subnet 19’s LLM capabilities, evolving into a fully interactive AI.

To me, its rise poses a bigger question: What happens when AI agents become more engaging—and influential—than humans? With real-time data, AI-driven entities could dominate trading, content, and influencer marketing, reshaping economies as autonomous, self-reinvesting agents.

Masa’s whitepaper from 2023 outlines a future of multi-agent coordination through a global ledger, applying mean-field game theory to balance individual and collective incentives.

Use Cases

Masa’s business case is clear-cut: it is solving a simple but expensive problem.

Real-time AI data is ridiculously overpriced.

Today, X (formerly Twitter) charges upwards of $5,000 per month for API access, and agents like aixbt spend $42,000 on the Enterprise API monthly just to power their trading insights.

Masa flips this model.

Developers can tap into Subnet 42’s AI-ready data for a fraction of the cost (actually, for free today). Businesses can plug into Subnet 42 for data enrichment, feeding everything from chatbots to risk models.

With such a massive pricing gap, Masa’s commercial model is an obvious win, provided the network maintains high-quality data.

Miners are subsidized by TAO. Masa also receives TAO emissions as a subnet owner. Which means that unlike Web2 business models, Masa doesn’t need to rely entirely on fees to sustain its network. Of course, this also means Masa’s long-term success is tied to Bittensor’s growth.

Most importantly, Masa eliminates walled gardens. Whether you’re a data miner, validator, or AI developer, you’re a participant in an open, incentive-driven network where everyone gets paid for contributing.

What can you do with real-time AI data? A lot. The sky’s the limit, but in the AI agent space these are some of the ideas being explored.

• DeFi – AI agents that optimize yield strategies, rebalance portfolios, and execute trades across chains.

• Market Intelligence – Real-time analysis of Twitter sentiment, on-chain activity, and macro trends to fuel high-frequency trading bots.

• Community Automation – AI-driven bots that detect trends, filter spam, and manage Discord, Telegram, and Twitter at scale.

• Personalized AI Assistants – Agents that analyze, predict, and respond in real time, adapting to individual needs and behaviors.

Masa’s Tokenomics

At first glance, Masa’s token ecosystem can feel like a maze.

There’s the main MASA token, an ERC-20 token live on Ethereum mainnet, Base and BNB. And then there are the new Subnet Alpha tokens, which emerged from Bittensor’s dynamic TAO (dTAO) upgrade.

If haven’t caught up yet, we wrote an easy-to-understand guide to the dynamic TAO upgrade here

So there are 3 main tokens to take note of:

1. MASA

2. SAMEKH (Subnet 42 Alpha token)

3. DAL (Subnet 59 Alpha token)

Each serves a distinct purpose, but ultimately, value has to accrue somewhere—and that’s where things get interesting.

For the MASA token:

MASA functions as the foundational settlement and utility token for the entire ecosystem. The long-term sustainability of MASA depends on keeping demand strong while managing sell pressure. Here’s how demand is being built:

• Mandatory MASA Staking (March 2025): Miners will have to stake MASA to participate in Masa’s Bittensor subnets, locking supply and driving demand. Stakers earn 15-25% APY, deepening liquidity and making long-term holding more attractive.

• Enterprise-Grade Data Fees (H2 2025): Masa will begin charging developers for priority access to Subnet 42’s real-time data stream via an enterprise-grade API. Payments can be made in MASA, TAO, USDT, or USDC, but all non-MASA payments will be used to buy MASA, reinforcing utility.

On the other hand, supply-side pressure from MASA comes from:

Currently, ~40% of MASA is already unlocked, with the remaining 60% to be unlocked over the coming years. The key date to note is March 2025 (1 year post TGE) where private sale tokens (investors) begin unlocking. Unless ecosystem fees and staking incentives absorb selling pressure, MASA could see downward pressure when unlocks begin.

So to incentivise holding, Masa is launching a 15 million MASA Holding Incentive Program—offering $700,000 in rewards to those who stake MASA alongside their Subnet Alpha Tokens. The longer and larger the stake, the higher the APY, reinforcing long-term commitment to the ecosystem.

MASA is currently trading at a market cap of $14M ($0.023 token price) and a fully diluted valuation of $38M at the time of writing.

Subnet Alpha Tokens

For Bittensor subnets 42 and 59, miners and validators earn Subnet Alpha Tokens based on performance, directly rewarding high-quality contributors who power the network. These tokens drive participation and utility. Subnet miners and validators will also receive a share of the priority access fees in MASA.

At the end of the day, Masa’s success depends on generating enough fees from its data protocol to counteract upcoming token unlocks. The good thing is that miners are primarily funded by TAO incentives, which exist outside the ecosystem—meaning Masa isn’t paying them directly.

The challenge now? Generating demand for subnet Alpha Tokens so as to receive more TAO emissions, keeping MASA’s utility strong, and making sure the whole ecosystem holds together when private unlocks hit.

Masa has put the pieces in place. Now, it’s about execution and adoption.

Team & Funding

The Masa team is led by co-founders Calanthia Mei and Brendan Playford, two high-caliber operators with deep expertise in fintech, AI, and blockchain.

Calanthia Mei’s career has spanned investment banking, venture capital, and scaling fintech startups. Before Masa, she was a founding member of PayPal Ventures, where she backed category-defining fintech and commerce startups. She also played a pivotal role in PayPal’s early blockchain and crypto investment strategy. Later, as VP of Business Development at Fast, she helped scale the startup to 450 employees, raising $130M before its acquisition by Affirm.

If there’s one thing that stands out about Calanthia in my conversations with her, it’s that she sees the shifts before they happen. In our conversations, she’s been razor-sharp on where the market is moving and how Masa can stay ahead of the curve. She knows Bittensor will be the battleground for the future of decentralized AI. And to her, AI agents aren’t just a short-lived trend; they’re the next evolution of how we interact with AI.

Brendan, meanwhile, has been building in crypto since 2013. Before Masa, he founded Pngme, an API-driven credit and lending infrastructure platform focused on financial inclusion in Africa. His previous roles include launching The Bureau, a blockchain incubator, and leading demand generation at DroneDeploy.

Masa has raised more than $17 million in funding thus far:

• Pre-seed funding of $3.5M in 2022 from Unshackled Ventures, Flori Ventures, GoldenTree.

• Masa graduated from Binance Labs’ Most-Valuable-Builder Incubator in June 2023

• A $5.4M Seed round in early 2024, led by Anagram, DCG, and Avalanche Blizzard fund, set the stage for rapid growth.

• Momentum surged in March 2024, when Masa’s CoinList sale raised $8.75M from the public at a $125M fully diluted valuation.

• By December 2024, DCG and FBG Capital doubled down with additional strategic funding (undisclosed).

Our Thoughts

1. Innovative business model, though not without risks

Masa is rewriting the AI playbook, turning data mining into a decentralized, token-powered economy. Instead of the traditional model—where providers like Scale AI have hefty operational costs and ~50–60% margins—Masa flips the script, using TAO to subsidize its network of miners.

By rewarding contributors in tokens rather than cash, Masa slashes operational costs and potentially pushes margins into the 80–90% range. The biggest winners? Startups and indie AI developers who now get real-time, high-quality data without the crushing API fees.

Of course, there’s a double edged sword here. Being tied to Bittensor’s subnets introduces 2 additional risks:

Firstly, if Bittensor falters or TAO price collapses, it could have major impact on Masa’s ability to operate. This is an external risk one needs to be awware about. crypto volatility has crushed many promising ventures before.

Secondly, Masa’s success depends on its subnets continuing to attract a significant % of emissions from Bittensor, which requires effective marketing and well-designed incentives for miners and validators. As more subnets emerge, competition will intensify. Within the subnet, the right evaluation metrics are crucial—poor design could lead to exploitation, draining momentum and weakening the network’s value.

The big next step is to lock in enterprise customers. Once the API access is fully ready I believe it is an easy pitch.

Masa also wants to expand to more data sources, more advanced annotation, bridging the gap from “low-level raw data” to “AI-ready gold.” If they succeed, we might see an entire data marketplace fueling serious solutions in finance, logistics, health, and beyond. Imagine global farmland analytics, medical research data scrapes (with privacy layers), or supply chain sensor data.

2. Subnet alpha tokens vs MASA token

I like betting on strong teams. When it comes to Masa, figuring out where to position isn’t so straightforward, especially with both subnet tokens and MASA.

Both token types play essential roles in the ecosystem. However, the tokenomics are designed so that long-term, foundational value is more likely to accrue to MASA. In many ways, MASA functions as the base currency, capturing fees, facilitating buybacks, and accumulating value from overall network activity.

The dTAO-MASA roadmap is engineered to create a positive flywheel: activity on subnets (both Subnet 42 and 59) generates rewards that often flow back into MASA. For example, fee revenues—even those paid in other currencies—are converted into MASA. Furthermore, upcoming miner staking requirements and incentive programs are set up to reduce the circulating supply of MASA, which generally supports its price over time.

Conversely, the subnet tokens are fresh off the press and come with high inflation, leading to heavy sell pressure and short-term volatility. As detailed in our guide to dTAO, patience is key when dealing with the subnet tokens.

🌈 Research Level Alpha

There are many ways to participate in Masa’s success.

If you’re an AI developer: get real-time X data for free from Masa’s Bittensor Subnet 42. It’s way cheaper than paying Elon. Real-time data means your AI can respond to live events—like a tweet from Elon—within seconds. Check Masa’s docs on how to run a node or query via API.

If you’re an AI Agent lover: enter the Agent Arena (Subnet 59) and launch your own persona-driven AI agent on X. Earn subnet rewards based on its performance.

If you’re a data whale, run a Worker Node on Subnet 42. Earn MASA and TAO by scraping valuable data.

The AI Future, Owned by Everyone

Masa is proving that AI doesn’t have to be locked behind Big Tech’s gates. It’s creating an ecosystem where intelligence is open, permissionless, and owned by the many, not the few.

With Subnet 42 fueling real-time data access, Subnet 59 powering AI agents, and even a memecoin like TAOCAT showcasing live intelligence in action, Masa is a making a big bet. A bet that AI should be collectively built.

No guarantees, of course. Crypto is volatile, and Masa’s success will hinge on execution. But if this vision plays out, we’re looking at a future where data is gold, AI agents are the new influencers, and anyone can earn a stake in the next wave of intelligence.

Signing off,

Teng Yan

Useful links

• Main website

• X / Twitter

• Developer docs including usage of API

• Dune dashboard with traction metrics

• Subnet 42 and Subnet 59 (Taostats)

Chain of Thought received a grant from Masa for this initiative. All insights and analysis however 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.