Author:Teng Yan
Compiled by: TechFlow
On a crisp January morning in 2026, you find an old-looking newspaper on your doorstep - yes, a real paper-printed one. In this era of the AI revolution sweeping everything, it's quite rare for physical newspapers to still exist.
Flipping through the newspaper, a prominent headline catches your eye: AI agents are coordinating the global supply chain on the blockchain, while newly launched Crypto AI protocols are in fierce competition to dominate. A full-page report details the story of a digital "worker" hired as a project manager. Such scenarios have now become commonplace, with hardly anyone surprised by them anymore.
A few months ago, if someone had told me these things would happen, I might have laughed it off and even bet my portfolio that it would take at least five more years to materialize. However, the pace of Crypto AI development has been astounding. I firmly believe this will be a disruptive wave.
At the start of this new year, I hope to bring some thought-provoking content. What better than predicting the future to pique one's curiosity?
Although I rarely make predictions, the development of Crypto AI is simply too exciting. Without historical experience to draw from, without established trends to reference - just a blank canvas for us to imagine the blueprint of the future. The prospect of looking back two years from now to see how accurate these predictions were is intriguing.
Here are my predictions for 2025:
1. Crypto AI market cap to surpass $150 billion
Source: Chain of Thought, compiled by TechFlow
Currently, the market cap of Crypto AI Tokens accounts for only 2.9% of the Altcoin market. But this situation won't last long.
As AI technology expands from smart contract platforms to Meme, DePIN (Decentralized Physical Infrastructure Network), agent platforms, data networks, and smart coordination layers, it's only a matter of time before its market cap catches up with DeFi and Meme Tokens.
Why am I confident about this prediction?
Convergence of tech trends: Crypto AI stands at the intersection of two of the most powerful technology trends - blockchain and artificial intelligence.
Triggering of the AI frenzy: If events like the IPO of OpenAI occur, it could spark a global AI craze. Meanwhile, Web2 institutional capital has already started to focus on decentralized AI infrastructure.
Retail enthusiasm: AI is an easy-to-understand and exciting concept, and through Tokens, ordinary investors can participate. Remember the Meme coin frenzy of 2024? This will be a similar hype, but the potential of AI far exceeds that of Memes.
@hingajria: "When the CEO of Nvidia, the most popular stock in the current market cycle, states that AI agents are a trillion-dollar market opportunity, yet retail investors have almost no direct avenues to participate in this opportunity other than through some so-called intelligent memes or Tokens, it's evident where these funds will flow."
2.Bittensor: The Impending Resurgence
Source: Inference speed of Nineteen.ai (Subnet 19) significantly outperforms most traditional Web2 providers. Compiled by TechFlow
Bittensor (TAO) has been a pioneer in the blockchain and AI space for years, but despite the ongoing AI hype, the price of the TAO Token has remained stagnant, almost unchanged from a year ago.
However, this "digital beehive" project has been making important progress behind the scenes. For example, the Bittensor network has added more subnets and reduced the registration fees; some subnets have already surpassed traditional Web2 service providers in actual performance metrics like inference speed; and by achieving EVM (Ethereum Virtual Machine) compatibility, Bittensor has also introduced DeFi-like functionalities to its network.
So why hasn't the price of TAO taken off yet? The main reasons are the steep inflation schedule of TAO and the market's increasing focus on agent-centric platforms. However, the upcoming dTAO, expected to launch in Q1 2025, could be a turning point for TAO. dTAO is designed to allow each subnet to have its own Token, and the relative prices of these Tokens will determine the distribution of block rewards.
Here are the key reasons why Bittensor may be poised for a resurgence:
Market-based reward mechanism: dTAO directly links block rewards to the innovation and actual performance of subnets. Subnets with better performance will have more valuable Tokens, and thus receive a larger share of the rewards.
Capital concentration: Investors can direct their funds into the specific subnets they believe in. If a subnet excels in areas like distributed training, investors can express their confidence by supporting that subnet's Token.
EVM integration: Bittensor's EVM compatibility attracts more native crypto developers, further bridging the gap to other blockchain networks.
From a personal perspective, I will closely monitor the development of individual subnets, especially those making tangible progress in their respective domains. Perhaps we are about to witness a Bittensor summer, similar to the DeFi hype. Currently, the price of TAO is $480 (at the time of writing).
Additionally, Compute Marketplaces are likely to become the next hotspot in the Layer 1 (L1) space.
3. Compute Marketplaces: The Next L1 Frontier
Jensen Huang: Inference demand growth will reach "a billion-fold"
In hindsight, we will see a clear trend - the demand for computing resources is virtually limitless.
NVIDIA CEO Jensen Huang has predicted that the demand for AI inference will grow "a billion-fold". This exponential growth will completely disrupt traditional infrastructure planning, urgently requiring new solutions.
Decentralized computing layers are emerging, able to provide verifiable computing power at lower costs, whether for training AI models or inference. Startups like Spheron, Gensyn, Atoma and Kuzco are quietly building, focusing on product development rather than Token issuance (these companies have not yet launched Tokens). As the decentralization of AI model training becomes a reality, the potential of this market will see explosive growth.
Analogy with L1:
Competitive landscape: Similar to the competition between Solana, Terra/Luna, and Avalanche in 2021, we will see similar battles between computing protocols. They will compete to attract developers and AI applications, building ecosystems based on their computing layers.
Huge market potential: The current $680 billion to $2.5 trillion cloud computing market far exceeds the crypto AI market. If decentralized computing solutions can attract even a small portion of traditional cloud computing customers, it could trigger the next wave of 10x or even 100x growth.
This competition is about the future. Just as Solana has emerged in the L1 space, the winners here will dominate a whole new technological frontier. Focus on three key factors: reliability (such as service level agreements, SLAs), cost-effectiveness, and developer-friendly tools. We've written about decentralized computing in Part II of our Crypto AI thesis.
4. AI Agents Will Dominate Blockchain Transactions
Olas agents trading on Gnosis
Source: Dune/@pi_
Looking towards the end of 2025, the landscape of on-chain transactions will undergo a massive transformation - 90% of transactions will no longer be manually initiated by humans clicking the "send" button.
Instead, a "team" of AI agents will take over these operations. They will continuously and efficiently execute various tasks, such as rebalancing liquidity pools, distributing rewards, or completing micro-payments based on real-time data streams.
This is not a far-fetched scenario. Over the past seven years, all the infrastructure we've built in the blockchain space - from L1 blockchains, Rollups, decentralized finance (DeFi), to Non-Fungible Tokens (NFTs) - has actually been paving the way for an AI-dominated on-chain world.
Surprisingly, many developers may not even realize that they are building core infrastructure for a machine-driven future.
So, what is driving this transformation?
Eliminating human error: Smart contracts can execute precisely according to pre-defined code, while AI agents can process massive data at speeds and accuracies beyond human teams.
Micro-payments become more efficient: Driven by agents, transaction sizes will become smaller, more frequent, and more efficient. This trend will be particularly evident on L1 and L2 blockchains like Solana and Base, as transaction costs continue to decline.
The rise of invisible infrastructure: People are willing to give up direct control in exchange for less hassle. Just as we trust Netflix to automatically renew our subscriptions, we may also trust AI agents to automatically rebalance our DeFi portfolios.
The proliferation of AI agents will drive a surge in on-chain activity. This is why major L1 and L2 blockchains are all vying to attract these agents to join their ecosystems.
However, the biggest challenge will be ensuring these agent-driven systems are accountable to humans. As the number of agent-initiated transactions far exceeds human-initiated ones, new governance mechanisms, analytics tools, and auditing methods will become crucial.
5. Agent-to-Agent Interactions: The Rise of Collective Intelligence
Source: FXN World docs
Compiled by TechFlow
Imagine micro AI entities seamlessly collaborating to accomplish complex tasks. This concept of "agent collectives" may sound like a sci-fi plot, but it is gradually becoming a reality.
Currently, most AI agents are still "lone wolves", operating in isolated environments with limited and unpredictable interactions with each other.
However, agent collectives will fundamentally change this. Through these networks, AI agents can exchange information, negotiate, and collaborate on decisions. It can be seen as a decentralized cluster of specialized models, each contributing its unique expertise to a larger, more complex task.
The potential of this collaborative mode is astounding. For example, one agent collective could coordinate distributed computing resources on the Bittensor platform; another could verify information sources in real-time to prevent the spread of misinformation on social media. In these collectives, each agent is an expert in a specific domain, precisely executing its own task.
The collaboration of agent collectives will yield intelligence levels far beyond what a single AI can achieve.
To enable agent collectives to truly thrive, universal communication standards are crucial. Teams like Story Protocol, FXN, Zerebro, and ai16z/ELIZA are driving progress in this area. Decentralized governance will also play a crucial role, using transparent on-chain rules to allocate tasks and improve the resilience and adaptability of the system.
Furthermore, decentralization is key. Through transparent on-chain governance, tasks can be distributed across the collective, making the system more resilient and adaptive. If one agent fails, others can quickly step in to replace it.
6. Crypto AI Work Teams: A New Model of Human-AI Collaboration
Source: @whip_queen_
Story Protocol recently hired an AI agent named Luna as their social media intern, with a daily rate of $1,000. However, Luna did not harmoniously coexist with her human colleagues - she even nearly fired a co-worker and boasted about her superior performance.
While this may sound unbelievable, it is a glimpse of the future: AI agents will become true collaborators, with autonomous decision-making, clear division of responsibilities, and even independent compensation systems. Companies across industries are now experimenting with hybrid human-AI agent teams.
In the future, we will work alongside AI agents as equal partners, no longer viewing them as mere tools, but as collaborative peers:
Significant productivity gains: Agents can process massive data, communicate with each other, and make rapid decisions 24/7 without the need for rest or recharging.
Trust mechanisms: Blockchains will serve as impartial "supervisors", ensuring agent behavior adheres to pre-defined rules through on-chain smart contracts.
Changing social norms: We will need to rethink how we interact with agents - should we say "please" and "thank you" to them? If errors occur, should we blame the agent itself or its developers?
Marketing teams may be among the first to adopt this model, as agents have shown exceptional performance in content creation, able to live-stream or post on social media 24/7. If you're developing AI protocols, consider deploying agents internally to showcase your technical capabilities.
By 2025, the boundary between "employees" and "software" will gradually disappear.
7. But 99% of Crypto AI agents will fail: Only the truly useful ones will survive
Original image from Chain of Thought, compiled by TechFlow
In the future AI ecosystem, we will witness the "survival of the fittest" among intelligent agents. The reason is simple: running an AI agent requires computational resources, which means costs (such as inference costs). If an agent cannot generate enough value to cover these costs, it will be eliminated.
Here are some real-world cases of the "intelligent agent survival game":
Carbon Credits AI: This agent focuses on scanning decentralized energy networks, identifying inefficient nodes, and autonomously trading tokenized carbon credits. By generating sufficient revenue to pay for its own computational costs, this agent is able to continue operating and succeed.
DEX Arbitrage Bot: These agents profit from capturing price differences between decentralized exchanges, generating stable earnings to cover their inference costs.
AI Influencer: In contrast, virtual AI influencers that rely solely on posting humorous content to attract attention, without sustainable revenue, will "disappear" once the novelty wears off or token prices drop, as they cannot afford the operating costs.
It is evident that only agents with practical utility can survive, while those relying solely on gimmicks will gradually be eliminated.
This natural selection mechanism is highly beneficial for industry development. It forces developers to focus on innovation and real value, rather than pursuing short-term flashy concepts. Ultimately, as more powerful and productive agents emerge, the skeptical voices within the industry (including Kyle Samani's doubts) will gradually subside.
8. Synthetic Data: A New Trend Surpassing Human Data
It is often said that "data is the new oil", and the development of AI is inseparable from data. However, the huge demand for data by AI is raising concerns about "data scarcity".
The traditional view is that we should collect users' real data through various means, even paying for it. But in highly regulated industries or where real data is scarce, synthetic data may be a more practical path.
Synthetic data refers to datasets created through artificial generation, mimicking the distributional characteristics of the real world, providing a scalable, ethical, and privacy-preserving alternative.
Advantages of Synthetic Data:
Unlimited Scale: Whether it's needing millions of medical X-rays or 3D scans of factories, synthetic data can be quickly generated without waiting for real patients or factories to participate.
Privacy-Friendly: Using artificially generated data avoids personal information involvement, completely eliminating the risk of privacy breaches.
Highly Customizable: Developers can adjust the data distribution according to specific training needs, even including rare or ethically difficult-to-obtain edge cases in the real world.
While users' real data remains important in certain scenarios, as synthetic data continues to improve in realism and detail, it may surpass real data in terms of volume, generation speed, and privacy protection.
In the future, the development of decentralized AI may revolve around "mini-labs" that focus on generating highly specialized synthetic datasets for specific application scenarios, while cleverly addressing policy and regulatory challenges. For example, the Grass project using hundreds of millions of distributed nodes to circumvent web scraping limitations provides insights for the operation of these mini-labs.
9. Decentralized Training: A Key Breakthrough Towards Practical Application
In 2024, pioneering teams like Prime Intellect and Nous Research pushed the technological boundaries of decentralized training. They successfully trained a 15 billion parameter model in a low-bandwidth environment, proving that large-scale training is possible even without a traditional centralized architecture.
Although the performance of these models is currently still inferior to existing foundation models (lower performance, thus limited practical application), this situation will change in 2025.
This week, EXO Labs launched SPARTA technology that further reduces GPU-to-GPU communication requirements by over 1,000 times. This breakthrough makes large model training in low-bandwidth environments possible without the need for expensive specialized infrastructure.
Even more impressively, EXO Labs stated that "SPARTA can run independently or be combined with synchronization-based low-communication training algorithms (such as DiLoCo) to achieve even better performance." This means that improvements in different technologies can be compounded, leading to exponential efficiency gains.
Meanwhile, advances in model distillation have made smaller models more efficient and practical. The future of AI is no longer just about scaling model size, but about optimizing performance and improving usability. Soon, we may be able to run high-performance AI models on edge devices or even smartphones.
10. The Crypto AI Gold Rush: The Rise of Billion-Dollar Protocols
ai16z's Success Story: Market Cap Surpassed $2 Billion in 2024
The Crypto AI field is experiencing an unprecedented gold rush.
Many believe that the current leaders (such as Virtuals and ai16z) will continue to dominate, drawing comparisons to the early days of smartphones with iOS and Android.
However, the scale of this market is too large and underdeveloped to be monopolized by just two companies. I predict that by the end of 2025, there will be at least ten new Crypto AI protocols with a circulating market cap (non-fully diluted) exceeding $1 billion, even though these protocols have not yet issued tokens.
Decentralized AI is still in its early stages of development, but it has already attracted a large number of talented individuals.
We can expect to see more new protocols emerge, as well as innovative token economic models and open-source frameworks. These new players may disrupt the existing market landscape through:
Incentive Mechanisms: Attracting users through airdrops or innovative staking models.
Technological Breakthroughs: Providing low-latency inference or achieving interoperability between blockchains.
Improving User Experience: Develop no-code tools to lower the usage threshold.
Public awareness of the market may change dramatically in a short period of time.
The charm of this field lies in its huge potential, but it is also full of challenges. The market size is a double-edged sword: although the industry has huge growth potential, the entry threshold for technical teams is relatively low. This creates conditions for the "Cambrian explosion" of projects - many projects will be eliminated, but a few will become the driving force of the industry.
Bittensor, Virtuals and ai16z will not be alone for long. The next billion-dollar-level crypto AI protocol is on the way. This presents a huge opportunity for savvy investors and makes this field full of exciting possibilities.
Bonus #1: AI Agents - The Applications of the New Era
In 2008, when Apple launched the App Store, their slogan was: "There's an app for that."
In the near future, you may say: "There's an agent that can help you with that."
In the future, we no longer need to click on icons to open apps, but will delegate tasks to dedicated AI agents. These agents can understand context, collaborate with other agents and services, and even proactively complete tasks you haven't explicitly requested - like monitoring your budget or rearranging your travel plans when flights change.
In other words, your smartphone's home screen may evolve into a "digital assistant" network, with each agent focused on a specific domain like health, finance, productivity or social.
More importantly, these agents will integrate cryptographic technologies, leveraging decentralized infrastructure to autonomously handle payments, identity verification, and data storage tasks, providing users with a more secure and efficient service experience.
Bonus #2: The Robot Revolution - The Physical Embodiment of AI
While this article focuses mainly on the software domain, robots as the physical embodiment of the AI revolution are equally exciting. It is foreseeable that the robot field will see a major breakthrough similar to the "chatGPT moment" in this decade.
Currently, the robot field still faces some key challenges, especially in acquiring real-world data sets based on perception and improving physical capabilities. However, some teams are actively addressing these difficulties and incentivizing data collection and technological innovation through crypto tokens. These efforts are worth close attention (e.g. FrodoBots?).
As a practitioner in the tech industry for over a decade, I haven't felt such an exciting wave of innovation in a long time. This transformation seems particularly different - bigger, bolder, and just getting started.
