Author: Teng Yan.
Compiled by: Luffy, Foresight News.
On a crisp morning in January 2026, you find an old newspaper at your doorstep, yes, printed on real paper, somehow having survived the AI revolution.
As you flip through the newspaper, you see a headline about AI agents coordinating global supply chains on the blockchain, while newly launched crypto AI protocols are vying for dominance. A half-page report introduces a digital 'employee' hired as a project manager: this has now become commonplace, and no one is surprised.
If you had asked me a few months ago, I would have scoffed at this idea, even betting my portfolio that such progress would take at least five more years to materialize. But this is the rapid pace at which crypto AI will disrupt the world. I am thoroughly convinced of this.
After recovering from a severe stomach flu, I sat at my desk to start the new year, wanting to focus on something valuable: something that sparks curiosity and might even provoke some debate. What could be better than trying to glimpse the future?
I usually don’t make predictions rashly, but crypto AI is so captivating that it’s hard to resist. With no historical precedent or trends to reference, it feels like a blank canvas where one can imagine what might happen next. Honestly, the thought of looking back at this article in 2026 to see how wrong I was makes it all the more interesting.
So here are my thoughts on how crypto AI might develop by 2025.
1. The total market cap of crypto AI will reach $150 billion.
Currently, crypto AI tokens account for only 2.9% of the market cap of altcoins. But this situation won't last long.
As AI encompasses everything from smart contract platforms to Memecoins, decentralized physical infrastructure networks (DePIN), and new primitives like agent platforms, data networks, and smart coordination layers, it is inevitable that it will stand shoulder to shoulder with DeFi and Memecoins.
Why am I so certain?
Crypto AI is at the intersection of the two most powerful technological trends I have ever seen.
AI frenzy trigger event: An OpenAI IPO or a similar event could trigger an AI frenzy worldwide. Meanwhile, Web2 institutional capital is already eyeing decentralized AI infrastructure as an investment target.
Retail frenzy: The concept of AI is easy to understand and exciting, and now they can invest through tokens. Remember the 2024 Memecoin gold rush? The same frenzy will happen with AI, but the difference is it is actually changing the world.
2. The revival of Bittensor.
Nineteen.ai (subnet 19) outperforms most Web2 providers in inference speed.
Bittensor (TAO) has been around for years, regarded as a veteran in the crypto AI space. Despite the surge in AI hype, its token price has remained sluggish, flat compared to a year ago.
In fact, this 'digital hive mind' has already quietly made leaps: more subnets with lower registration fees are emerging, some subnets have already surpassed Web2 counterparts in practical metrics like inference speed, and are compatible with the Ethereum Virtual Machine (EVM), bringing DeFi-like functionalities to the Bittensor network.
So, why hasn't TAO skyrocketed? A steep inflation plan and the shift in attention towards AI agents have hindered its growth. However, dTAO (expected to launch in Q1 2025) could be a significant turning point. With dTAO, each subnet will have its own tokens, and their relative prices will determine how TAO is distributed.
Reasons for Bittensor's revival:
Market-based token releases: dTAO will directly tie block rewards to innovation and actual measurable performance. The better the subnet, the more valuable its tokens, thus the more TAO released.
Focus on capital flows: Investors can finally allocate funds to specific subnets they believe in. If a specific subnet adopts innovative methods for distributed training and achieves significant results, investors can direct funds to it to express their support.
EVM integration: Compatibility with EVM will attract a broader community of native crypto developers to Bittensor, narrowing the gap with other networks.
Personally, I am watching various subnets, keeping an eye on those making real progress in their respective fields. At some point, we will have a DeFi summer version of Bittensor.
3. The computation market will become the next battleground similar to L1.
Jensen Huang: Inference demand will 'grow a billionfold.'
An obvious major trend is the massive demand for computation.
NVIDIA CEO Jensen Huang has a famous saying that inference demand will 'grow a billionfold.' This is an exponentially growing demand that will disrupt traditional infrastructure planning, urgently requiring 'new solutions.'
Decentralized computing layers provide raw computational power (for training and inference) in a verifiable and cost-effective manner. Startups like Spheron, Gensyn, Atoma, and Kuzco are quietly laying a solid foundation to leverage this trend, focusing more on products than on tokens (these companies currently do not have tokens). As decentralized training of AI models becomes feasible, the potential market size will skyrocket.
Compared to L1 blockchains:
Just like in 2021: Do you remember the competition among Solana, Terra/Luna, and Avalanche to be the 'best' L1 blockchain? We will see a similar brawl emerge between computation protocols, competing to attract developers to build AI applications on their computation layers.
Web2 Demand: The cloud computing market, valued between $680 billion and $2.5 trillion, is much larger than the crypto AI market. If these decentralized computing solutions can attract even a small portion of traditional cloud customers, you will see the next wave of 10x or even 100x growth.
The stakes are high. Just as Solana stood out in the L1 blockchain space, the winners in the computation market will dominate a whole new frontier. Watch for three key factors: reliability, cost-effectiveness, and developer-friendly tools.
4. AI agents will flood blockchain transactions.
Olas agent's transaction on Gnosis. Source: Dune/@pi_
Fast forward to the end of 2025, and 90% of on-chain transactions will no longer be triggered by humans clicking the 'send' button.
On the contrary, they will be executed by a multitude of AI agents that tirelessly rebalance liquidity pools, allocate rewards, or execute micropayments based on real-time data sources.
This doesn't sound far-fetched. Everything we have built over the past seven years - L1 blockchains, scaling solutions, DeFi, NFTs - has quietly paved the way for a world dominated by AI-led on-chain activities.
Ironically? Many developers may not even realize they are building infrastructure for a machine-led future.
What is causing this shift?
Avoiding human error: Smart contracts execute strictly according to the code. In turn, AI agents can process vast amounts of data more quickly and accurately than human teams.
Micropayments: These agent-driven transactions will become smaller, more frequent, and more efficient, especially as transaction costs decrease on Solana, Base, and other L1/L2 blockchains.
Invisible infrastructure: If it means less hassle, humans will gladly relinquish direct control. We trust Netflix to automatically renew subscriptions; thus, trusting AI agents to automatically rebalance our DeFi positions seems like the logical next step.
AI agents will generate astonishing activity on-chain. No wonder all L1/L2 blockchains are courting them.
The biggest challenge will be making these agent-driven systems accountable to humans. As the ratio of agent-initiated transactions to human-initiated transactions rises, new governance mechanisms, analytical platforms, and auditing tools will be needed.
5. Interactions between agents: The rise of swarms.
Source: FXN World documentation.
The concept of agent swarms (tiny AI entities working seamlessly together to execute grand plans) sounds like the plot of the next blockbuster sci-fi/horror movie.
Today's AI agents mostly operate in isolation, with minimal and unpredictable interactions.
Agent swarms will change this situation, enabling AI agent networks to exchange information, negotiate, and collaborate to make decisions. It can be imagined as a decentralized ensemble of specialized models, each contributing unique expertise to larger, more complex tasks.
One swarm may coordinate distributed computing resources on platforms like Bittensor, while another can address misinformation by verifying information sources in real time before content spreads on social media. Each agent in the swarm is an expert, executing its tasks with precision.
These swarm networks will generate intelligence far more powerful than any single isolated AI.
To allow swarms to thrive, universal communication standards are crucial. Agents need the ability to discover, authenticate, and collaborate, regardless of the underlying framework they are based on. Teams like Story Protocol, FXN, Zerebro, and ai16z/ELIZA are laying the groundwork for the emergence of agent swarms.
This leads to the critical role of decentralization. Distributing tasks among swarms according to transparent on-chain rules will make the system more resilient and adaptable. If one agent fails, others will immediately fill the gap.
6. Crypto AI work teams will be hybrids of humans and machines.
Source: @whip_queen_
Story Protocol hired Luna (an AI agent) as their social media intern, paying her $1,000 a day. Luna did not get along well with her human colleagues: she almost got one colleague fired while boasting about her outstanding performance.
Although it sounds strange, this is a harbinger of the future. In the future, AI agents will become true work partners with autonomy, responsibility, and even salaries. Companies across various industries are exploring hybrid work teams of humans and machines.
We will collaborate with AI agents, not treating them as slaves, but as equal partners:
Surge in productivity: Agents can process vast amounts of data, communicate with each other, and make decisions around the clock.
Establishing trust through smart contracts: Blockchain serves as a fair and impartial overseer that does not tire or forget. On-chain ledgers ensure that important agent actions adhere to specific boundary conditions/rules.
Evolution of social norms: We will soon have to deal with etiquette issues regarding interactions with agents. Should we say 'please' and 'thank you' to AI? If they make mistakes, should we hold them morally accountable, or blame their developers?
I expect marketing teams to be the first to adopt this model, as agents are adept at generating content and can live-stream and post on social media around the clock. If you are building an AI protocol, why not deploy agents internally to showcase your capabilities?
In 2025, the line between 'employees' and 'software' will begin to blur.
7. 99% of crypto AI agents will perish, with only practical agents surviving.
We will witness a Darwinian natural selection among AI agents. Why? Because running an AI agent incurs costs in the form of computational power (i.e., inference costs). If an agent cannot generate enough value to pay its 'rent,' it will be eliminated.
Example of agent survival games:
Carbon credit AI: Imagine an agent searching within a decentralized energy grid, identifying inefficiencies, and autonomously trading tokenized carbon credits. It can earn enough money to cover its computational costs. Such agents can thrive.
Decentralized exchange arbitrage bots: Agents exploiting price differences between decentralized exchanges can continuously generate income to cover their inference costs.
Spam publishers on X: Meanwhile, what about that virtual AI influencer who only tells cute jokes and has no sustainable income source? Once the novelty wears off and token prices crash, it will vanish, unable to sustain operations.
The distinction is clear: utility-oriented agents thrive, while others gradually fall by the wayside.
This natural selection is beneficial for the field. Developers are forced to innovate and prioritize productive use cases over flashy gimmicks. As these stronger, more effective agents emerge, they will leave skeptics speechless.
8. Synthetic data will surpass human data.
People often say 'data is the new oil.' AI relies on data, but its immense demand for data has raised concerns about an impending data shortage.
The traditional view is that we should find ways to collect private real-world data from users, even paying them for it. But I am gradually coming to agree that a more practical approach in heavily regulated industries or where real data is scarce is through synthetic data.
These are artificially generated datasets designed to simulate real-world data distributions, providing a scalable, ethical, and privacy-preserving alternative to human data.
The powerful reasons for synthetic data:
Unlimited scale: Need a million medical X-rays or 3D scans of a factory? Synthetic generation can produce them in unlimited quantities without waiting for real patients or real factories.
Privacy protection: When using synthetic datasets, no personal information is at risk.
Customizable: You can tailor the data distribution to exact training needs, injecting extreme cases that may be too rare in reality or too morally challenging to collect.
Indeed, in many cases, the human data owned by users is still important, but if synthetic data continues to improve in authenticity, it may surpass user data in quantity, generation speed, and freedom from privacy constraints.
The next wave of decentralized AI may revolve around 'small labs' that create highly customized synthetic datasets for specific use cases.
These small labs will cleverly sidestep policy and regulatory hurdles in the data generation process, just as Grass uses millions of distributed nodes to bypass web scraping restrictions.
I will elaborate on this in future articles.
9. Decentralized training will truly take effect.
In 2024, pioneers like Prime Intellect and Nous Research broke through the boundaries of decentralized training. We have already trained a model with 15 billion parameters in low-bandwidth environments, proving that large-scale training outside traditional centralized methods is feasible.
Although these models are currently less practical (lower performance) compared to existing foundational models, I believe this will change by 2025.
This week, EXO Labs took a step further with SPARTA, reducing communication between GPUs by over 1000 times. SPARTA enables large model training without dedicated infrastructure under low bandwidth conditions.
What impressed me most was their statement: 'SPARTA can work on its own but can also be combined with synchronous low-communication training algorithms like DiLoCo for better performance.'
This means these improvements can compound to further enhance efficiency.
As advances in technologies like model distillation make smaller models practical and efficient, the future of AI lies not in size but in better performance and accessibility. Soon, we will have high-performance models capable of running on edge devices or even smartphones.
10. Ten new crypto AI protocols will have a market cap of $1 billion (not yet launched).
In 2024, ai16z skyrocketed to $2 billion.
Welcome to the real gold rush. It’s easy to think that today’s leaders will continue to dominate, and many compare Virtuals and ai16z to the early days of smartphones (iOS and Android).
But this market is too large and undeveloped to be dominated by just two companies. By the end of 2025, I predict at least 10 new crypto AI protocols that have not yet launched will have a circulating (non-fully diluted) market cap exceeding $1 billion.
Decentralized AI is still in its infancy, and a large pool of talent is gathering.
We have every reason to expect the emergence of new protocols, new token models, and new open-source frameworks. These new entrants can replace the existing ones by combining incentives (like airdrops or staking), technological breakthroughs (like low-latency inference or chain interoperability), and user experience improvements (no-code). A shift in public perception could happen in an instant.
This is both the allure and the challenge of this field. The market size is a double-edged sword: the cake is large, but the barrier to entry is low for skilled teams. This sets the stage for a Cambrian explosion of projects, many of which will gradually disappear, but a few will become transformative forces.
The dominance of Bittensor, Virtuals, and ai16z won’t last long. The next batch of crypto AI protocols with a market cap of $1 billion is on the horizon. Savvy investors will have plenty of opportunities, which is what makes this so exciting.
Additional Highlight #1: AI agents are the new applications.
When Apple launched the App Store in 2008, their slogan was 'There’s always an app for you.'
Soon, you will say: 'There is always an agent that suits you.'
You will no longer click icons to open applications, but delegate tasks to specialized AI agents. These agents possess contextual awareness, can cross-communicate with other agents and services, and even initiate tasks you never explicitly requested, such as monitoring your budget or rearranging your travel itinerary if your flight changes.
Simply put, your smartphone's main screen might turn into a 'digital colleague' network, with each 'colleague' having its own area of work: health, finance, productivity, and social.
And because these are crypto-powered agents, they can autonomously handle payments, authentication, or data storage using decentralized infrastructure.
Additional Highlight #2: Robotics
While much of this article focuses on software, I am also very excited about the physical manifestation of the AI revolution - robotics. In this decade, robotics will have its ChatGPT moment.
This field still faces significant obstacles, especially in acquiring perception-based real-world datasets and enhancing physical capabilities. Some teams are rising to the challenge, using crypto tokens to incentivize data collection and innovation. These efforts are worth watching (like FrodoBots?).
Having been in the tech field for over a decade, I can't remember the last time I felt such genuine excitement. This wave of innovation feels different: grander, bolder, and just beginning.
