Pantera: Crypto’s role in the AI ​​revolution

Source: Pantera Capital October Blockchain Letter; Compiled by: 0xjs@黄金财经

Crypto: A Weapon for the AI ​​Gold Rush

Author: Matt Stephenson, Pantera Capital Research Partner; Ally Zach, Pantera Capital Research Engineer

"AI is infinitely abundant, while Crypto is absolutely scarce."  

Sam Altman's observation in 2021 has since become a mantra for enthusiasts of both technologies. At first glance, abundance seems to be more influential than enforced scarcity, suggesting that AI may be a more prudent investment. In fact, Nvidia's market capitalization is larger than the entire cryptocurrency.

But Altman's remarks are reminiscent of Adam Smith's "diamond and water paradox." Smith pointed out that while water is essential for survival, its abundance makes it almost worthless.

In contrast, diamonds have little practical use but are valuable because they are scarce. This paradox suggests that even if AI becomes as important as water, its market value may still be limited. In contrast, the scarcity of cryptocurrencies is more strategically important and valuable than it might initially appear.  

Large language models (LLMs) have achieved remarkable feats, including passing the Turing test and reportedly outperforming humans on standard IQ tests. But this raises the question: if humans can’t tell the difference between humans and intelligent AI (on a Turing test), can they tell the difference between intelligent AIs? If humans can’t tell the difference, then future improvements in AI performance may yield diminishing returns in terms of perceived benefits to consumers.  

Just as the jump from 4K to 8K TV resolution is a marginally noticeable improvement to the average viewer, the difference between a high-performance AI model and a slightly more advanced model may be imperceptible to most users. This could lead to the commoditization of much of the AI ​​market, where state-of-the-art models will be used only for specialized applications in research, industry, or government, while more cost-effective “good enough” models will become the standard for everyday use. Top AI models may become “expensive boutiques that mainstream consumers will never consider upgrading to.”

So even as we speculate on the potential growth of AI, we should also consider the alternative: the powerful capabilities currently known in AI already exist and will become increasingly commoditized. This is where the intersection of crypto and AI (“Crypto x AI”) really comes into focus. Crypto’s potential may not be a high-beta bet on AI meme value, but a practical value capture mechanism for AI’s distributed future. Once everyone has a 4k TV in their home, its value will be what we do with them.  

By serving as an important and reliable input to AI and a rail for distributed AI coordination and transactions, cryptocurrencies are closer to conservative “shovel and pickaxe” bets on AI. This may surprise investors who view Crypto x AI primarily as a volatile proxy for potential growth in AI. But interestingly, over the past six months, using Nvidia as a proxy for AI growth sentiment, cryptocurrencies look more like hedges against AI growth sentiment than high beta investments.

We will first assess the bright prospects for “AI agents” and how crypto will play a role. Then, we will discuss the potential for crypto to support the current inputs to AI: data, compute, and models.

AI Agents: Programs with Programmable Money

By Matt Stephenson, Research Partner at Pantera Capital

Last year, before most people were talking about AI agents on blockchain, I co-authored a paper that was accepted to NeurIPS, the top AI conference in the US. Since then, I’ve had the pleasure of speaking at crypto and agent AI events at universities like Stanford, Columbia, Cornell, and Berkeley, in addition to attending many technical and investment conferences. Next week, I’ll be speaking on AI with an Oxford professor, the president of the IEEE, and a member of the GBBC, all in an effort to better understand, explore, and communicate what the future of agent AI is and how it intersects with blockchain. Of course, I’m also invested in this future, including investments in agent infrastructure like Sentient and other undisclosed positions.

The future is here. While OpenAI says AI agents won’t be ready until 2025, in the cryptocurrency space, we already have AI agents trading and exploring on the blockchain space today. An AI agent that has promoted its own token (note: Truth Terminal) currently has about $300,000, and by the time you read this, it may become the first AI agent millionaire.  

But what are these agents? How do they differ from the "robots" we are more familiar with?

Agents are more than robots

Defining an "agent" is more nuanced than it seems. The AI ​​field has a less practical definition of an agent: "anything that senses its environment through sensors and acts on that environment through actuators." Economists' view of an agent is closer to what we want: "an agent is someone who acts on your behalf in a specific decision domain."

If an agent is acting on your behalf, then the robot is essentially a difficult agent to communicate with. First, you have to write code for the robot to execute, which means communicating in a (programming) language that most people don't understand. And for those who know language, they still have to program what the robot should do under a variety of different conditions, which means specifying those conditions in advance. Both of these are communication costs.  

As an analogy, suppose you have a friend who is going abroad and you ask him to buy you a souvenir. If your friend is like a robot, he will ask you to write a program to specify what souvenir they should buy you. What if your friend is like an agent? Then you can make requests in language, and you can trust that your friend will buy you what you want. Using language, without having to specify preferences for gifts you might receive abroad, reduces communication costs. Clearly, this is a better agent.  

Having to know the conditions in advance (because you have to program them) limits the usefulness of the robot as an agent. Then, the mere fact that the robot has to be programmed means that it is out of reach for those who don't program it. We model the move to AI agents as a reduction in these communication costs and the corresponding release of economic value.

Despite the high communication costs of existing bots, over $2 trillion in monthly cryptocurrency stablecoin trading appears to be bot-based. As bots become better agents, perhaps able to trade USDC and USDT based on relative risk like you, we should expect this number to increase.

AI Agents Will Use Crypto

One ​​reason AI agents could benefit crypto is that it helps alleviate the user experience issues that crypto is notorious for. The complexity of blockchain interactions, wallet management, and decentralized finance protocols has long been a barrier to widespread adoption. AI agents can act as intuitive interfaces, translating user intent into the precise technical actions required on the blockchain. They can guide users through complex transactions, explain risks, and even suggest the best strategy based on market conditions and user preferences.

Another reason is that agents cannot have bank accounts, but can trade with wallets. This limitation of the traditional financial system fits perfectly with the spirit of cryptocurrency. In the crypto world, agents do not need permission from a central authority to operate. They can interact directly with smart contracts and decentralized protocols to hold and manage digital assets on behalf of users. This opens up new possibilities for automated wealth management, 24/7 trading, and personalized financial services that operate entirely within the crypto ecosystem.

Finally, a mature agent ecosystem means that agents need to trade and coordinate with each other. Modern smart contracts, as programmable, always-online international legal systems, are well suited for this task. AI agents can leverage crypto infrastructure to engage in complex, multi-party transactions and agreements. They can negotiate terms, execute trades, and even resolve disputes within the parameters set by human principals. This creates a new paradigm of autonomous economic activity, where agents can form ad hoc coalitions, pool resources, and collaborate to accomplish tasks that humans cannot or should not manage directly.

We believe each of these activities will add value to crypto infrastructure. But there are also indirect effects, making crypto itself better. For example, decentralized autonomous organizations (DAOs) have been inactive due to attention constraints in crypto. A DAO actively managed by a network of AI agents, each representing the interests of the DAO’s voters, would be a game changer. These agents could analyze proposals, allocate resources, and execute strategies at a speed and scale beyond human capabilities, while adhering to the core principles and goals of their human creators.

AI agents and cryptocurrencies are more than a perfect match, they are two technologies that need each other. Agents need programmable money to operate autonomously in the digital economy. Cryptocurrencies need AI to improve user experience and deliver on their promise of bringing a financial revolution to everyone. As this synergy develops, we may see core blockchain infrastructure such as Solana, Ethereum, Near, and Arbitrum become the main beneficiaries of this new agent-driven economy. They are poised to do this by facilitating agent transactions, hosting decentralized applications that agents interact with, and providing the secure, transparent environment needed for inter-agent coordination. As agent activity increases, these networks are likely to see increased transaction volume, greater demand for their native tokens, and stronger network effects. This isn’t just about technical compatibility — it’s about creating a new economic paradigm in which AI and crypto work together to make finance more efficient, accessible, and maybe even a little sci-fi.  

Crypto Powers Current AI

By Ally Zach, Research Engineer at Pantera Capital

Imagine you’re on the verge of a major breakthrough, only to find the tools you need are just out of reach. Innovation often feels like that — a journey filled with breakthrough highs and challenging lows. Take the automotive industry, where the quest for more efficient engines once hit a dead end. Engineers were eager to push the limits, but the materials needed didn’t exist yet. Progress stalled until new alloys and composites reignited the innovation engine. Similarly, new technologies like crypto could unlock AI’s untapped potential.

For years, AI has progressed incrementally, first slowly and then rapidly, similar to an S-curve. In 2017, we achieved a key breakthrough that gave rise to Transformer-based architectures, as outlined in the influential paper Attention Is All You Need. These Transformers revolutionized sequential data processing in models, enabling efficient training of large datasets. This sparked the rapid development of powerful new LLMs and generative AI models.

Despite progress in AI development, significant bottlenecks in data, compute, and model generation must be overcome to achieve the next leap. Combining AI with blockchain technology can help decentralize resources and democratize access, making innovation open to global contributors.

Data

Data is the lifeblood of AI, the fuel that drives its accuracy and reliability. High-quality, representative data is essential for building effective models, but acquiring this data is challenging due to privacy concerns, limited access, and inherent biases. Additionally, users are increasingly reluctant to share personal information, making data collection resource-intensive and often hampered by trust issues.

Blockchain technology offers a promising solution by introducing a decentralized, secure, and transparent approach to data aggregation. Platforms like Sahara fit into our long-term strategy of advancing decentralized infrastructure for AI by enabling individuals to contribute and monetize their data while retaining control. Additionally, token economics incentivize high-quality contributions by rewarding users accordingly. This approach helps address privacy concerns by giving users ownership and control over their own data. It democratizes access to data, enabling smaller businesses that previously lacked the resources to compete with large tech companies. By incentivizing data sharing in a secure manner, blockchain-based platforms turn data into a commodity, enriching the available data pool and potentially producing more robust and unbiased AI models.

However, despite its innovative nature, blockchain-based data aggregation is not a standalone solution for AI development. If used alone, practical challenges such as scalability, data quality assurance, and integration complexity limit its effectiveness. With their massive datasets and mature infrastructure, large tech companies still have significant advantages that decentralized platforms have difficulty matching.

Therefore, solutions, including blockchain-based ones, introduce new avenues for data collection and collaboration that complement rather than replace traditional approaches. Synergies between decentralized efforts and established technology leaders can lead to partnerships that leverage the strengths of both parties and promote innovation and inclusivity in AI development.

Compute

The rising cost and scarcity of GPUs present significant barriers to smaller players in AI development. Due to high demand and supply chain issues, GPU prices have continued to rise since the outbreak of the pandemic, and large companies have increasingly monopolized access to basic hardware. This limits innovation, as many startups and researchers need help affording tools for advanced model training. This reduces the diversity of AI research and slows progress at smaller institutions.

Crypto, however, has the potential to level the playing field by commoditizing computing power. Platforms such as Exo and io.net are democratizing GPU access through decentralized marketplaces where anyone can access or lend computing resources. Individuals with spare computing power can offer it on the network and earn rewards for it. The commoditization of high-performance computing enables a wider range of innovators to participate in AI development, breaking down barriers that once restricted access to advanced tools.

In the future, as the supply of GPUs increases, decentralized computing markets may compete directly with traditional cloud services. These platforms lower the barrier to access and provide cost-effective alternatives that enable broader participation in the AI ​​ecosystem. However, ensuring that users have access to reliable computing power remains a challenge. Validating GPU standards and maintaining consistent, secure resources are critical to building trust and preventing fraud. While decentralized solutions may not replace traditional services, they can provide a competitive alternative where flexibility and cost are more important than guaranteed performance.

Models

Today, AI development is often concentrated in a small number of organizations such as OpenAI, Google, and Facebook. This concentration limits opportunities for global innovators and raises concerns about whether AI reflects diverse human values. Centralized control can lead to models that reflect narrow viewpoints and ignore the needs and perspectives of a wider user community.

A shift is underway to distribute the power of AI development through decentralized platforms. In line with our vision that AI will increasingly run on crypto rails, platforms like Sentient and Near are democratizing development by creating open-source, community-driven ecosystems. Using blockchain technology, they transparently manage contributions, ensuring developers are recognized and compensated through token rewards. This allows anyone to build, collaborate, own, and monetize AI products, ushering in a new era of AI entrepreneurship. Illia Polosukhin, co-author of the seminal paper “Attention is All You Need” and co-founder of Near, is working to foster an open environment for developing general artificial intelligence (AGI) through crowdsourcing. Collaborative initiatives like this one aim to align AI development with broad human values.

These platforms act as catalysts for change, driving an AI economy that is both competitive and collaborative. By broadening participation, they encourage diverse ideas to flourish, leading to more innovative solutions and potentially reducing bias in AI models.

Crypto x AI offers a unique opportunity to democratize AI development, but it also presents significant challenges. Balancing the need for large-scale collaboration with high-quality, expert-driven work is critical to ensuring models are robust and ethical. By decentralizing access to data, compute power, and model development, crypto breaks down traditional barriers and enables talent from around the world to participate in the advancement of AI. This influx of diverse perspectives fosters collaboration and builds a more inclusive ecosystem. Embracing this collaborative model will not only accelerate innovation, but ensure that the global community shapes the future of AI.