Author: Sandy Kaul (Head of Digital Assets and Innovation at Franklin Templeton)
Compiled by: Jiahuan, ChainCatcher
AI is evolving and dominating the investment narrative
AI has been continuously evolving. Around the 2010s, early capabilities such as machine learning, natural language processing, and predictive analytics ignited the "big data" era, enabling people to process structured and unstructured data at speeds and scales previously unimaginable. At that time, AI was more like a tool assisting human work.
By the early 2020s, generative AI emerged, significantly elevating its applications and role. AI transitioned from a helper to a "co-creator," capable of generating content, responding to diverse questions, and even performing parts of human tasks. The potential of generative AI has still not been fully realized—products continue to grow stronger and are increasingly integrating into more aspects of daily life.
As its influence grows, AI's status as a leading investment theme is now undeniable.
On July 14, 2026, IBM's stock price plunged 25.2% in a single day. Prior to this, it issued a warning that corporate technology budgets are increasingly being allocated to AI infrastructure, while spending on traditional software and IT projects is being delayed or cut.
Today, the S&P 500 has reached its highest concentration levels since the late 1990s tech bubble. The top 10 companies by market capitalization are all AI-themed stocks, collectively accounting for nearly 40% of the index’s total market value. In comparison, this figure was only 25% during the dot-com bubble and just 15% in 1980.
Institutional investors, in particular, view AI as a structural mega-trend, heavily investing in AI infrastructure, data centers, and semiconductor stocks.
But this layout may still not be enough to capture the next wave of AI evolution.
The Rise of AI Agents
Generative AI represented a leap forward compared to early AI tools in terms of capabilities, effectiveness, and application scenarios. Today, as agent-based AI matures and becomes widely adopted, its impact on daily life may be equal to or even greater than that of the previous generation.
The agent AI transforms the interaction model from a passive, response-based chatbot into an autonomous system capable of perceiving its environment, formulating its own plans, and executing multi-step tasks to achieve high-level goals without continuous human oversight.
Agents can interact directly with external software systems and code repositories, thereby redefining the role of AI. Thirty-eight percent of organizations say that by 2028, agents will become team members alongside human colleagues, jointly enhancing productivity and driving innovation.
Following this trend, the tasks delegated to AI will become increasingly complex. Generative AI excels at gathering knowledge and organizing content, while agents will take on more "transactional" tasks—initiating, tracking, completing, and managing outcomes on their own.
Some forecasts suggest that by 2030, the agent economy could reach a scale of $3 trillion to $5 trillion.
For institutions, these types of transactions typically occur within enterprise software. Predictions indicate that by 2028, 33% of enterprise software will integrate agent AI, with up to 15% of daily decisions handled by these agents.
Software applications will pay small fees to each other for computing power, API calls, data usage, and various services. This is a new way of interaction, enabling precise accounting and settlement for each individual task.
Protocols enabling "software to pay software" are emerging.
Protocols supporting this "machine-to-machine" trading are gradually emerging. Stripe and Visa have already launched Machine Payment Protocols (MPP).
Open-source solutions are also gaining momentum. In the early 1990s, the designers of the World Wide Web specifically reserved response code "402," labeled as "payment required," for communication rules between browsers and servers.
Coinbase developed an "x402" protocol based on this, enabling agents to initiate and complete such payment instructions, and subsequently transferred the related intellectual property to the Linux Foundation to make it an open industry standard.
Today, major credit card networks, Web2 giants such as Stripe, Shopify, Google, and Amazon Web Services (AWS), as well as an increasing number of Web3 service providers, have integrated this payment standard. The goal is to enable "software-to-software payments" without any human intervention.
Over the coming years, Agent payments are likely to reshape how consumers interact. Projections suggest that by 2030, Agents could account for 15% to 25% of U.S. e-commerce sales.
Currently, ChatGPT processes 2.5 billion queries per day, including 53 million shopping-related inquiries initiated through AI platforms. OpenAI is also integrating checkout functionality into third-party ChatGPT applications, such as Target, DoorDash, and Instacart.
How blockchain enables transactions between machines
Supporting these "machine-to-machine" transactions requires a secure, autonomous, verifiable, and high-throughput ledger system.
Traditional credit card and banking systems are not suited for the microtransactions required by agents. A standard credit card transaction incurs an average fee of 2% to 3%, plus a fixed fee of about $0.30; meanwhile, an agent typically spends only $0.001 to purchase one second of computing power or perform one data query.
For AI agents to operate effectively, they will likely rely on cryptography and blockchain, as this underlying infrastructure is inherently well-suited for such scenarios. In fact, due to the following characteristics, blockchain and cryptographic technologies are poised to serve as the foundational layer for these transactions.
Automatic generation and execution of contracts. Payment agents generate tokens to complete purchases and settlements. Each token contains a set of predefined transaction rules: which merchants can accept the token, the maximum amount that can be spent per transaction, and the token’s expiration period. Once a purchase is completed, these one-time tokens are automatically invalidated. The blockchain can hold, send, and receive these tokens, strictly enforcing the rules embedded within them, just like executing a smart contract.
Decentralized identity verification. Each Agent has a unique, cryptographically verifiable identity. Every token it generates carries its own credential, which is required to sign blockchain transactions. When validating transactions, the blockchain verifies these credentials; if an identity is deemed invalid, the consensus mechanism blocks the transaction.
Fully auditable. Every decision, transaction, and data exchange made by the Agent on-chain is recorded on an immutable ledger, accessible to anyone via a blockchain explorer, ensuring accountability and transparency.
Connect to decentralized computing power and data. Through blockchain, Agents can access distributed computing resources (such as GPU networks) and data, reducing reliance on centralized, private cloud infrastructure and helping to lower the operational costs of high-frequency trading models.
Speed and settlement. Bitcoin can process only about 7 transactions per second (TPS), Ethereum about 75 TPS, but newer high-speed blockchains have significantly increased these peaks: Aptos reaches up to 12,933 TPS, Solana at 6,284 TPS, and BNB Chain at 3,252 TPS.
This speed is already comparable to Visa’s network, which processes 1,700 to 10,000 transactions per second under normal conditions. But even this comparison underestimates on-chain systems: while blockchain both records and settles transactions within that TPS window, Visa only records the transaction—the actual settlement still takes 1 to 3 business days.
With these features, blockchain will play a key role in enabling agent AI for consumer-level transactions. Conversely, the growth of agent AI is likely to become the "killer app" that drives blockchain adoption.
How to invest in the agent AI opportunity
Currently, investors seeking to benefit from AI growth typically buy stocks of AI-themed companies and related supply chain firms, become limited partners in private equity funds, or invest in energy providers and data centers that support AI operations.
But to seize the opportunity presented by agent AI, these portfolios may need to extend their exposure to native tokens of public blockchains, as well as project tokens issued by on-chain applications and initiatives. Several key forces are driving this shift, as outlined below.
Demand for cryptocurrencies will rise. To record a transaction on a particular blockchain, an Agent must pay transaction fees in that chain’s native token—for example, SOL to record a transaction on Solana. As Agent payments increase, demand for the native tokens of the blockchains supporting these operations may surge, creating value for every token holder. Initially, this demand will likely come from machine-to-machine micropayments between enterprise software systems.
The blockchain ecosystem will expand. The more transactions on a chain and the higher the demand for its native token, the more funds flow into the chain’s treasury. The blockchain foundation uses these treasury funds to grant developers funding to build applications on the chain, offer "bug bounties" to those who discover security vulnerabilities, and incentivize individuals who validate transactions for the network. The more funds available for allocation, the more likely the ecosystem is to grow and become more secure, attracting even more developers to build applications, issue their own tokens to fund projects, and share ownership of their applications.
Web3 applications will take market share from Web2. As more applications move on-chain and more developers enter the space, the advantages of Web3 applications over Web2 will become increasingly clear. This is already happening in Web3 gaming: the industry is shifting from Web2’s “single-player” model to Web3’s “player-owned economy.”
Tap-to-earn apps have attracted hundreds of millions of users worldwide. Players can now trade and buy/sell in-game items (NFTs) on secondary markets and across platforms, truly owning and monetizing the assets they accumulate in games. Similar experiences and shifts in ownership may unfold across a wide range of consumer applications, which will also drive market interest in the tokens issued by these projects.
The flywheel effect will kick in. The protocols for embedding Agent payments into blockchain applications already exist, and they could bring a flywheel effect to these newly issued tokens. Users simply need to instruct their Agent to handle transactions and payments, without having to set up wallets, buy cryptocurrencies, or manage various tokens and digital assets themselves.
For users, the experience of using a Web3 application looks no different from using a Web2 product. But because the tokens here have both utility and represent ownership, they can gain more benefits within the Web3 ecosystem.
In some ways, this shift will resemble the transition from Web1 to Web2: moving from Web1’s web servers and static websites to Web2’s cloud-based services and interactive applications. In both transitions, the protagonists—whether the underlying technology providers or the businesses built on top of them—shifted from established incumbents to a new generation of players driving growth in the new era. This time, the torch is being passed to blockchain and various decentralized applications and projects.
Currently, investors haven’t fully figured out how to capture the value created by blockchain and its ecosystem. They are accustomed to a centralized, corporation-driven business world: to share in the value created by a company, they simply buy its stock.
But I believe it will become increasingly clear over the coming years that: to capture the value of decentralized networks and businesses, investors need to buy related crypto assets. These assets are likely to become important holdings in portfolios, especially for those looking to capitalize on new opportunities in agent AI.