Wall Street has spent two years pricing artificial intelligence through a single lens: compute. Who has the most chips, the most data center capacity, the most cloud infrastructure? That framework has been enormously rewarding for investors who recognized it early.
A different infrastructure gap is now opening up, and most investors have not started pricing it yet.
Why AI agents require a different kind of infrastructure than AI models
The AI systems that have dominated public attention since late 2022 are fundamentally information tools. They answer questions, summarize documents, and generate content.
The user still has to take action. That changes with agentic AI, software that does not simply advise but executes tasks autonomously on behalf of users.
The difference sounds incremental. It is not. When software can act rather than just advise, every action carries legal, financial, and reputational consequences for the person who authorized it.
Related: Sam Altman makes stunning admission about AI
That introduces infrastructure requirements that conversational AI never needed: identity systems, permission frameworks, payment rails, and audit mechanisms that allow humans to review and reverse what happened.
Mau Ledford, CEO and co-founder of Sogni AI, describes the shift plainly: “The key shift is that AI is moving from ‘help me think’ to ‘help me finish,'” Ledford said. “A chatbot suggests a trip. An agent books it.”
The first real-world deployments are already appearing. Travala launched what it describes as the first end-to-end agentic AI travel protocol on June 10, enabling autonomous agents to search, reserve, and settle payments across more than 2.2 million hotel properties, including Marriott, Hilton, and IHG.
The system processes transactions at approximately $0.01 per booking with near-instant settlement, according to The Block. Final payment authorization still requires user approval, but everything before that step runs autonomously.
Juan Otero, CEO of Travala, said travel is a natural starting point because the booking process remains one of the most fragmented consumer experiences in existence.
“Agentic AI can do all this for you in one chat,” Otero said. “No more tabs, no more forms, no more friction.” he added.
More AI:
- Micron sits at the center of a red-hot chip rally
- IBM CEO sends blunt message on AI and quantum computing
- Anthropic CEO makes shocking admission about AI
Travel is only one of the early domains. Experts point to business procurement, customer support, financial management, research, and creative production as sectors likely to follow. Gartner projects that 40% of enterprise applications will embed task-specific AI agents by end of 2026, up from under 5% in 2025, according to Svitla.
Creative production is one area already showing traction: rather than generating a description of an ad, an agent uploads a product photo, produces a storyboard, renders video, localizes copy, and prepares distribution assets, compressing a multi-step workflow into a single instruction. As Sogni AI’s CEO said, “The winning products will not be the ones that make agents the most autonomous. They will be the ones that make autonomy feel controlled.”
Trust and identity are the hardest problems agentic AI infrastructure must solve
As AI agents move from answering questions to taking actions, the infrastructure gap that matters most is not compute. It is accountability. Booking the wrong flight or executing an unintended transaction carries consequences that conversational AI never had to reckon with.
Every autonomous action needs a clear owner, a clear scope, and a clear audit trail.
The good news is that the problem is not entirely new. Enterprises already use identity and access management systems defining what each employee can access and what requires approval.
Consumers are already familiar with the concept: a parent setting a spending limit on a child’s credit card is the same delegation logic agentic systems will need at scale. The engineering challenge is extending those proven frameworks to autonomous software.
Sydney Huang, CEO of Human API, believes that the identity problem is fundamentally one of adaptation rather than invention. “The challenge is not inventing an entirely new security model,” Huang said, “but adapting existing identity, permissioning, and governance frameworks to autonomous software agents in a way that remains transparent and easy for people to understand.”
The emerging architecture looks like layered controls: a verified identity for the agent, clear delegation from the user, defined spending limits, an audit log of every action, and a mechanism for humans to revoke access at any point. She said the long-term solution combines cryptographic identity, verifiable credentials, and programmable permission systems, so trust is enforced at the infrastructure level rather than relying on any single platform’s policies.
Travala’s system already implements several of these principles using session keys that ensure payment requests originate from the agent but final signing authority remains with the user’s wallet, and a machine-verifiable performance layer that lets agents build reputation from verified completed bookings over time, according to The Block.
But the catch-22 of agentic access remains an unsolved problem. Ran Hammer, VP of Business Development at Orbs, frames it plainly: “To be useful, it needs real access to your stuff, but that much access is dangerous.”
The same capability that makes an agent valuable, the ability to act across multiple systems on a user’s behalf, is the same capability that makes a compromise or a hallucination consequential in a way that a wrong answer never is.
Nitat/Getty Images
Why traditional payment infrastructure is not built for autonomous software
Identity solves one half of the agentic commerce equation. Payments solve the other. An agent that can plan and confirm a transaction but cannot execute the payment is only halfway to useful, and traditional banking infrastructure was designed for humans, not for software.
Opening a bank account, completing identity verification, and executing transactions have historically required human intervention. Autonomous software cannot fulfill those requirements the same way.
That friction is pushing agentic commerce toward programmable payment infrastructure that can settle instantly and enforce spending rules automatically. As Travala’s CEO put it: “Banking is a bottleneck for agentic AI. Traditional banking rails are fundamentally incompatible with autonomous software.”
This is where digital assets become functionally relevant, not as a speculative asset class, but as payment infrastructure aligned naturally with how autonomous software operates. Programmable settlement runs around the clock without a banking intermediary. Spending rules can be encoded directly into the transaction.
The alignment is structural. “Agents and crypto fit naturally because blockchain is part of the internet layer,” Hammer said. “An agent paying in stablecoins doesn’t care about banking hours or middlemen.” The practical evidence: x402 agentic transactions on Base grew from near-zero in mid-2025 to over 100 million cumulative transactions through Q1 2026, according to BitcoinKE.
Ben Goertzel, CEO of SingularityNET, argues the financial infrastructure for agents must be built as open, collectively governed infrastructure. “Crypto is the native financial substrate for AI agents,” Goertzel said. “Programmable money, permissionless access, and verifiable state are exactly what autonomous economic actors need.” he added.
Key context on the agentic AI infrastructure buildout:
- Travala’s Travel MCP protocol is live through Claude Desktop and open to external developers; the company is offering a 10% rebate in Coinbase Wrapped Bitcoin for developers building agents that complete hotel bookings through its system, a direct financial incentive to accelerate third-party ecosystem development, according to The Block.
- The long-term solution is a combination of cryptographic identity, verifiable credentials, and programmable permission systems that allow trust to be enforced at the infrastructure level rather than relying solely on platform policies; that distinction matters because platform-level trust can be revoked or gated, while infrastructure-level trust is available to any agent meeting the technical standard.
- His OmegaClaw project at SingularityNET is building agents that do hypothesis-driven financial risk analysis, including decomposing unsustainable yield rates and warning leveraged traders of liquidation cascades below their entry price; it is one of the first deployments of agentic AI specifically designed for autonomous financial risk management rather than general assistance.
- The identity challenge extends to API access as a separate bottleneck; many services simply block agents from accessing them, meaning that even a capable and trustworthy agent can be locked out at the service level regardless of how well the identity and permission architecture is designed, a friction point he expects to ease over time as adoption grows.
- The Travala launch uses ERC-7715 session keys to ensure payment requests originate from the agent while final signing authority remains isolated to the user’s secure wallet environment; ERC-8004 adds a machine-verifiable trust layer that tracks agent performance based on verified real-world outcomes, creating a reputation system for autonomous software that mirrors how human track records are evaluated, according to The Block.
What the agentic AI infrastructure gap means for investors
The history of major technology transitions shows that the first phase rewards the companies building the capability and the second rewards those building the infrastructure to deploy it at scale. During the internet era, enormous value accrued not to the websites but to the payment systems, identity frameworks, and cloud services that websites ran on.
Agentic AI appears to be establishing a similar structure. The models exist and the use cases are materializing. IDC projects that 40% of roles in Global 2000 companies will involve direct engagement with AI agents by end of 2026, according to Gartner and IDC research.
What is not yet built at scale is the identity layer, the permission framework, and the payment infrastructure that allows agents to transact safely on behalf of hundreds of millions of users. Those gaps represent the next concentrated infrastructure investment opportunity.
The companies most likely to capture durable value are those solving the identity and permissioning problem with proven enterprise architecture. This is an adaptation challenge, not an invention challenge.
Organizations extending familiar IAM frameworks to autonomous software will reach institutional adoption faster than those requiring the market to accept new trust models. The infrastructure race to capture that value is just beginning.
Related: AI shopping apps could change how Americans find deals
