AI Agent Payments: The Flow Gap Between Reported Growth and Real Transaction Volume

The market is projecting explosive growth. Juniper Research forecasts spend on AI agents for customer experience will surge from $1.3 billion in 2025 to $6.6 billion by 2027. Broader projections are even more staggering, with the total agentic commerce market potentially reaching $3 trillion to $5 trillion globally by 2030. This represents a clear, large-scale demand signal for the technology.

Yet the current reality of actual transaction volume is constrained. Consumer trust in AI making payments remains a critical bottleneck. Only 16% of US consumers currently trust AI to make payments, with similar low trust levels reported in the UK. This lack of confidence directly limits the flow of real money through these systems.

The result is a massive 'flow gap' between projected value and current transaction volume. The infrastructure and market demand are being built for a future of autonomous commerce, but the foundational element of consumer trust-and the resulting transaction flow-is lagging far behind the optimistic growth curves.

The Infrastructure Bottleneck

The most immediate constraint is an extreme shortage of compute resources. Demand for training and running AI models is outstripping supply by a factor of ten. This scarcity forces companies to spend over 80% of their total capital raised on compute alone, directly capping their ability to scale services. In essence, the foundational engine for AI agents is already operating at maximum capacity, leaving little room for expansion. Together, these bottlenecks create a hard ceiling on monetizable scale. The compute shortage limits how many agents can run, while the payment infrastructure gap prevents those agents from executing and being paid for transactions. Until both are resolved, the projected trillions in agentic commerce will remain a theoretical future, not a flowing present.

Even if compute were abundant, the financial plumbing for autonomous transactions is missing. Legacy payment processors lack the specialized infrastructure for real-time metering and agent-to-agent settlements. They are built for human-initiated, batched transactions, not the hundreds of micro-transactions per conversation that AI agents will execute. This gap means that even if agents could act autonomously, the systems to bill and settle those actions don't exist.

The Path to Real Transaction Flow

The immediate catalyst for closing the flow gap is the emergence of intermediate control layers. Companies are building these layers not for capability, but for control. The core need is to evaluate policy, block or escalate risky requests, and maintain an immutable audit trail before any financial action. This is a foundational requirement for any business to delegate spend authority to an AI agent, addressing critical risks like wrong payees, duplicate execution, and unclear accountability.

Simultaneously, the major payment networks are rolling out solutions to plug the infrastructure gap. In the last month alone, both Visa and Mastercard launched AI agent payment solutions. These initiatives signal that the established financial plumbing is beginning to adapt, providing the specialized rails needed for real-time metering and agent-to-agent settlements that legacy systems lack.

The key watchpoint is whether consumer trust can accelerate faster than this secure infrastructure is built. The control layers and network solutions address the business and technical bottlenecks, but they do not directly solve the fundamental consumer hesitation. Until trust in AI payments rises from its current low levels, the flow of real transaction volume will remain constrained, no matter how advanced the underlying systems become.