Why the projections are so far apart
The major research firms disagree by a factor of 35 on how big agentic commerce will be in 2030. The gap reflects the definition each one uses more than the quality of its models. Some count only purchases executed by an agent. Others count every purchase influenced by AI, including search, comparison, and recommendations.
| Source | Horizon | Metric | What is counted |
|---|---|---|---|
| McKinsey (Oct. 2025) | 2030 | $3T–$5T globally; up to $1T in the US | Commerce “orchestrated” by AI: influence plus execution, broad scope |
| Morgan Stanley (Dec. 2025) | 2030 | $190B (base case) to $385B (high case), US | US e-commerce actually executed by agents |
| eMarketer (2025) | 2029 | $144B | Strict definition of an agentic purchase |
| Gartner (2025) | 2028 | Over $15T, B2B | B2B purchases mediated by agents, a separate scope |
| Emerging consensus | 2030 | 10%–25% of US e-commerce, or roughly $200B–$500B | Purchases in which an agent plays a significant role |
Three scenarios for 2030
Forecasting here works through scenarios rather than a single prediction. The three paths below bracket most published analyst views. Shares are expressed as a percentage of e-commerce in developed markets.
| Conservative | Base case | Disruption | |
|---|---|---|---|
| Share of e-commerce executed by agents | ≈ 5 % | 10%–20% (Morgan Stanley range) | Over 25% (McKinsey high end) |
| Main trigger | Agents remain a search channel; the human still makes the final click | Instant Checkout and similar features go mainstream; mandates become a de facto standard | Recurring purchases (groceries, consumables, subscriptions) move to autopilot at scale |
| Payment rails | Almost exclusively tokenized cards | Cards dominate, stablecoins hold machine-to-machine, and A2A waits in the wings | A reshuffle: the rail becomes invisible, and the agent picks the cheapest one for each transaction |
| Main obstacle | Fraud, disputes, and reluctance from consumers and merchants (the Amazon-Perplexity precedent) | How fast merchants publish structured data | Regulation and concentration: two or three agent platforms control access to the customer |
The 2026 data still fits both the base case and the disruption scenario: AI traffic is growing fast, it converts better than conventional traffic, and Europe has seen its first live transactions. Payments history counsels caution. Contactless took 10 years to catch on, and that was with the infrastructure already in place.
Potential winners and losers
Agentic commerce puts three separate profit pools in the value chain up for grabs. Advertising and SEO own discovery today, e-commerce platforms and PSPs own checkout, and the card networks own the payment rail. Each scenario redistributes them differently.
| Company | Best case | Main threat |
|---|---|---|
| Card networks (Visa, Mastercard) | Agentic tokens win out: every agent purchase stays a card transaction, with richer data | Machine-to-machine payments slip away to stablecoins; interchange comes under pressure if agents pick rails on cost |
| A2A / instant payments (Wero, Pix, UPI) | Agents ignore marketing and pick the cheapest rail, so A2A wins payments where card protections matter little | No native framework for agent mandates; payer protection still has to be built |
| Stablecoins (USDC, EURC) | De facto standard for machine-to-machine settlement (x402, pay-per-crawl, APIs) | Tougher AML rules; reliance on the dollar is a problem outside the US |
| Agent platforms (OpenAI, Google, Anthropic…) | Become the front door to commerce and take a cut (already about 4% on Instant Checkout for Shopify merchants) | Gatekeeper regulation (DMA), merchant pushback, mandated interoperability |
| PSPs and orchestrators | New services to sell: multi-protocol integration, KYA, mandate management, rail routing | Margin squeeze if agents negotiate acquiring costs in real time |
| Merchants | Merchants that control their product data get global distribution with no ad spend | Disintermediation: the brand disappears, prices are compared constantly, and merchants depend on agent platforms |
| Traditional search advertising | – | The most likely loser: agents skip the banner ads and read data, not slogans |
Checkout and the PSP business in 2030
If the base case or the disruption scenario plays out, checkout as we know it disappears for a growing share of transactions. For those flows, there is no longer a payment page, a form, or a purchase funnel to optimize. In their place, the agent, the merchant, and the rail negotiate through APIs, out of the human’s sight.
- Conversion optimization (A/B testing buttons, one-click) loses value; the quality of data and APIs gains it.
- The PSP becomes a rail orchestrator and trusted third party: it verifies agents (KYA), executes mandates, routes to the lowest-cost rail, and arbitrates machine disputes.
- Fraud prevention shifts from “Is this click human?” to “Is this mandate valid, and can this agent be trusted?”
- Customer support partly becomes agent-to-agent support: systems negotiate disputes, returns, and refunds with each other, under human supervision.
For payments professionals, the practical takeaway is the same in every scenario. The skills gaining value are machine identity management, mandate compliance, and multi-rail orchestration. None of these three projects depends on how big the channel ultimately gets, so all of them can start now.
Open questions
As of July 2026, five questions still have no settled answer. They cover consumer appetite, the outcome of the litigation, standards consolidation, where regulators land, and the economics of the channel.
- Actual consumer appetite: handing off search is one thing; handing off payment with no confirmation step is a much bigger leap of trust. Adoption of “human not present” mandates at scale is still unknown.
- The legal outcome: the Amazon-Perplexity case (injunction in March 2026, appeal pending) will decide whether a merchant can block agents, and so whether the channel develops with or against the big marketplaces.
- Standards consolidation: will AP2, ACP, UCP, and x402 converge, or will we face 10 years of costly fragmentation, as with wallets?
- Where regulators land: SCA and PSD3 rules for mandates in Europe, AML treatment of machine micropayments, and whether agent platforms count as gatekeepers. That makes three open questions and not one final text.
- The economics of the channel: if agent fees (about 4% on Instant Checkout) stack on top of existing fees, some merchants will not find it worthwhile. Final pricing has not settled yet.
Three topics cover each building block in depth: Agentic payments: the fundamentals (protocols), Agent identity and mandates (trust), and Stablecoins and machine-to-machine settlement (alternative rails).