Goldman Sachs: AI Agent Commerce Could Tap $2.6 Trillion in US Consumer Spending, Names 18 Potential Winners

By Zhitong Finance

Wall Street financial giant Goldman Sachs has released its latest research report on AI agent commerce, indicating that new AI agent business models—fully ignited by Meta Muse and OpenAI Astra—are poised to drive a new wave of e-commerce penetration and significantly expand and redistribute commercial value across product discovery, digital advertising, digital payments, and online transaction security.

As consumer-facing personal agents like Muse increasingly search for products, compare prices, and execute purchases on behalf of users, the competitive focus in the agent business is shifting toward who can capture consumer purchase intent and convert it into trusted, executable transactions. Goldman Sachs' latest estimates show that, based on current consumption levels, AI agent commerce in the U.S. could tap into a potential annual consumer spending pool of approximately $2.6 trillion over the next few years, covering categories with high adoption likelihood in retail and services. Growth opportunities primarily stem from agents gradually taking over existing shopping expenditures.

$2.6 Trillion Addressable Market

Goldman Sachs analysts have identified approximately $2.6 trillion in existing U.S. consumer spending that falls into categories more readily adoptable for agent commerce—meaning the annual spending pool that agent commerce can potentially capture. The firm highlights key beneficiaries including:

• Consumer entry platforms: Meta (parent of Facebook and Instagram), Alphabet (parent of Google)
• Retail and merchant infrastructure: Amazon, Walmart, and Shopify
• Payment and security providers: Visa, Mastercard, Cloudflare, and key identity and risk management service providers

The report lists a total of 18 publicly listed companies positioned to benefit from this $2.6 trillion consumption migration theme. Ratings and 12-month price targets are as of the report's publication date, with reference stock prices as of September 23, 2026, in U.S. dollars. Potential upside/downside is calculated based on those reference prices.

For global equity investors, opportunities span both the "transaction acquisition" and "transaction security" ends. Goldman Sachs emphasizes this is a structural shift unfolding over the next 3–5+ years, with broader adoption potentially taking even longer. Companies that can sustain benefits must possess user distribution, consumer trust, merchant participation, and transaction execution capabilities. The deployment of Muse and Astra's multi-step execution capabilities provide the technological foundation for this trend, while sustained strong demand for CPUs, storage, and broader AI data center infrastructure hardware represents a long-term, trillion-dollar positive impact on the AI computing chain as applications scale.

Six Pathways Reshaping Online Commerce Profit Distribution

1. Agent Commerce First Captures Existing Spending, Then Drives E-Commerce Growth

Goldman Sachs notes that agent commerce will initially capture existing consumer spending before reducing transaction friction to fuel e-commerce growth. Third-party surveys cited by Goldman show:

• Approximately 44% of online shoppers already use AI for product discovery
• 42% of consumers use AI for price comparison
• Only 16% use AI to complete purchases, indicating significant room for development between product research and actual transactions

Goldman categorizes U.S. retail and service spending by adoption likelihood:

• High-likelihood categories: ~$2.6 trillion
• High + medium likelihood categories combined: ~$12.3 trillion (approximately 60% of relevant spending)

Sensitivity analysis shows that converting just 2.3% of in-person, face-to-face payments in high-likelihood categories to e-commerce could accelerate e-commerce growth by about 1 percentage point. Expanding the denominator to high + medium likelihood categories, the corresponding conversion rate is approximately 0.5%. These are scenario estimates for existing consumption channel migration and should not be interpreted as $12.3 trillion in new market revenue. Adoption speed depends on the subjectivity of product preferences, transaction complexity, and the cost of purchase errors, meaning standardized, low-risk purchases will likely be adopted first, while high-end apparel and luxury goods may take longer.

2. Advertising Value Will Migrate with Purchase Intent

Meta and Alphabet hold advantages in competing for new entry points. AI is already reducing advertising creative costs, optimizing targeting, and improving ad spend returns. Longer-term, consumers may express purchase needs directly within agent interfaces, causing commercial intent to gradually flow from search results pages and merchant websites to AI platforms. Goldman compares this to the shift from desktop to mobile internet: merchants or brands consistently prioritized by agents may gain more stable repeat purchase relationships.

Advertising budgets will gradually shift toward AI-native sponsored recommendations, product information feeds, and other commercial display formats. Meta (leveraging its app ecosystem, user relationships, and Muse) and Alphabet (via Gemini, AI Overviews, AI Mode, and full-stack AI infrastructure) are identified as primary long-term beneficiaries in digital advertising. However, value realization requires solving attribution and performance measurement: brands must know whether ads truly influenced agent choices and final transactions. Goldman also notes that agent commerce has not yet materially impacted retail media performance.

3. More Open Product Discovery Benefits Both SMB Infrastructure Providers and Large Retailers

Goldman's analyst team states that Shopify's opportunity lies in merchants still needing unified management of product catalogs, inventory, payments, orders, fulfillment, and after-sales service, even as consumer shopping entry points evolve. Its collaboration with Google on UCP and Agentic Storefronts helps merchants connect to multiple AI channels. Amazon and Walmart hold advantages in pricing, supply, delivery speed, and transaction reliability—metrics that agents can directly measure when comparing products. Both models can benefit simultaneously.

A key divergence lies in customer relationship ownership: Shopify actively embraces new traffic, while Amazon restricts unauthorized shopping agents, reflecting the latter's focus on protecting first-party data, membership relationships, and retail media revenue. Consumer brands will also diverge: standardized products face greater price comparison and private-label substitution risk, while personalized, high-engagement brands like Estée Lauder are more resilient. Goldman also favors SharkNinja and Tapestry for their brand-building and technology adaptation capabilities. Brand marketing will increasingly optimize for generative search and agent recommendations—a model known as GEO (Generative Engine Optimization).

4. Agent Payments Will Build on Existing Card Networks

Goldman believes Visa and Mastercard are well-positioned due to network effects, consumer payment habits, tokenization, and risk management capabilities. If agents split a shopping basket across multiple merchants, transaction count increases while average order value decreases, potentially boosting per-transaction fee revenue. Online transactions will also increase attachment rates for identity verification, fraud prevention, and other value-added services.

Payment processing technical barriers will also rise, favoring native e-commerce processing platforms like Stripe and Adyen. PayPal's two-sided ecosystem has near-term strategic value, but its competitive position depends on actual value delivered to consumers and merchants. Financing could benefit Affirm and Klarna: agents can compare financing costs and terms at purchase time, increasing the discovery and usage of buy-now-pay-later options, though final payment choice remains driven by consumer preference.

5. Greater Automation Requires Stronger Identity, Authorization, and Liability Frameworks

Technical infrastructure is being built in layers: MCP connects data and tools, UCP and ACP standardize commercial interactions between merchants and agents, and payment protocols handle identity, authorization, and payment credentials. True scale requires proving "which consumer authorized which agent to buy what, with what budget," and clarifying liability for erroneous purchases, fraud, chargebacks, and returns. Traditional signals like IP addresses, devices, and browsing behavior may lose independent judgment capability when legitimate agents also operate at high speed via the cloud, creating new demand for Cloudflare, Akamai, and payment security providers.

Goldman estimates that related cybersecurity spending currently represents about 1%–2% of U.S. e-commerce revenue, with long-term room for growth. In information services, Equifax covers consumer, merchant, income, and employment verification; TransUnion specializes in persistent identity linkage and digital network risk; FICO's clearer opportunity lies in automated decision-making and anti-fraud software. Goldman views agent opportunities for the latter three as long-term growth potential, not yet quantified as explicit earnings upside—the key is whether new workflows generate more paid verification and decision services.

6. Agents Can Improve Sales Efficiency for Existing Supply, but Unique Supply Still Determines Platform Pricing Power

Goldman cites ticketing industry data showing that agents can improve sales efficiency for existing inventory, but unique supply still determines platform bargaining power. Data from Live Nation indicates:

• Approximately 95% of concerts do not sell out
• Amphitheater ticket sales rates: ~60%–70%
• Theater ticket sales rates: ~65%–75%

Agents matching events based on city, time, budget, and personal preferences could increase ticket sales for long-tail events and venue utilization. Ticketmaster (owned by Live Nation) is relatively advantaged due to differentiated ticket inventory and venue partnerships. Secondary platforms like StubHub may initially benefit from lower customer acquisition costs but face challenges from price transparency, service fee competition, and reduced add-on services and ad sales as consumers bypass original pages. Goldman maintains a "Buy" rating on both companies but notes that long-term profit distribution depends on whether agent platforms continuously lower distribution costs or gradually capture more transaction revenue through referral fees, commissions, and display charges.

Every AI Agent Transaction Adds New AI Inference Workload

From a technology paradigm perspective, Muse and Astra are indeed expected to expand agent usage, but adoption will accelerate gradually across different tasks. Muse already features dedicated cloud virtual machines, browser operations, background persistence, and memory mechanisms. Astra enhances computer operation, software usage, and multi-step professional task execution. Applied to shopping, a single request may trigger product search, specification verification, inventory checks, price comparison, delivery assessment, identity verification, and payment confirmation.

Higher AI inference success rates and lower human intervention costs will make more daily tasks worth delegating to AI, potentially expanding user scale, usage frequency, and task coverage simultaneously. However, Goldman emphasizes that account connectivity, merchant participation, and liability rules will determine how quickly technical capabilities translate into real transaction volume.

AI Infrastructure Implications

This inference workload expansion will increase demand for model inference, tool execution, and state management. Key hardware requirements include:

• GPUs/TPUs: Accelerator computing for models
• CPUs: Running browsers, virtual machines, product searches, database queries, and transaction orchestration
• HBM and server DRAM: Carrying model data, context, and concurrent working environments
• Enterprise SSDs: Storing product indexes, task records, and persistent states
• High-speed networking and optical interconnects: Supporting distributed data exchange

NVIDIA's latest engineering documentation explicitly notes that CPU execution speed affects wait times between agent model calls and discusses hierarchical context caching using HBM, DRAM, local NVMe, and remote storage. This provides strong engineering rationale for long-term demand for x86 CPUs (AMD, Intel), Arm-based high-performance CPUs, storage chip components, data center high-speed optical interconnects, and power supply chains.

Storage Remains a Clear Bottleneck

Market research firm TrendForce estimates:

• 2026: Server DRAM contract prices cumulative increase of approximately 270% ; enterprise SSD prices cumulative increase of approximately 235%
• 2027: HBM contract prices may still rise 70%–140%

These figures reflect the combined effect of AI computing expansion and storage price increases. TrendForce's latest estimates show that DRAM and NAND combined will account for 47% of major cloud service providers' capital expenditures in 2026, rising to 68% in 2027, driven by both volume growth and price increases.