AI Retail and E-Commerce: A Value-Chain Study

Core Conclusions

• AI retail and e-commerce is not the endpoint of "model capability." It is the application layer that forms once model capability is embedded into transactions, advertising, payments, fulfillment, membership, and merchant software, creating profit pools that are billable or amplifiable. U.S. online retail was roughly $1.234 trillion in 2025; U.S. retail media ad spend was roughly $60.3 billion in 2025 and roughly $71.1 billion in 2026. The first thing to monetize is not "chat" itself but the advertising and transaction loop wrapped around shopping traffic.

• The clearest, most mature AI use cases with real revenue are still retail media / onsite advertising, search and recommendation optimization, merchant ad automation, payment risk control, customer-service automation, and supply-chain forecasting and fulfillment optimization. Amazon's advertising services revenue reached $68.635 billion in 2025; Walmart global advertising grew 37% in Q4 FY26, with U.S. Walmart Connect up 41%; DoorDash disclosed that its advertising business already exceeded $1 billion in annualized revenue in 2024, covering 30-plus countries and serving 150,000+ advertisers; eBay advertising revenue was $581 million in Q1 2026, or 2.6% of GMV.

• Shopping agents and conversational commerce have entered the transition from "product launch–user trial" toward a "partial transaction loop," but most platforms have not yet broken it out as a separate revenue line. Amazon says its shopping AI assistant helped more than 300 million customers research, compare, and buy products in 2025; Walmart launched Sparky in 2025 and announced in October 2025 that it would support Walmart Instant Checkout through ChatGPT; Shopify has connected Agentic Storefronts to ChatGPT, Google AI Mode/Gemini, and Microsoft Copilot; Instacart's Cart Assistant covered only about 25% of U.S. users in Q1 2026.

• The profit pool will most likely stay with platform companies that own transaction data, the consumer gateway, merchant relationships, payment authorization, fulfillment networks, and ad inventory, rather than with standalone AI shopping tools. Google, OpenAI, and Perplexity can capture the "discovery gateway" and some high-intent traffic, but without payments, refunds, inventory, delivery, and after-sales accountability, they cannot monopolize retail economics. That is precisely why Shopify, PayPal, Walmart, and Amazon are proactively opening up catalogs, protocols, and AI checkout capabilities.

• The biggest financial leverage for platforms is advertising, not agent subscriptions. Advertising combines high gross margin, low fulfillment burden, and strong coupling to GMV and traffic. Amazon advertising growth keeps outpacing total revenue; Walmart lists advertising and membership side by side as "diversifying profit sources"; DoorDash treats advertising as its local-commerce growth engine; Instacart also lists advertising and data capabilities as growth pillars.

• AI's more certain impact on e-commerce is not "fully replacing platforms" but rewriting the logic of traffic allocation and shelf governance inside platforms. Keyword search, homepage shelves, cashback shopping guides, and traditional SEO will matter less; structured product attributes, real-time inventory, delivery promises, price credibility, review summaries, payment authorization, and an "agent-ready catalog" will matter more. Google has connected AI Mode, virtual try-on, price alerts, and agentic checkout to its Shopping Graph; Shopify has explicitly launched Catalog, MCP, and UCP-compatible infrastructure; OpenAI has launched shopping research, product discovery, and Instant Checkout.

• The direct beneficiaries truly worth tracking are the platforms that have already turned AI from an internal efficiency tool into "higher GMV / higher ad revenue / higher merchant revenue / higher payment penetration": Amazon, Walmart, Shopify, JD, Alibaba, Instacart, DoorDash, MercadoLibre, and PayPal. By contrast, much of AI content generation, general-purpose customer-service bots, single-point merchant copilots, standalone visual search, and pure dynamic-pricing tools remain in the high-risk stage of pilots, low-price competition, or being absorbed into platforms.

• The profit leverage of supply-chain and fulfillment AI is underestimated by the market. Amazon's fulfillment expense was $109.074 billion in 2025, or 15.2% of net sales; Walmart keeps using automation, in-store fulfillment, and digital delivery as efficiency levers; JD has deeply integrated AI into its "super supply chain" and sees large-scale use in customer service and merchant marketing; MercadoLibre's logistics network handles more than 1.8 billion items a year. Every percentage point of efficiency gain in the fulfillment chain is more sensitive for profit and cash flow.

• Dynamic pricing is one of the "hottest in narrative, most heavily regulated" tracks. The FTC issued a dedicated "surveillance pricing" study and request for comment in 2025; California privacy regulators have finalized new rules on automated decision-making technology; Instacart directly halted on-platform item price tests in December 2025 amid controversy. This shows that in many categories dynamic pricing is still a concept or experiment, not a profit pool that can be scaled with confidence.

• The highest probability of disruption falls not on Amazon/Walmart but on traditional shopping-guide cashback, price comparison aggregators, low-end customer-service outsourcing, manual ad operations agencies, basic SEO copy tools, and single-point SaaS that lacks transaction data. In the agent era the core is not generating a piece of copy; it is getting real product data, prices and inventory, coupons, payment and authorization, and after-sales and fulfillment capabilities. Shopify, PayPal, Google, and OpenAI are all turning these capabilities into protocols and platforms.

• The most important catalysts over the next 12–24 months will come from four threads: first, whether ChatGPT, Google AI Mode, Copilot, and others push shopping from recommendation to scaled checkout; second, whether platforms disclose the real conversion rates, average order values, repurchase rates, and ad lift that agents bring; third, whether retail media keeps growing fast and extends to off-site and in-store; fourth, whether regulators tighten dynamic pricing, personalized-recommendation transparency, and platform accountability.

• The biggest risk is not that "AI cannot be built" but that "AI gets built yet cannot form an independent profit pool"—showing up as insufficient consumer authorization, platforms making it free, ad-slot repricing, high compute costs, liability for wrong recommendations, cross-border and data-sovereignty constraints, and agents turning platform front-end traffic into back-end fulfillment plumbing. Gartner, as relayed by Reuters, expects that by 2027 more than 40% of agentic AI projects may be cancelled over unclear cost and value—particularly relevant for high-valuation, low-validation AI e-commerce narratives.

The Value-Chain Landscape and Profit Pools

Start by distinguishing five stages: product launch, user trial, conversion validation, revenue landing, and scaled adoption. The key dividing line in AI retail today is not "AI or no AI" but whether a company has crossed the two thresholds of "conversion validation" and "revenue landing." Leading platforms such as Amazon, Walmart, Shopify, JD, Instacart, and DoorDash have broadly completed product launch and user trial; their advertising, merchant tools, payments, customer service, and supply-chain optimization have entered revenue landing; scenarios where "agents charge on their own" remain rare.

From the standpoint of where profit pools accrue, three groups are most likely to win over the long run. The first is platform winners: Amazon, Walmart, Shopify, Alibaba, JD, MercadoLibre, Instacart, and DoorDash, because they hold consumer intent, product catalogs, payments, merchant relationships, and fulfillment all at once. The second is settlement/authorization-layer winners: companies like PayPal that turn payments, identity, risk control, and authorization into cross-agent infrastructure. The third is picks-and-shovels vendors: Criteo, Constructor, Gorgias, and the like—provided they can prove they are not just a "model shell" but have real e-commerce data, workflows, and ROI.

The table below further separates "real revenue" from "high narrative but unvalidated":

Business Models, Value Capture, and Scenario Analysis

AI retail's business models split roughly into six types. The core is not "who has the best model" but who can embed AI into a continuously billable transaction flow. The best business models usually share three traits: tying to GMV or ad budgets, reusing existing traffic and data, and having the platform bear fulfillment, trust, and payment responsibility. Amazon, Walmart, Instacart, DoorDash, and eBay all show that advertising is lighter than commissions and often higher-margin; Shopify and PayPal show another way to extract a profit pool—the "infrastructure + protocol layer + payment layer" model.

If you press the question "will shopping agents expand platform GMV, or move the gateway from e-commerce platforms to model companies and general-purpose agents," it currently looks more like a two-way reallocation. The gateway will indeed migrate toward general-purpose agents, but the profit pool will not necessarily migrate in step: Google, OpenAI, and Perplexity can control discovery and intent formation, while Amazon, Walmart, Shopify, and PayPal are fixing themselves as the transaction and fulfillment layer through Catalog, Checkout, MCP/UCP, and payment authorization. For investors, the most critical variable is not "who wins the conversation" but "who gets the order, the payment, the refund, the logistics, and the ad attribution."

A large e-commerce platform's AI budget today goes mainly to five places: search/recommendation and catalog standardization, ad optimization, customer-service automation, payment risk control, and fulfillment and inventory optimization. This is clear from Amazon's sharp increase in AI-related capex, Walmart's emphasis on supply-chain automation and in-store fulfillment, JD's emphasis on the super supply chain and AI customer service, Instacart's launch of Catalog Intelligence and Store View, and DoorDash's use of AI in local commerce for advertising, menus, and an autonomous delivery platform.

An offline retailer's AI budget skews more toward store inventory accuracy, store-delivery order allocation, out-of-stock/replenishment, labor scheduling, in-store media, and membership personalization; a DTC merchant's AI budget tends to skew toward creative and product feeds, customer service and returns, CRM and ad delivery, onsite search, and A/B testing; a cross-border merchant's AI budget skews more toward translation, localization, pricing and tax, customs compliance, cross-border risk control, and multi-currency checkout. Within these budgets, the items most likely to create incremental revenue are advertising and conversion lift; the items most likely to create profit improvement are customer service, risk control, inventory, and fulfillment.

Scenario analysis is as follows:

The base scenario deserves the most attention. The reason is simple: it does not require consumers to immediately hand over all shopping to agents; it only requires platforms to turn AI into faster discovery, higher conversion, better advertising, and lower customer-service and fulfillment costs. This is already validated by leading platforms' product roadmaps and earnings-call language.

Track Maturity and Competitive Reshaping

The table below compresses the thirty tracks listed in the prompt into twelve groups for investment judgment; each group maps to multiple sub-scenarios in the original question.

Among these tracks, the five points most worth emphasizing are as follows.

First, retail media is still the best profit pool. AI will not weaken it; instead it will evolve from keyword advertising into "conversational sponsored slots," "agent-readable product promotion," and "off-site closed-loop budgets." Amazon, Walmart, DoorDash, and eBay have already proven this with real revenue.

Second, shopping agents will erode part of the value of traditional search and homepage shelves, but not the value of the transaction loop itself. For platforms, the homepage, search box, ranking, and ad auction do not disappear; they are rewritten into more "machine-consumable" product understanding, inventory availability, logistics promises, and commercial placements. Google and Shopify are both pushing open protocols; Amazon and Walmart lean more toward closed in-platform agents.

Third, fresh produce, groceries, daily necessities, and local quick-commerce will be agent-enabled earlier than fashion and luxury. The reason is that these scenarios are high-frequency, have strong replenishment logic, standardized SKUs, and obvious price comparison, and consumers have higher tolerance around "fast and cheap." The first landing capabilities of Instacart, DoorDash, Walmart, and Alibaba/Qwen are all clearly concentrated in groceries, dining, local services, and high-frequency replenishment.

Fourth, the payment and checkout layer will be underestimated. Once consumers no longer open each app one by one but place orders directly inside AI, payment authorization, identity confirmation, refund reversals, merchant-liability allocation, coupon invocation, and tax calculation become the layer with the most pricing power again. PayPal's intensive partnerships with OpenAI, Google, Microsoft, and Perplexity in 2025–2026 are precisely about claiming this position.

Fifth, dynamic pricing is not the best investment track in the short term. Not because it has no value, but because commercial upside and regulatory risk are amplifying almost simultaneously. The FTC, CPPA, and the EU are all tightening rules on automated decision-making, transparency, and platform accountability; Instacart halting its price tests further shows this track is not yet mature enough on consumer trust.

Company Screening, Tiering, and Scoring

First, a practical tiering.

For the scoring model, we suggest following the prompt's weights: direct revenue or GMV exposure 20%, transaction data and gateway barrier 20%, fulfillment supply chain 15%, ad monetization 15%, financial quality 10%, growth leverage 10%, valuation reasonableness 10%. Because valuation conventions, forward consensus, and EV/GMV measures across markets are hard to fully reconcile at the same point in time, the table below leads with a capability and commercialization-certainty score, gives only directional valuation judgments, and marks figures that cannot be rigorously verified as "needs further validation."

Next, a more "research-card"-style master table of key listed companies. Given length and public-disclosure limits, the table focuses on the fifteen most worth tracking, prioritizing the latest public earnings or company announcements; figures without a unified standard, such as forward PE/EV/GMV, are marked "needs further validation."

A directional judgment on "which companies already reflect AI e-commerce expectations and which still have an expectation gap":

• Fairly fully reflects AI e-commerce expectations: Shopify, Amazon, DoorDash. The first two's AI platform status is already market consensus; DoorDash's valuation is usually quite sensitive given its strong local-commerce, advertising, and autonomous-delivery narrative. In 2025 the market repeatedly used Shopify's AI-channel capability as a valuation anchor.

• May still have an expectation gap: JD, PayPal, Instacart, MercadoLibre. JD's AI shows up more in supply chain and customer service, and the market may not fully award an "AI retail" premium; PayPal's "settlement layer" value in agent checkout is not yet fully digested by the market; if Instacart can keep scaling advertising and enterprise retail tech, it would re-rate from "grocery platform" to "grocery tech + media"; MercadoLibre's three-in-one advertising/logistics/payments long-term value is higher than its current explicit AI disclosure.

• Strong narrative but weak financial validation: Etsy's external AI gateways, many standalone merchant content tools, and open-web dynamic-pricing tools. They are not without potential; they simply lack, for now, evidence of "standalone revenue" and "scaled adoption."

On the private and primary-market side, based on current public information, the highest-confidence names worth tracking are as follows:

Risks, Regulation, and the Disrupted

The main regulatory thread in AI retail has shifted from "can it be done" to "how can it be done compliantly." The U.S. FTC issued a public request for comment on surveillance pricing in 2025 and disclosed interim findings; California has finalized new rules covering automated decision-making, risk assessment, and cybersecurity audits; the EU's DSA and DMA impose constraints on recommendation transparency, platform accountability, and self-preferencing, respectively; China's PIPL and E-Commerce Law require personal-information processing to be minimal and transparent, and strengthen platform-operator responsibility.

This brings four kinds of investment risk.

The first is consumer-trust risk. If AI recommendations cause wrong purchases, hidden price differences, unclear promotional terms, or misleading checkout, regulators will go straight to the platform. Instacart halting item price tests in December 2025 is the most direct example.

The second is platform free-ification risk. Shopify, Amazon, Walmart, Google, and PayPal are all building in or protocolizing agent-ready catalog, search, merchant copilot, checkout, and payment capabilities. Standalone tools without exclusive data, cross-platform execution, or industry workflows are easily marginalized fast.

The third is gateway-replacement risk. If AI shopping gateways take shape quickly, traditional off-site shopping guides, cashback affiliates, comparison sites, and some paid-search traffic will be diluted. For platforms this is not catastrophic, because they can still defend the checkout, logistics, and ad loop; but for middle-layer traffic businesses it will hurt a lot.

The fourth is fulfillment and liability risk. The closer to automated purchasing, the more one faces inventory errors, substitution disputes, refunds, chargebacks, late deliveries, cross-border taxes, counterfeits, and review governance. Platforms with physical stores, warehouse networks, and payment risk-control systems are actually stronger in this round. The defensive strength of Amazon, Walmart, JD, MercadoLibre, and Coupang comes precisely from these "clunky but important" foundations.

The disrupted parties can be understood more concretely:

Final Conclusions

AI retail and e-commerce is the layer of the AI value chain closest to real payments, real orders, and real profit pools. For investing, what matters more than "demo effect" is closed-loop capability and the billing path. Returning to the ten core judgments in the prompt, the conclusions compress as follows:

The five sub-tracks most worth watching are retail media, in-platform search and recommendation, payments and agent checkout, supply-chain/inventory optimization, and shopping agents in groceries and local quick commerce. The first two determine revenue leverage; the latter two determine profit leverage; and high-frequency groceries determine user penetration.

The ten listed companies most worth further research are Amazon, Walmart, Shopify, JD.com, Alibaba, Instacart, DoorDash, MercadoLibre, PayPal, and Sea. For a more conservative tilt, Amazon/Walmart/JD/MELI place more weight on the loop and cash flow; for a more leveraged tilt, Shopify/Instacart/DoorDash/PayPal are more worth watching on new protocols and new gateways.

The private companies and internal product lines most worth tracking are OpenAI, Perplexity, Constructor, Gorgias, and Sierra, plus Amazon Alexa for Shopping, Walmart Sparky, Shopify Catalog/Agentic Storefronts, Alibaba Qwen-Taobao, and Instacart Cart Assistant. The first five represent the agent gateway and the picks-and-shovels ecosystem; the latter five represent the internal product lines where large platforms truly have a chance to scale.

The five points the market most easily misreads are also clear: First, AI shopping does not mean AI shopping companies can monopolize the profit pool. Second, shopping agents need not replace platforms to reshape platform traffic and ad structure. Third, the highest-quality revenue today still comes mainly from advertising and payments, not agent subscriptions. Fourth, the profit leverage of inventory/fulfillment optimization is often larger than front-end chat. Fifth, dynamic pricing is not a "naturally high-margin track"; its regulatory and trust thresholds may surface faster than its commercial value.

Over the next 6–12 months, what to track is not "which agent is the most human-like" but the following metrics: whether platforms disclose the conversion rates, average order values, and repurchase rates that AI search/agents bring; whether retail media keeps growing faster than GMV; whether Shopify Catalog and UCP/MCP produce visible merchant new-customer acquisition; whether the AI checkout of PayPal/Google/OpenAI generates scaled orders; whether Instacart and DoorDash's AI gateways improve GTV/orders in groceries and local retail; and whether regulation on dynamic pricing, privacy, and algorithm transparency escalates.

For a narrower follow-up research direction, we suggest prioritizing "shopping agents + the payment/checkout layer." This thread best answers the most critical question of the next two years: will e-commerce platforms degrade into a product catalog and fulfillment plumbing, or will they keep holding the profit pool through Catalog, payment authorization, and ad restructuring. At this stage, the public materials most worth reading together on this question are Google AI Mode shopping and agentic checkout, OpenAI shopping research and Instant Checkout, Shopify Catalog/MCP/UCP, PayPal's Agentic Commerce Services, the Walmart–OpenAI partnership, Amazon Alexa for Shopping, and Instacart and DoorDash's AI gateway efforts in grocery scenarios.

Open questions and limitations: This report prioritizes company filings, investor materials, official product announcements, and regulatory materials, but three classes of issues remain not fully resolved. First, valuation conventions across markets such as forward PE, EV/GMV, and EV/Sales are hard to reconcile on a single date, so the valuation section uses directional judgment rather than mechanically listing figures. Second, the revenue, ARR, funding rounds, or customer details of some private companies are not fully disclosed, and this report includes only high-confidence names. Third, some second-tier retail tech, merchant SaaS, shopping-guide cashback, and logistics-automation names in China's A-shares/Hong Kong stocks require separate deep dives into local-language materials to reach more granular conclusions.

This report is based on public information and does not constitute investment advice. Markets carry risk; invest with caution.