Pinterest Bets $4 Billion on AWS and Launches a Conversational AI Shopping App

A central element of the deal is Pinterest's adoption of AWS's custom silicon. The company will use AWS Trainium chips for AI model training and AWS Graviton processors for general-purpose compute and inference . This marks a strategic shift toward specialized cloud hardware, giving Pinterest the "compute flexibility, hardware optionality, and infrastructure efficiency" to accelerate its AI development, as Chief Technology Officer Matt Madrigal put it . The move also reduces dependence on a single dominant chip vendor at a time when AI hardware supply remains constrained .

Pinterest's stock rose nearly 5% on the day of the announcement .

The New AI-Powered Advertising Tools

At Cannes Lions, Pinterest introduced three products aimed at advertisers, alongside the consumer-facing shopping app. All are in limited release or testing phases as of mid-June 2026.

1. Business Assistant

An AI assistant integrated directly into Pinterest Ads Manager, currently in a closed beta for U.S. advertisers . Instead of requiring marketers to navigate dashboards manually, the tool uses natural-language conversation to surface platform insights, identify campaign opportunities, and spot trends. According to Pinterest, it avoids walls of text and instead shows trend graphs and top-performing Pins to promote .

2. Pinterest Model Context Protocol (MCP)

This is an AI-native infrastructure layer that acts as a standard integration protocol, letting advertisers connect their own AI tools directly to Pinterest's campaign data, analytics, and keyword insights . The protocol aims to make Pinterest’s ad platform accessible and programmable within the broader ecosystem of agentic and copilot tools marketers already use, with grounding in Pinterest-specific signals like taste, trends, and intent .

3. Performance+ Creative Model

An expansion of the existing Performance+ automation suite, this generative AI model evaluates and tests a wider set of ad creatives—images, copy, layouts—to maximize campaign performance automatically . It builds on Pinterest’s earlier efforts to turn product images into lifestyle imagery through generative AI .

Ask Pinterest: A Standalone Conversational Shopping App

The most visible—and experimental—product is Ask Pinterest, a standalone web app in limited access on mobile and desktop . Rather than scrolling through a feed of curated Pins, users open a chat window and type natural-language descriptions of what they are looking for. A query like "cozy small balcony" yields concrete product recommendations and visual suggestions based on the user’s own saved Pins and Boards, personalized further by Pinterest's Taste Graph .

TechCrunch characterized the app as an experiment that could one day make its way into the main Pinterest experience . It represents a fundamental departure from passive browsing toward active, conversational product discovery—a test of whether Pinterest can turn vague inspiration into measurable commerce .

What It Means for Users, Advertisers, and the Platform

For shoppers, the immediate change is philosophical rather than practical: most users won't experience Ask Pinterest unless the limited-access test expands. But the app signals where Pinterest believes product discovery is headed—toward AI-mediated, personalized conversations rather than keyword searches or algorithmic feeds.

For advertisers, the changes are more concrete. The Business Assistant and Performance+ creative tools target faster campaign setup, reduced manual optimization, and better return on ad spend through automation . The MCP protocol, meanwhile, is an architectural play: rather than forcing marketers into Pinterest's own interfaces, it positions the platform as a data layer that external AI agents can query directly, which could make Pinterest ad budgets more defensible as the AI tools market fragments .

For Pinterest itself, the $4 billion AWS deal locks in cloud infrastructure capable of handling enormous AI workloads for the rest of the decade. By committing to custom silicon, the company is hedging against GPU shortages and potentially lowering the unit cost of AI inference at scale—a strategic necessity if conversational search and generative ad creation become default rather than experimental .