Pinterest Taste Graph expansion targets connected TV after tvScientific acquisition

The move into connected television signals a shift from intent-based search to persistent affinity-based targeting in the living room. By leveraging a corpus of hundreds of billions of saved images, Pinterest bypasses the need for third-party cookies or device graphs, relying instead on first-party behavioral decay logic to maintain relevance. This technical development places Pinterest in direct competition with established video ad networks by offering a targeting layer that claims to improve outcomes by 27% through automated modeling. Strategists should monitor whether the 79% precision rate for content labeling remains sufficient as the platform scales its $4 billion AWS-backed infrastructure to handle 640 million monthly users.

Pinterest's push into connected television advertising places it in direct competition with established CTV ad platforms that rely on device graphs and deterministic identity. In early 2025, The Trade Desk announced its OpenPath direct-path integration had reached over 100 publishers, giving advertisers programmatic access to premium CTV inventory without intermediary hops. Meanwhile, Roku reported its advertising revenue grew 27% year over year in Q2 2025, driven by increased CTV ad spend from performance marketers who previously concentrated budgets on social platforms. Pinterest's Taste Graph approach, which relies on affinity signals rather than device-level identity, represents a distinct architectural bet against the identity-resolution model that underpins most CTV targeting today.

The tvScientific acquisition closed in a period of active consolidation among CTV ad-tech vendors. In March 2025, Innovid completed its acquisition of MediaMath, combining CTV measurement with programmatic buying capabilities to offer advertisers a unified platform for planning, activation, and attribution across streaming. Amazon's advertising business surpassed $56 billion in annual revenue for fiscal year 2024, underscoring how retail media networks with first-party purchase data are expanding into video. Pinterest's $4 billion AWS commitment, announced alongside the tvScientific deal, signals the company's intent to scale its graph-based targeting infrastructure to compete with these larger players on measurement and reach.

On the technical side, Pinterest's graph neural network lineage predates the Taste Graph branding. The company's PinSage algorithm, detailed in a 2018 KDD paper as the first graph convolutional network applied at web scale with 3 billion nodes, established the foundation for embedding-based recommendation that Taste Graph now extends to advertising. More recently, Pinterest published research on PinnerSage, a multi-modal user embedding framework that captures diverse interest clusters per user, enabling the kind of multi-faceted affinity modeling that CTV targeting requires. The 79% content-labeling precision cited in the Taste Graph announcement aligns with Pinterest's broader machine learning infrastructure investments, which the company said support over 640 million monthly active users as of mid-2025, suggesting the system has been stress-tested at scale before its CTV deployment.