open-source AI – Misryoum reports how Pinterest is shifting to open-source and mixed AI models to reduce costs while rolling out new tools for advertisers and users.
Pinterest is quietly reworking the way it buys and builds AI, aiming to deliver more features while shrinking the share of spending tied to expensive models.
Misryoum says the visual discovery platform is pursuing what it calls a “model-agnostic” approach to generative AI. a strategy that blends multiple types of systems rather than relying on one primary vendor.. The goal is straightforward: help its AI-supported tools and advertising products work in the real world. while meeting investor expectations that rapidly growing AI budgets must translate into clear returns.
In practice. Pinterest’s mix combines proprietary models built by its own engineers. closed-source systems from major AI providers. and open-source models from the wider tech ecosystem.. Closed-source models are often designed for fast integration and less hands-on maintenance, but they can be costly.. Open-source models can be downloaded and adapted more deeply. though they typically demand more internal engineering effort to build. maintain. and improve.
Insight: This is a familiar dilemma in AI spending, where performance alone is no longer enough. Companies increasingly manage “how” they deploy models, not just “which” model they choose, to control ongoing compute and usage costs.
Misryoum reports that Pinterest is also testing and iterating its approach through beta programs and changes to how it hires.. The company has been bringing in more software development and AI-focused talent to steward model customization and infrastructure needs. signaling that cost savings from open-source come with an internal operational tradeoff.. Pinterest has also pointed to investments aimed at supporting the blended strategy. including upgrades that enable it to run models efficiently at scale.
The company’s AI rollout has included advertiser-facing tools that turn product catalogs into shopping-oriented pins. alongside voice-enabled functionality that helps users interact with search-like experiences.. In both cases. Pinterest has described using a combination of internal models and third-party systems. with open-source models playing a role in tasks like visual and content understanding and assistant support.
Insight: For advertisers, the value of AI is measured less by novelty and more by speed, relevance, and controllable costs. Blending models can make it easier to experiment without locking the business into one expensive compute path.
At the heart of Pinterest’s plan is a balancing act: prioritize internal models for personalization. use open-source options for cost effectiveness. lean on multimodal capabilities where they fit best. and keep closed-source models when they offer stronger performance for particular workloads.. Misryoum notes that the company has also tied its strategy to infrastructure investment. including plans to expand cloud-computing needs such as graphics processing capacity to power the more customized model mix.
Insight: If Pinterest’s approach holds up in production, it could become a template for other consumer and advertising platforms trying to keep AI feature development moving at investor-friendly cost levels.