Meta is doubling down on artificial intelligence with two major announcements this week — unveiling its third-generation custom inference chip and rolling out AI-powered features for Facebook Marketplace.
MTIA 300: Meta's Latest AI Silicon
On March 12, Meta announced the MTIA (Meta Training & Inference Accelerator) 300, its third-generation AI inference chip specifically designed for generative AI workloads. This represents Meta's continued investment in custom silicon to power AI capabilities across its platforms, following earlier generations of the MTIA chip family.
The MTIA 300 marks a significant step in Meta's strategy to reduce dependence on third-party chipmakers like NVIDIA and build proprietary hardware optimized for its unique AI inference needs. As Meta integrates AI more deeply across Facebook, Instagram, WhatsApp, and its metaverse ambitions, having purpose-built chips allows the company to improve performance while potentially reducing operational costs.
While Meta did not disclose specific performance benchmarks for the MTIA 300, the company's trajectory suggests meaningful improvements in inference speed and efficiency over previous generations. Custom chips have become essential for major tech companies as AI workloads scale — Google has its Tensor Processing Units (TPUs), Amazon has its Trainium and Inferentia chips, and Microsoft has partnered on custom silicon.
Facebook Marketplace Gets Smart with Meta AI
On the same day, Meta announced new AI features for Facebook Marketplace, its peer-to-peer shopping platform with hundreds of millions of users worldwide. The company is introducing AI-powered auto-replies that will automatically draft responses to the most common buyer question: "Is this still available?"
Sellers can enable this feature when creating listings, and Meta AI will generate editable responses based on information from the listing, including description, availability, pickup location, and price. This addresses a major pain point for Marketplace sellers, who frequently receive identical inquiries about item availability.
Beyond auto-replies, Meta is also using AI to speed up the listing process itself. The platform now analyzes photos uploaded by sellers to help populate listing details, reducing the manual effort required to create a complete listing.
These features represent Meta's broader push to make Marketplace more competitive against platforms like eBay, Craigslist, and OfferUp by reducing friction for both buyers and sellers. AI-powered communication could help close more deals by ensuring prompt responses, while automated listing assistance lowers the barrier for new sellers.
Two-Pronged AI Strategy
Together, these announcements illustrate Meta's dual-track approach to AI: building powerful underlying infrastructure while simultaneously deploying user-facing AI products. The MTIA 300 chip will power inference for Meta's AI products behind the scenes, while the Marketplace features put AI directly in the hands of everyday users.
This strategy mirrors broader industry trends. As AI competition intensifies among Microsoft, Google, Amazon, and OpenAI, Meta must demonstrate both technical sophistication and practical applications. The Marketplace AI features provide tangible value that could increase user engagement and transaction volume, while custom silicon ensures Meta isn't solely reliant on external chip suppliers for its expanding AI ambitions.
Looking ahead, expect Meta to expand AI features across its family of apps. The company has previously integrated Meta AI (its conversational assistant) into Facebook, Instagram, and WhatsApp. The Marketplace additions suggest Meta is finding specific, high-utility applications for AI rather than just launching general-purpose chatbots.
The MTIA 300 chips are expected to begin deployment in Meta's data centers later this year, powering inference for Meta's various AI services.