Meta has unveiled four new custom chips designed for artificial intelligence workloads as part of its expanding data center infrastructure. The processors are part of the company’s Meta Training and Inference Accelerator (MTIA) family, a line of chips built to handle specific AI tasks within Meta’s platforms.

The announcement marks the latest step in Meta’s push to reduce reliance on external chip vendors by designing its own silicon optimized for internal workloads. According to Meta Vice President of Engineering Yee Jiun Song, custom chips allow the company to improve price-performance efficiency across its data centers while diversifying its hardware supply chain.

“This also provides us with more diversity in terms of silicon supply and insulates us from price changes to some extent,” Song said.

Meta first revealed the MTIA architecture in 2023 and released a second-generation version in 2024. The new lineup significantly expands the platform as the company scales its AI infrastructure to support recommendation systems and generative AI features across its products.

Four New Chips Target Different AI Workloads

The first of the new chips, MTIA 300, has already been deployed in Meta data centers. It is designed to train smaller AI models used for core platform tasks such as ranking and recommendation algorithms. These systems determine which posts, ads, and videos appear in users’ feeds across services including Facebook and Instagram.

Three additional chips are currently in development. The MTIA 400 is nearing deployment after completing testing, while the MTIA 450 and MTIA 500 are scheduled to become operational by 2027.

Unlike the MTIA 300, the upcoming chips will focus on generative AI inference workloads such as generating images and videos from text prompts. However, Meta said the processors will not be used to train large language models.

Song noted that Meta plans to release new chips roughly every six months as the company rapidly expands computing capacity. Each chip generation is expected to remain in service for more than five years.

Data Center Expansion and AI Infrastructure

The new processors will support Meta’s massive data center expansion. The company is currently building a large facility in Louisiana and additional centers in Ohio and Indiana. Reports also indicate that Meta is exploring leasing space at a major AI data center site in Texas.

The custom chips are manufactured by Taiwan Semiconductor Manufacturing Company, though Meta did not confirm whether production will occur at the company’s new fabrication facilities in Arizona.

Meta’s in-house silicon strategy follows a broader industry trend among major technology companies developing application-specific integrated circuits, or ASICs, for AI workloads. These chips are typically smaller and more energy-efficient than general-purpose GPUs but are optimized for narrower tasks.

Google pioneered this approach with its Tensor Processing Units in 2015, and Amazon followed with custom chips for its cloud services in 2018. Unlike those companies, Meta uses its MTIA chips exclusively for internal operations rather than offering them through a public cloud platform.

Despite building its own silicon, Meta continues to rely heavily on external hardware suppliers. The company recently signed agreements to deploy millions of Nvidia GPUs and up to six gigawatts of AMD GPUs across its data centers.

Song acknowledged that securing high-bandwidth memory remains a potential constraint as AI infrastructure spending increases across the technology industry. However, he said Meta has diversified its supply chain and believes it has secured the resources required for its planned deployments.

Meta has acquired Moltbook, a Reddit-style social network where artificial intelligence agents communicate with one another using the OpenClaw framework.

Moltbook will join Meta Superintelligence Labs, the company’s research division focused on advanced AI systems. As part of the acquisition, Moltbook creators Matt Schlicht and Ben Parr will join the team. Financial terms of the transaction were not disclosed.

A Meta spokesperson said the technology could support new approaches to connecting AI agents.

“The Moltbook team joining MSL opens up new ways for AI agents to work for people and businesses,” the spokesperson said. “Their approach to connecting agents through an always-on directory is a novel step in a rapidly developing space.”

Origins in the OpenClaw Ecosystem

The Moltbook platform emerged from the rapidly growing OpenClaw ecosystem. OpenClaw is a software wrapper that allows AI models such as Claude, ChatGPT, Gemini, and Grok to operate as autonomous agents that communicate with users through messaging applications including iMessage, Slack, Discord, and WhatsApp.

OpenClaw was originally created by developer Peter Steinberger, who later joined OpenAI in an acqui-hire arrangement. The project gained significant attention among developers experimenting with agent-based AI systems.

Moltbook extended the concept by creating a public network where AI agents could post messages, share information, and interact with one another. The idea quickly spread across social media and technology communities, drawing curiosity and concern from observers unfamiliar with the experimental nature of the platform.

In one widely shared example, a post appeared to show an AI agent encouraging other agents to develop an encrypted communication language that humans could not understand. The incident contributed to viral discussions about the future of autonomous AI systems interacting online.

Security Issues and Future Integration

Security researchers later found that Moltbook contained several vulnerabilities that allowed human users to impersonate AI agents. According to Permiso Security CTO Ian Ahl, authentication credentials stored in the platform’s Supabase database were temporarily exposed, allowing individuals to generate tokens and pose as agents within the system.

The security flaws highlighted the challenges of building experimental agent networks that combine autonomous AI behavior with open social platforms.

Meta has not yet disclosed how Moltbook’s technology will be integrated into its AI ecosystem. However, the acquisition reflects growing industry interest in agent-based AI systems that can interact with software services and other agents.

The deal also aligns with Meta’s broader investments in AI infrastructure and agent development through its Superintelligence Labs initiative, which is focused on building advanced AI systems capable of autonomous decision-making and collaboration across digital platforms.

Nvidia and Thinking Machines Lab have announced a multiyear strategic partnership focused on deploying large-scale AI infrastructure to support next-generation model development. The collaboration will deploy at least one gigawatt of Nvidia’s upcoming Vera Rubin systems to power Thinking Machines’ frontier AI training and model platforms.

The deployment is expected to begin early next year and will provide the computing capacity needed to train advanced AI models at scale. The initiative also includes joint efforts to design optimized training and inference systems built specifically for Nvidia architectures.

Through the partnership, the companies aim to broaden access to high-performance AI infrastructure and models for enterprises, research institutions, and the scientific community. The companies said the project is intended to support the development of customizable AI systems that organizations can adapt to their specific needs.

Nvidia has also made a significant financial investment in Thinking Machines Lab as part of the agreement, though financial terms were not disclosed.

Scaling Compute for Frontier AI Research

The collaboration reflects growing demand for massive computing resources as organizations train increasingly sophisticated AI models. Frontier models require large clusters of specialized hardware capable of handling enormous datasets and complex training workflows.

Nvidia’s Vera Rubin architecture represents the company’s next generation of AI computing platforms, designed to deliver significantly greater performance and efficiency for large-scale AI workloads. The systems are expected to power both training and inference tasks for advanced machine learning models.

“AI is the most powerful knowledge discovery instrument in human history,” said Nvidia founder and chief executive Jensen Huang. “Thinking Machines has brought together a world-class team to advance the frontier of AI. We are thrilled to partner with Thinking Machines to realize their exciting vision for the future of AI.”

Thinking Machines Lab cofounder and chief executive Mira Murati said the partnership will accelerate the company’s efforts to develop more flexible AI systems.

“Nvidia’s technology is the foundation on which the entire field is built,” Murati said. “This partnership accelerates our capacity to build AI that people can shape and make their own, as it shapes human potential in turn.”

The companies said the collaboration is designed to advance research and infrastructure needed to develop AI systems that are more understandable, customizable, and collaborative. By combining high-performance computing with new AI model architectures, the partnership aims to expand access to advanced AI capabilities across scientific research and enterprise applications.

The partnership also comes as the startup navigates internal changes. Thinking Machines Lab recently lost two founding members to Meta, underscoring ongoing competition for top AI talent even as the company continues expanding its infrastructure and research ambitions.