Meta is poised to commence production of the latest iterations of its custom AI-specific chips in September, a strategic move aimed at mitigating escalating GPU costs amidst an unprecedented component shortage. This development was reported by Reuters, citing an internal memo from the company.
According to the memo, at least one of these chips successfully completed its testing phase within approximately six weeks. While Meta is collaborating with Broadcom on the chip's design, manufacturing will be handled by Taiwan's TSMC. The company is also sourcing key components from various suppliers, including RAM from Samsung, storage solutions from Sandisk, and fiber optic equipment from Sumitomo Electric.
The four new chips, developed under Meta's Training and Inference Accelerator (MTIA) program, were first detailed by the company in March. Some of these are either currently being deployed or are slated for deployment this year or next. Meta has adopted a modular design philosophy for these chips, anticipating that the rapid evolution of AI will necessitate adaptability by the time the chips reach full production.
As the company articulated at the time, "Each MTIA generation builds on the last, using modular chiplets, incorporating the latest AI workload insights and hardware technologies, and deploying on a shorter cadence."
These proprietary chips are expected to generate significant cost savings by reducing Meta's reliance on GPUs from major chipmakers like Nvidia and AMD, though the company still anticipates substantial expenditures with these established providers. Meta plans to integrate MTIA chips for critical tasks such as training models for its ranking and recommendation algorithms, handling broader AI workloads, and facilitating inference for its diverse applications. Meta has been developing its own AI chips since 2023.
Meta is making colossal investments to secure the vast computing capacity required to power its extensive AI initiatives. In April, the company projected capital expenditures for the current year to range between $125 billion and $145 billion, a substantial portion of which is dedicated to its AI endeavors.
Globally, Meta has been actively forging data center and power agreements, committing tens of billions of dollars to ensure sufficient computing power for training and deploying its new Muse Spark series of AI models. The company aims to deploy 7 gigawatts of compute capacity this year, with plans to double that figure next year, as reported by Reuters based on the internal memo.
Further bolstering its infrastructure, Meta signed a deal with ARM last year to secure compute for its recommendation systems. This is in addition to multi-billion dollar agreements with AMD for its Instinct GPUs and with Amazon for utilizing the cloud giant’s homegrown CPUs to address AI-related requirements.
Meta is not alone in its efforts to curb the considerable capital outflow to Nvidia. Last month, OpenAI unveiled an inference processor developed in collaboration with Broadcom, while Anthropic is reportedly exploring the development of its own chips with Samsung. Both Amazon and Google already design their own chips for AI training and inference, and a burgeoning ecosystem of startups is also emerging to meet the burgeoning demand in this space.
Meta declined to comment on these developments.
The Editorial Staff at AIChief is a team of professional content writers with extensive experience in AI and marketing. Founded in 2025, AIChief has quickly grown into the largest free AI resource hub in the industry.