[ad_1]
These alternatives, known as application-specific integrated circuits (ASICs), are designed for narrower AI workloads and can offer lower costs and improved efficiency compared to traditional GPUs.
Google was an early pioneer in the field, debuting its Tensor Processing Units in 2015. Amazon followed with its first custom AI chip in 2018. Unlike Google and Amazon, which provide their chips through cloud services, Meta uses its MTIA processors exclusively for internal AI systems.
The latest agreement arrives shortly after Broadcom announced expanded chip partnerships with Google and Anthropic, including plans involving 3.5 gigawatts of Google TPU deployments.
Broadcom stock has gained roughly 10% so far in 2026, outperforming the S&P 500’s approximately 2% increase during the same period.
Meta also said Tracey Travis, the former finance chief of Estée Lauder, will leave the company’s board after serving since 2020.
The company has aggressively increased AI spending in recent months after announcing in January plans to invest up to $135 billion in AI infrastructure throughout the year.
On Broadcom’s March earnings call, Tan also said OpenAI is expected to deploy its first custom AI chip generation in 2027.
Meta’s recent AI infrastructure commitments include plans for up to 6 gigawatts of AMD GPUs, millions of Nvidia AI chips, and additional custom silicon developed with chip design company Arm Holdings.
In total, Meta is planning 31 AI-focused data centers worldwide, including 27 facilities across the United States.
Leave a Reply