Tech Giants Race for Custom Silicon: Meta and Broadcom Lead AI Chip Revolution

While NVIDIA continues to dominate the AI chip market with record-breaking revenues, Silicon Valley's biggest players are quietly mounting a rebellion. On March 4-5, 2026, both Meta and Broadcom unveiled ambitious plans—not just to reduce reliance on NVIDIA, but to claim a decisive stake in the AI computing landscape. This silicon arms race is reshaping the global semiconductor industry in real time.

Meta's Silicon Gambit: From Recommendation Algorithms to AI Training

Meta's Chief Financial Officer Susan Li dropped a bombshell at the Morgan Stanley Technology Conference on March 4, revealing that the company is developing custom silicon chips to train next-generation AI models. "Some of our workloads really are very customized to us," Li stated. "The ranking and recommendations workloads have been where we have started, and that's the place where we have rolled out custom silicon at the most scale. But we expect and are hopeful that we are going to expand that over time, including eventually to training AI models."

This isn't Meta's first rodeo with custom silicon. The company launched its first in-house chip, the Meta Scalable Pixel Processor (MSPP), back in 2023 for video processing and recommendation algorithms. However, according to The Information, Meta previously abandoned a more advanced AI training chip project due to design roadblocks. The latest pivot underscores Meta's strategic determination to control its AI infrastructure.

Meta is currently one of NVIDIA's largest customers, purchasing billions of dollars worth of H100 GPUs in 2024 alone. But as AI models balloon in scale and training costs soar, developing custom chips has become imperative—not just for cost reduction, but for greater autonomy. Industry analysts suggest that if Meta successfully deploys custom silicon for AI training, it could significantly reduce dependence on NVIDIA while gaining more leverage during chip shortages.

Broadcom's Bull Run: AI Chip Sales to Surpass $100 Billion

Just as Meta revealed its chip ambitions, Broadcom delivered news that sent shockwaves through Wall Street. CEO Hock Tan said during the earnings call on March 4 that the company expects AI chip sales to exceed $100 billion in 2027—a projection that would mark a staggering growth in market share.

The announcement sparked a near-3% surge in Broadcom's stock price, with investors betting big on the chip designer's AI prospects. Tan emphasized that demand for custom chips is accelerating, particularly from major tech companies. He highlighted Broadcom's unique position in the custom silicon space, offering highly optimized solutions tailored to specific customer requirements.

Notably, Broadcom serves as the primary custom chip supplier for Google's TPU (Tensor Processing Unit), while also providing tailored silicon for Meta, Amazon, and other tech giants. Unlike NVIDIA's general-purpose GPUs, Broadcom specializes in customization, delivering superior performance per watt for specific AI workloads. This differentiated strategy has secured Broadcom a unique standing in the AI chip ecosystem.

A Shifting Paradigm: From General-Purpose to Application-Specific

The recent moves by Meta and Broadcom signal a profound shift in the AI chip industry. While NVIDIA still reigns supreme in AI training with its CUDA ecosystem and GPU dominance, new contenders are emerging—Google's TPU, Meta's custom silicon, and Broadcom's application-specific chips are forming a new competitive landscape.

This race is fueled by explosive growth in AI computing demand. Tech giants have realized that merely purchasing NVIDIA chips is neither cost-effective nor sustainable. Custom silicon offers not only lower costs but also supply chain resilience—ensuring they're not held hostage during shortages.

From a broader perspective, the AI chip market is transitioning from "general-purpose dominance" to "customization wins." As AI applications become increasingly specialized, application-specific chips optimized for particular tasks demonstrate superior efficiency and cost-performance. Broadcom's $100 billion sales projection perfectly encapsulates this trend.

*参考来源：Bloomberg、Reuters、The Information