[Trend] Escaping NVIDIA: Meta and Broadcom to Create Custom AI Silicon

Following Google, Mark Zuckerberg's corporation is betting on hardware independence. On April 15, 2026, Meta announced an expanded partnership with Broadcom to co-develop custom AI chips.

The deal reflects a fundamental trend: no Big Tech company wants to pay a massive margin to the monopolist NVIDIA. Training the recently announced Muse Spark model (and future iterations of AGI) requires millions of accelerators. Broadcom serves as an ideal partner, providing network interfaces and expertise in ASIC design. The joint development of proprietary AI accelerators (MTIA) will allow Meta to drastically reduce inference CAPEX and optimize its data centers specifically for the architecture of its neural networks.

Source: Meta / Reuters