Meta pledges 1 GW of custom chips as Broadcom’s Hock Tan exits board, boosting AI push

Meta’s decision to allocate a full gigawatt of power to its MTIA custom silicon marks a decisive escalation in the company’s hardware‑first AI strategy, according to CNBC. The gigawatt figure, which translates into roughly the same power consumption as a small data‑center campus, signals that Meta intends to run its own large‑scale inference workloads rather than rely on third‑party GPUs. By co‑designing the chips with Broadcom, Meta can tailor the architecture to the specific demands of its Llama‑derived models and the massive volumes of user‑generated content that feed them. The partnership also gives Broadcom a foothold in the burgeoning AI‑accelerator market, a sector traditionally dominated by Nvidia and AMD.

The timing of the multiyear agreement coincides with the departure of Broadcom founder Hock Tan from Meta’s board, a move CNBC notes underscores the shifting dynamics of the collaboration. Tan’s exit does not appear to affect the technical roadmap; instead, it reflects a broader realignment as both companies double down on their AI ambitions. For Meta, the board change removes a potential conflict of interest while preserving the strategic benefits of Broadcom’s silicon expertise. For Broadcom, the deal offers a steady pipeline of high‑volume orders that could offset the competitive pressure it faces in its core semiconductor businesses.

From a financial perspective, the gigawatt commitment represents a substantial capital outlay. While CNBC does not disclose the exact cost, industry benchmarks suggest that powering a gigawatt of AI accelerators can run into the low‑hundreds of millions of dollars in electricity alone, not to mention the upfront R&D and fabrication expenses. Meta’s willingness to shoulder these costs indicates confidence that the resulting performance gains and cost efficiencies will outweigh the investment, especially as the company seeks to monetize its AI services across advertising, creator tools, and the nascent Metaverse ecosystem.

Strategically, the move positions Meta to compete more directly with the AI‑centric roadmaps of rivals such as Microsoft, Google, and Amazon, all of which have announced sizable investments in custom chips. By controlling the silicon stack, Meta can optimize power‑efficiency, latency, and integration with its existing data‑center infrastructure—advantages that off‑the‑shelf GPUs cannot easily match. CNBC’s reporting suggests that the MTIA chips will be deployed across Meta’s global data‑center network, providing the computational backbone for next‑generation features like real‑time translation, advanced content moderation, and generative media tools.

Finally, the partnership may have broader implications for the semiconductor supply chain. Broadcom’s involvement in AI accelerators could diversify the market beyond the current duopoly of Nvidia and AMD, potentially easing some of the capacity constraints that have plagued AI hardware procurement in recent years. As Meta scales the MTIA fleet, Broadcom will likely need to expand its manufacturing capacity, a development that could spur further investment in fabs and packaging technologies. In sum, the gigawatt pledge, coupled with Tan’s board exit, underscores a calculated bet by Meta to own more of the AI value chain while giving Broadcom a strategic entry point into a fast‑growing segment of the chip market.