Meta expands multibillion-dollar alliance with Broadcom for its own AI chips

Meta has expanded its multi-billion partnership with Broadcom, betting on joint development of custom chips for its AI systems. At the same time, the corporate relationship between the companies is changing: Hock Tan is stepping down from Meta's board of directors.

New Level of Alliance

The expanded agreement between Meta and Broadcom takes their relationship beyond a typical "supplier-customer" scheme. The companies plan to jointly design and produce specialized chips that will operate within Meta's AI infrastructure. For a company that simultaneously develops advertising algorithms, recommendation systems, and generative models, this is a matter not only of performance but also of control over the entire computational chain — from hardware to software stack.

The formulation about a multi-billion partnership demonstrates the scale of the bet. Meta has long been increasing its spending on data centers and accelerators, and Broadcom gets a chance to establish itself as one of the key technology partners in the most expensive segment of the AI market. If previously the industry mainly discussed purchases of universal GPUs, then increasingly attention is shifting toward custom solutions created specifically for the internal loads of a large platform.

Why Meta Needs Its Own Chips

Meta's own or jointly designed accelerators are not needed for the sake of image. With such a volume of computations, even small optimization in energy consumption, memory bandwidth, or the cost of a single server quickly translates into millions of dollars in savings. Moreover, custom hardware provides more freedom in how to run model training, inference, and services that work for billions of users. For Meta, such an approach has several practical advantages:

• more precise tuning of chips for its own AI workloads
• reduced dependence on universal solutions and external roadmaps
• more predictable planning of supplies for data centers
• a chance to reduce the total cost of infrastructure as it scales
• the ability to more closely link hardware with its internal AI stack and network architecture

This does not mean an immediate rejection of third-party accelerators: large AI companies typically build a hybrid stack and combine different types of computational hardware. But the trend is clear: the more money goes into models and servers, the stronger the interest in own chips. For Broadcom, such dynamics are beneficial because the company knows how to work as a partner in custom design and manufacture of complex semiconductor solutions.

Changes on the Board

A separate storyline of the news is Hock Tan's departure from Meta's board of directors. Against the backdrop of an expanding commercial partnership, this move looks logical from a corporate hygiene perspective. When a CEO of one of the most important suppliers or technology partners sits on the board of a client, the market inevitably raises questions about conflicts of interest, access to sensitive information, and boundaries of influence.

In such a configuration, a clearer division of roles can even be more beneficial for both sides. Meta gains a cleaner management structure at a moment when it is making a major bet on AI hardware. Broadcom, in turn, maintains and grows its commercial contract without the extra burden of discussions about corporate control.

The fact that board changes are happening simultaneously with the deepening of the hardware alliance underscores: this is not about distancing but rather about a more formalized model of cooperation.

What This Means

The AI market is increasingly moving away from a race purely for models toward a race for infrastructure. Meta's decision to deepen its alliance with Broadcom shows that the next field of competition will be custom chips, energy efficiency, and control over the cost of computations. This is where margins, the speed of launching new services, and dependence on external suppliers will be determined.