Broadcom will supply the technology behind more than 1 gigawatt of Meta’s custom AI chips, the two companies confirmed on April 14, 2026, in what amounts to one of the largest disclosed commitments to in-house silicon by any tech giant. The deal covers Meta’s MTIA accelerator and is expected to scale into “multi-gigawatt” territory over the coming years, according to Broadcom’s press release.
At the same time, Broadcom CEO Hock Tan is stepping down from Meta’s board of directors, severing a direct governance link between supplier and customer just as their commercial relationship reaches a new scale.
What the deal actually covers
Broadcom’s announcement describes an expansion of an existing partnership, not a new contract from scratch. The company will provide custom silicon design and advanced packaging technology for MTIA, Meta’s Training and Inference Accelerator. MTIA is the chip Meta has been developing in-house since at least 2023, when it publicly disclosed the first generation. A second version followed in 2024 with significantly improved performance for ranking and recommendation workloads across Facebook, Instagram, and WhatsApp.
The 1 GW figure signals enormous physical scale. A single large data center building typically consumes between 50 and 200 megawatts of power, meaning the initial MTIA rollout alone could fill the equivalent of several dedicated facilities. The “multi-gigawatt” language in the press release implies a roadmap stretching years into the future, though Broadcom did not attach specific dates or intermediate milestones.
Broadcom’s role centers on what it does best: designing application-specific integrated circuits (ASICs) and providing the high-bandwidth interconnect and packaging technology needed to link thousands of accelerators into training and inference clusters. The company has built a substantial business doing exactly this for hyperscale cloud operators, and Meta now joins that roster at a scale that appears to rival Broadcom’s largest existing custom-chip customers.
Why Meta is building its own chips
Meta is not alone in this push. Google has been running its Tensor Processing Units (TPUs) for nearly a decade. Amazon Web Services deploys its Trainium and Inferentia accelerators, designed by subsidiary Annapurna Labs. Microsoft unveiled its Maia AI chip in late 2023. The common motivation is the same: reduce reliance on Nvidia’s GPUs, which dominate AI training, and tailor hardware to the specific neural network architectures each company runs most often.
For Meta, the calculus is straightforward. The company operates some of the world’s largest recommendation and generative AI systems, serving billions of users daily. Custom silicon tuned to those workloads can deliver better performance per watt than general-purpose GPUs, and it gives Meta leverage in negotiations with external suppliers. That does not mean Nvidia disappears from Meta’s data centers. The company is widely expected to continue purchasing merchant GPUs for the most demanding frontier training runs while shifting more predictable inference tasks onto MTIA.
The financial terms of the Broadcom deal, including total contract value, per-chip pricing, and payment structure, have not been disclosed. Broadcom’s press release includes standard forward-looking risk language, and the company’s SEC filings warn that large customer programs can be delayed, resized, or canceled depending on market conditions. Those caveats are boilerplate, but they are worth noting: a multi-gigawatt roadmap is an ambition, not a binding purchase order.
Hock Tan’s board exit and what it signals
The timing of Hock Tan’s departure from Meta’s board raises questions that neither company has fully answered. Tan, who transformed Broadcom from a mid-tier chipmaker into a semiconductor and software conglomerate through a string of acquisitions, had occupied a seat that gave him direct visibility into Meta’s AI infrastructure strategy. That arrangement was always unusual: a CEO of a major supplier sitting on the board of one of his largest customers creates inherent conflicts around pricing, roadmap decisions, and competitive intelligence.
As of April 2026, no joint statement from Meta or SEC Form 8-K filing has surfaced to explain the reasoning. The most straightforward interpretation is that both companies decided cleaner governance separation was necessary as the dollar amounts grew. But other possibilities exist, from routine board refreshment to broader strategic shifts at either firm. Without a direct disclosure, the motivations remain opaque.
What is clear is the optics: Broadcom and Meta are tightening their technical partnership while loosening their boardroom ties. For investors tracking Broadcom’s customer concentration risk, the combination is worth watching closely.
Open questions for investors and the industry
Several gaps remain that upcoming earnings calls and regulatory filings should help fill. First, the timeline: Broadcom has not said when the multi-gigawatt expansion will materialize, and chip technology moves fast. By the time later phases of the MTIA program ramp, Nvidia’s next-generation Rubin architecture and competing custom efforts at Google and Amazon will have advanced as well. Meta will need to iterate on MTIA’s design without sacrificing the economies of scale that justify building custom chips in the first place.
Second, the revenue impact for Broadcom is hard to model without pricing data. The company already derives a significant share of its semiconductor revenue from a handful of hyperscale customers. Adding Meta at this scale could meaningfully shift that concentration, for better or worse.
Third, independent benchmarking of MTIA against current merchant GPUs, including Nvidia’s Blackwell B200 family, has not appeared publicly. Until third-party or academic comparisons surface, it is difficult to judge whether Meta’s custom silicon represents a genuine performance advantage or primarily a supply-chain diversification strategy.
The deal confirms that the era of hyperscalers designing their own AI chips is accelerating, not slowing down. Broadcom is positioning itself as the go-to partner for companies that want custom silicon without building an entire chip design operation from scratch. Whether that bet pays off at the scale both companies are projecting will depend on execution, market timing, and how quickly the AI hardware landscape shifts beneath them.
More from Morning Overview
*This article was researched with the help of AI, with human editors creating the final content.
