As the global contest for artificial intelligence supremacy shifts from software models to the infrastructure that powers them, Meta has made one of its clearest long-term bets yet: owning more of the silicon stack behind its AI future.
In a sweeping expansion of its partnership with Broadcom, the Facebook and Instagram parent said it will co-develop several generations of custom AI processors through 2029, committing an initial more than one gigawatt of computing capacity in what it described as the first phase of a sustained, multi-gigawatt rollout.
The scale of the commitment underscores the intensity of the capital race among Big Tech firms as they seek to reduce reliance on Nvidia’s costly and supply-constrained processors while building proprietary infrastructure for next-generation AI services.
Register for Tekedia Mini-MBA edition 20 (June 8 – Sept 5, 2026).
Register for Tekedia AI in Business Masterclass.
Join Tekedia Capital Syndicate and co-invest in great global startups.
Register for Tekedia AI Lab.
To put the size of the deployment into perspective, the initial one-gigawatt commitment is enough to power roughly 750,000 U.S. homes on average, making this one of the most significant disclosed custom silicon deployments in the consumer technology sector.
The announcement marks more than a supply agreement. It reflects Meta’s accelerating transition toward a vertically integrated AI infrastructure strategy, one in which the company designs increasingly specialized hardware for its own workloads rather than depending primarily on third-party GPUs.
Meta founder and chief executive Mark Zuckerberg framed the partnership in precisely those strategic terms.
“Meta is partnering with Broadcom across chip design, packaging, and networking to build out the massive computing foundation we need to deliver personal superintelligence to billions of people,” Zuckerberg said. “As we roll out more than 1GW of our custom silicon to start and then multiple gigawatts over time, this partnership will give us greater performance and efficiency for everything we’re building.”
The move ties Meta’s hardware roadmap directly to its broader AI ambition: delivering real-time intelligent experiences across Facebook, Instagram, WhatsApp, Threads, and its growing suite of generative AI products.
Meta has centered the collaboration on Meta Training and Inference Accelerator (MTIA) program, the company’s in-house chip initiative designed to support both recommendation systems and generative AI inference workloads. The first chip in the lineup, the MTIA 300, is already being used to power Meta’s ranking and recommendation systems, while three additional generations are scheduled through 2027.
These later generations are being optimized specifically for inference, the computationally intensive process through which AI systems respond to prompts, rank content, personalize feeds, and generate outputs in real time.
Training models is capital-intensive, but inference at Meta’s scale, serving billions of users daily, may ultimately become the more commercially consequential cost center. Every AI-generated response, recommendation, or ranking event carries an infrastructure cost. Custom chips tailored for Meta’s specific workloads can materially lower cost per inference and improve latency.
That is one of the clearest reasons hyperscalers are increasingly moving toward ASICs, or application-specific integrated circuits. Unlike general-purpose GPUs, ASICs are designed for narrower tasks but often deliver better power efficiency and lower total cost of ownership for those workloads.
Other companies, such as Alphabet and Amazon, have also expanded their custom silicon programs as AI demand pushes infrastructure spending to record levels. The broader industry trend is that Big Tech is moving aggressively to reduce dependence on Nvidia’s dominant GPU ecosystem.
That trend has made Broadcom one of the biggest winners of the AI infrastructure boom. The chipmaker has positioned itself as a critical enabler for hyperscalers seeking custom accelerators, advanced packaging, and high-performance networking. In Meta’s case, Broadcom’s Ethernet networking technology will connect the company’s rapidly expanding AI clusters, addressing one of the most important bottlenecks in large-scale AI systems: data movement across thousands of accelerators.
As AI clusters scale into the hundreds of thousands of chips, networking throughput becomes nearly as important as raw compute power. This is where Broadcom’s role extends beyond chip design. The company said the partnership is built on its XPU platform, which is specifically designed for custom AI accelerators and optimized for large-scale deployments.
Broadcom chief executive Hock Tan emphasized the long-term scope of the collaboration.
“We are pleased to expand our strategic collaboration with Meta as they pioneer the next frontier of artificial intelligence,” Tan said. “This initial MTIA deployment is just the beginning of a sustained, multi-generation roadmap to serve the trajectory of massive growth over the next few years that highlights Broadcom’s unmatched leadership in AI networking and the power of our foundational XPU custom accelerator platform.”
Another important element of the announcement is the governance shift. As part of the deal, Hock Tan will leave Meta’s board and transition into an advisory role focused on the company’s custom chip strategy. That move suggests the relationship is becoming deeper and more operationally embedded, with Broadcom’s leadership directly influencing Meta’s long-term silicon roadmap.
Wall Street responded positively.
Broadcom shares rose about 3.5% in extended trading, while Meta’s stock was little changed, a signal that investors view Broadcom as a major beneficiary of sustained hyperscaler AI capital expenditure.
Separately, Meta also disclosed a boardroom development unrelated to the Broadcom deal: Tracey Travis, who has served on the board since 2020, will not stand for re-election at the annual shareholder meeting.