Broadcom CEO says it will deliver 3 generations of MTIA accelerators to Meta
- Broadcom CEO confirms delivery of 3 generations of MTIA accelerators to Meta by end of 2027
- Company cites line of sight for Meta to deploy 3 gigawatts of power capacity through 2028
- Roadmap highlights deepening strategic partnership in custom AI silicon and infrastructure

*this image is generated using AI for illustrative purposes only.
Broadcom Inc CEO outlined a multi-year hardware roadmap for Meta Platforms, stating the company will deliver three generations of its custom MTIA accelerators between now and the end of 2027.
During a conference call, the executive also highlighted infrastructure scale, noting there is a line of sight for Meta to deploy 3 gigawatts of power capacity through 2028.
Strategic Hardware Roadmap
The commitment involves successive iterations of Broadcom’s custom silicon designed specifically for Meta’s AI workloads. The delivery schedule spans from the current period through the end of 2027.
Infrastructure Scale
Alongside the chip roadmap, the CEO pointed to significant energy requirements associated with Meta’s data centre expansion. The stated line of sight for 3 gigawatts of deployment through 2028 underscores the scale of capital expenditure and power procurement likely required by the social media giant.
What the Numbers Show
The linkage between the hardware delivery timeline (ending 2027) and the power deployment horizon (extending to 2028) suggests that infrastructure build-out may lag slightly behind or run parallel to the final generation of chip deployments. The 3 gigawatt figure represents a substantial load, equivalent to the output of several large nuclear reactors, indicating heavy reliance on grid-scale power solutions.
How might Broadcom's exclusive focus on Meta's custom MTIA accelerators impact its competitive positioning against NVIDIA in the broader AI chip market?
What specific power procurement strategies or renewable energy partnerships is Meta pursuing to secure the 3 gigawatts of capacity needed by 2028?
Will the slight lag between hardware delivery (2027) and full power deployment (2028) create bottlenecks in Meta's AI training or inference capabilities?
































