Meta scraps second-generation Olympus AI accelerator, boosts Nvidia and AMD supply deals

METAMETA

Meta Platforms halted development of its second-generation Olympus AI accelerator, ending its custom silicon push after heavy investment. The company has inked substantial supply agreements with Nvidia and AMD and plans to deploy Google tensor processing units, shifting AI compute sourcing to established GPU and TPU vendors.

1. Meta abandons second-generation Olympus project

Meta Platforms decided to cease development of its second-generation Olympus AI accelerator, ending a multi-year in-house silicon program that aimed to power its AI workloads.

2. Shift to merchant GPUs and TPUs

The company has finalized supply agreements with Nvidia and AMD for GPU provision and will deploy Google's tensor processing units, shifting AI compute sources to established external vendors.

3. Strategic and financial implications

The pullback underscores the technical and capital challenges of custom chip design, reducing the risk of vertical integration for GPU providers and raising questions about future hardware strategy and costs.

4. Impact on Broadcom and partnerships

Broadcom may lose custom silicon opportunities from the cancelled scheme but could see increased networking demand for TPU deployments; the collaboration with Google is expected to be tactical rather than long-term due to competitive overlaps in digital advertising.

Sources

FBFF