As global demand for artificial intelligence infrastructure continues to surge, Microsoft appears to be reshaping its hardware strategy to secure greater control over the chips powering its AI services. According to a report by The Information, the company is in discussions with Broadcom to co-design custom AI chips, signaling a deeper commitment to in-house and semi-custom silicon development.
The move comes as competition for high-end AI accelerators reaches unprecedented levels. With generative AI models growing larger and more resource-intensive, cloud providers are under pressure to ensure reliable access to specialized hardware while keeping costs and supply risks under control.
Microsoft Expands Its Custom Silicon Strategy
Microsoft has already been working with Marvell on aspects of chip design, but the report suggests the company is now broadening its partnerships. Broadcom’s involvement would mark a strategic expansion, bringing in a firm with deep experience in custom accelerators, networking silicon, and large-scale data center deployments.
Broadcom’s existing relationship with OpenAI—a key Microsoft partner—also makes it a natural fit for designing hardware optimised for large-scale generative AI workloads. While neither Microsoft nor Broadcom has publicly confirmed the discussions, the reported talks reflect a growing trend among hyperscalers to reduce reliance on off-the-shelf AI processors.
Microsoft’s reported plans mirror a wider movement across the technology sector. Major cloud and AI players are increasingly investing in custom-designed chips to complement—or partially replace—NVIDIA GPUs.
- Google continues to advance its in-house Tensor Processing Unit (TPU) platform and is preparing to make the chips more broadly available outside its own services.
- Amazon Web Services recently introduced Trainium3, its most powerful AI accelerator to date, aimed at large-scale model training.
- Meta is developing custom AI silicon in partnership with Marvell, with a potential launch slated for around 2027.
These efforts underscore a shared goal: greater control over performance, power efficiency, and long-term costs in AI data centers.
NVIDIA’s Market Position
Despite this wave of custom silicon projects, NVIDIA remains the dominant force in the AI accelerator market. Its GPUs continue to set benchmarks for performance and benefit from a mature software ecosystem, particularly CUDA and its extensive developer tools.
However, industry analysts note that custom chips do not need to outperform NVIDIA’s offerings across the board to have a meaningful impact. Even diverting a portion of internal workloads—such as inference or highly optimized training tasks—to custom hardware could reduce dependence on NVIDIA and reshape purchasing dynamics.
If Microsoft proceeds with Broadcom, it would further strengthen the trend toward vertically integrated AI infrastructure, where cloud providers design hardware closely aligned with their software and workloads. While NVIDIA is unlikely to lose its leadership position in the near term, the rise of custom accelerators suggests the AI hardware market is entering a more competitive and diversified phase.
As AI adoption accelerates across industries, the ability to control both silicon and software is quickly becoming a strategic advantage—and Microsoft’s reported talks with Broadcom indicate it does not intend to be left behind.
