The AI Chip Shake-Up: Nvidia's Grip Loosens as Giants Diversify
A recent report from Michael Cembalest, Chairman of Market and Investment Strategy at J.P. Morgan Asset Management, points to a fundamental reshaping of power dynamics within the AI compute market. The analysis, released on November 15th, examines the future compute infrastructure commitments already locked in by leading AI labs like OpenAI and Anthropic.
The Striking Numbers Behind the Shift
The most startling figure is the projected share of Nvidia GPUs in these future plans: a mere 39%. This stands in stark contrast to Nvidia's dominant 83% share of overall AI chip revenue from 2023 to 2025. This precipitous drop suggests a potentially significant contraction of Nvidia's influence in the next generation of AI hardware infrastructure.
The Challengers Emerge: A Diversified Vendor Landscape
So who is capturing the remainder of the market? The report outlines an increasingly competitive and varied field:
- Broadcom's custom chips rank second with a 24% share, indicating strong demand for specialized hardware from top AI firms.
- Google's Tensor Processing Units (TPUs) hold 16%, reflecting Google's continued investment in proprietary silicon.
- AMD GPUs have secured 15% of the share, becoming a key choice for cost-effective procurement.
- Amazon's Trainium chips account for 6%, bolstering the underlying hardware ecosystem for its cloud services.
The Tech Giants' "De-Nvidiazation" Playbook
This vendor breakdown reveals a broader industry trend: major former Nvidia clients like Google, Amazon, and Microsoft are actively pursuing two core strategies to reduce dependency and manage costs.
The first path is collaborative or in-house custom chip development The second path is adopting more cost-effective alternatives. AMD's GPUs, with their competitive pricing and improving performance, have become an attractive option. In the same announcement, OpenAI confirmed it would also procure 6 gigawatts of compute power from AMD GPUs. This multi-sourcing strategy grants AI companies greater flexibility in negotiations and supply chain resilience. This report is more than a procurement snapshot; it reads like a declaration of a new phase in the AI hardware race. When the most advanced AI labs begin to significantly diversify their compute sources, it sends a powerful message to the entire industry: the era of monolithic, single-vendor GPU architecture is fading. The future is likely a heterogeneous compute ecosystem built on a mix of general-purpose GPUs, custom ASICs, and cloud-native chips. This shift may represent a healthier turning point for the industry, promising lower overall AI development costs, spurred hardware innovation, and mitigated supply chain risks.What This Signals: From Monopoly to a Heterogeneous Future