A new report indicates that Broadcom has secured memory chip pricing comparable to NVIDIA's, eroding the latter's historical advantage in volume-based discounts. The development follows Samsung's ability to negotiate favorable terms through increased order volumes, a shift that could reshape pricing dynamics in the AI hardware supply chain.

The report, published by the Substack newsletter WK, notes that Samsung's expanded procurement of high-bandwidth memory has allowed it to offer prices aligned with NVIDIA's recent agreements. This comes as NVIDIA faces pressure to maintain gross margins, which declined to 75% in its fiscal second quarter and are projected to fall to 71–72% in the upcoming fiscal fourth quarter. Meanwhile, Samsung is leveraging elevated memory prices to offset losses in its foundry division, which analysts estimate could shrink by 42% by 2026 due to process complexity in next-generation chip fabrication.

Broadcom's strengthened bargaining position stems from growing procurement volumes, enabling it to negotiate on more equal footing with NVIDIA. This shift has implications for major customers like Anthropic, which is reported to be Broadcom's largest anticipated client by 2027, planning deployments of 5 gigawatts of TPU capacity in 2027 and 10 gigawatts by 2028. Reuters has separately reported that Broadcom has committed up to $42 billion in financing for Anthropic's AI infrastructure buildout, a figure representing a significant portion of the company's projected $125 billion in lease obligations over five years. The arrangement raises potential conflicts of interest regarding access to computing resources.

Industry analysts note that while NVIDIA previously held an edge through higher procurement volumes, Broadcom's increasing scale has narrowed the gap. The report underscores broader trends of consolidation and pricing pressure in the AI semiconductor market, where memory procurement and financing partnerships are becoming pivotal to competitive positioning.