Key Takeaways

  • Anthropic's enterprise momentum supports continued AI infrastructure investment, as the company captured 40% of U.S. enterprise LLM spending in 2025.
  • AMD, NVIDIA, and SanDisk remain exposed to concentrated purchasing by model developers, hyperscalers, and cloud providers.
  • Rapid infrastructure growth could support semiconductor sales while increasing valuation, competition, and oversupply risks.

Anthropic’s expanding position in enterprise artificial intelligence illustrates how shifting model leadership can alter hyperscaler purchasing patterns and hardware demand.

Menlo Ventures, as reported by ZDNET, estimated that Anthropic captured 40% of U.S. enterprise LLM spending in 2025, compared with 27% for OpenAI. The estimate covered 495 U.S. enterprise decision-makers and production API spending.

These figures indicate Anthropic's Claude is securing enterprise production workloads, which involve deep integration with corporate data and translate into sustained demand for computing capacity.

Shifting leadership among Anthropic, OpenAI, and other model developers can redirect chip demand across cloud platforms, accelerator architectures, memory configurations, and storage systems.

This dynamic creates an opportunity for AMD as it competes with NVIDIA in AI accelerators. Enterprises looking for additional capacity or alternatives to a dominant hardware supplier could support AMD’s accelerator business, though software compatibility, availability, and cloud-provider purchasing decisions ultimately influence which chips run the associated workloads.

IDC reports global AI infrastructure spending reached $89.9 billion in Q4 2025, up 62% year over year. Full-year spending totaled $318 billion, more than double 2024’s $153 billion. IDC forecasts spending of $487 billion in 2026 and more than $1 trillion by 2029, implying an approximate 31% compound annual growth rate from 2025.

AI servers represented 98% of AI-centric infrastructure spending in Q2 2025, while accelerated servers accounted for 91.8% of that server spending. Semiconductor revenue is forecast to reach $1.29 trillion in 2026, up 52.8% year over year, with DRAM revenue projected at $418.6 billion as high-bandwidth memory and AI systems lift demand.

SanDisk’s NAND flash and enterprise-storage exposure provides a route into AI-server expansion, as retrieval systems, training data, checkpoints, and generated content increase storage requirements. However, NAND has historically been cyclical; capacity added during a demand boom can pressure prices if deployment schedules slip.

A concentrated group of hyperscalers and model developers currently controls a large portion of accelerator purchases. This concentration can produce enormous orders, followed by abrupt changes in timing, architecture, or supplier mix.

While AI infrastructure demand remains robust, stock valuations may assume unusually high growth will continue with limited interruptions. As model developers like Anthropic secure enterprise adoption, semiconductor suppliers such as AMD, NVIDIA, and SanDisk must navigate shifting customer concentration, competition, and potential cyclical oversupply.