Anthropic and OpenAI May Buy Memory Directly From Manufacturers in 2027 as AI Infrastructure Race Intensifies
Anthropic and OpenAI are expected to begin purchasing memory chips directly from manufacturers in 2027, signaling a major shift in how leading AI companies build and control their computing infrastructure.
As artificial intelligence models become larger, faster, and more expensive to train, memory has become one of the most important components in modern AI data centers. High-performance memory is essential for training advanced language models, running AI inference at scale, and supporting the massive data movement required by next-generation AI systems.
Until now, AI companies have typically relied on GPU makers, server suppliers, and cloud infrastructure partners to handle much of the hardware sourcing. That approach may be changing. By buying memory directly, companies like Anthropic and OpenAI could gain more control over cost, supply, performance, and long-term data center planning.
The move would also reflect a broader trend in the AI industry: model developers are no longer just software companies. They are increasingly involved in chip design, server architecture, networking, power management, and full-scale data center strategy. As AI workloads grow, controlling the entire hardware stack is becoming a competitive advantage.
Direct memory purchasing could benefit major memory manufacturers by opening a new category of high-value customers. Instead of selling primarily to GPU companies, server builders, and system integrators, memory suppliers may soon work directly with AI labs that operate enormous computing clusters.
This could be especially important for advanced memory technologies such as high-bandwidth memory, or HBM, which plays a critical role in AI accelerators. HBM is designed to provide extremely fast data transfer between processors and memory, making it vital for training large AI models and running complex AI workloads efficiently.
Demand for AI memory has already surged as companies race to expand computing capacity. If leading AI developers begin locking in direct supply agreements, competition for premium memory chips could become even more intense. This may influence pricing, availability, and investment decisions across the semiconductor industry.
For Anthropic and OpenAI, direct deals with memory manufacturers could help secure long-term supply at a time when AI infrastructure demand continues to outpace availability. It could also allow them to customize hardware configurations more closely around their own model training and inference needs.
The shift may also reshape relationships across the AI hardware supply chain. GPU providers and server manufacturers would remain essential, but AI companies may take a more active role in deciding which memory technologies are used, how systems are configured, and how future data centers are designed.
By 2027, the AI infrastructure market could look very different from today. Instead of relying entirely on traditional hardware vendors, leading AI companies may act more like hyperscale cloud providers, negotiating directly with component makers and shaping the next generation of computing systems from the ground up.
If Anthropic and OpenAI move forward with direct memory purchases, it would mark another sign that artificial intelligence is transforming not only software, but the entire semiconductor and data center ecosystem. Memory chips are becoming a strategic resource in the AI race, and the companies building the most advanced models appear ready to secure that resource more directly.
