Google's massive expansion of AI infrastructure is creating a supply crunch for specialized memory chips, and two South Korean companies are positioned to capture most of the demand. As Google increases its 2026 capital expenditure budget to between $195 billion and $205 billion for AI data centers, the company is ramping up orders for high-bandwidth memory (HBM), a specialized form of DRAM (dynamic random access memory) that optimizes the performance of AI processors.

HBM is essential to how modern AI systems work. Unlike standard memory, HBM is packaged directly with AI chips like graphics processing units (GPUs) and tensor processing units (TPUs) to dramatically improve their speed and efficiency. As Google builds out its AI infrastructure, it needs vastly more of this specialized memory than ever before, but the supply chain is tightly controlled by just two major manufacturers.

Who Supplies Memory to Google's AI Data Centers?

The memory supply chain for AI infrastructure reveals a clear hierarchy. Samsung is the primary HBM supplier for Google's custom TPUs, supplying more than 60% of the company's HBM needs. SK Hynix provides the remainder of Google's TPU memory requirements. However, SK Hynix's role extends far beyond Google. The South Korean company is also the main HBM supplier for Nvidia's GPUs, which means it's positioned to benefit from spending increases across the entire AI infrastructure ecosystem.

This division of labor reflects the reality of modern chip manufacturing. Samsung and SK Hynix are among only a handful of companies worldwide capable of producing high-bandwidth memory at scale. Their duopoly on HBM production gives them significant leverage as hyperscalers race to build AI data centers. Google's decision to increase its capex budget signals that other companies like Meta, Microsoft, and Amazon are likely to follow suit, raising their own spending on AI infrastructure and multiplying demand for memory chips.

How to Understand the AI Memory Supply Chain

• Samsung's Primary Role: Samsung supplies more than 60% of Google's HBM needs for its custom TPUs, making it the dominant memory partner for Google's AI infrastructure expansion.
• SK Hynix's Dual Position: While SK Hynix provides the remaining portion of Google's TPU memory, it is the main HBM supplier for Nvidia's GPUs, giving it exposure to multiple hyperscaler customers and their capex increases.
• Supply Constraints: Only two companies can manufacture high-bandwidth memory at the volumes required by major AI chip makers, creating a bottleneck that limits how quickly hyperscalers can scale their AI infrastructure.
• Cascading Demand: Google's capex increase from $180-190 billion to $195-205 billion signals that competitors will likely raise their own AI spending, multiplying demand for memory chips across the industry.

The broader picture reveals why memory suppliers matter so much right now. Google's TPU architecture uses direct optical circuit switches to keep data in light form, reducing power consumption and latency. This advanced design requires cutting-edge HBM to function properly. As Google scales up its AI infrastructure, it needs more of these specialized chips than ever before, and the supply is limited.

Google's approach to AI infrastructure is not monolithic. While the company designs its own custom TPUs with Broadcom, it also purchases Nvidia GPUs for research and specialized tasks that TPUs cannot handle as efficiently. This dual approach creates demand across multiple memory suppliers. Google also works with Celestica for hardware integration, Lumentum for optical components, and Taiwan Semiconductor Manufacturing for advanced chip production. However, memory suppliers face the most acute demand pressures because HBM is the most supply-constrained component in the entire AI data center stack.

What Does This Mean for the AI Industry?

The timing of Google's capex increase is significant. The company plans to significantly increase capex spending in 2027 as well, suggesting that demand for memory chips will remain elevated for years to come. This multi-year commitment signals confidence in AI's importance to Google's future and creates a predictable revenue stream for Samsung and SK Hynix.

For investors and industry observers, the memory supply chain represents a critical chokepoint in the AI infrastructure boom. Samsung and SK Hynix have moved from being commodity memory makers to specialized suppliers of cutting-edge technology that powers the world's most advanced AI systems. As long as hyperscalers continue to invest heavily in AI infrastructure, both companies are likely to remain in high demand, though Samsung's larger share of Google's business and SK Hynix's dominance of Nvidia's memory needs create different growth trajectories for each company.