Samsung Pursues Multi-Year Memory Deals With Google and Microsoft as Big Tech Races to Lock In AI Chip Supply
Samsung Electronics is negotiating three-to-five-year memory supply agreements with Google and Microsoft that could include more than 10 billion dollars in prepayments, as hyperscalers move to secure chip capacity amid a structural AI-driven shortage.
Samsung Electronics is in discussions with Google and Microsoft over long-term memory supply agreements spanning three to five years, a sharp departure from the quarterly and annual contracts that have traditionally defined the semiconductor memory market. The deals under negotiation could include more than 10 billion dollars in upfront prepayments from the two hyperscalers, according to industry reports, underscoring the urgency with which Big Tech is moving to secure access to one of the most constrained inputs in the AI hardware supply chain.
From Commodity to Strategic Asset
The proposed contract structure would fix volumes over a multi-year period while linking pricing to spot market levels, with adjustments triggered if prices move beyond predefined ranges. Large upfront payments would be offset against committed purchase volumes, and any shortfall in purchases would be deducted from the prepayment balance. The approach differs fundamentally from the arrangements that have governed memory procurement for decades, where buyers typically negotiated prices on a quarterly cycle with limited forward commitment.
Samsung Co-CEO Jun Young-hyun has stated that quarterly or annual agreements are no longer sufficient to meet demand in the current environment, signaling the company’s intent to pursue longer-duration contracts that provide visibility into demand for three years or more. For Samsung, the benefit is clear: guaranteed demand enables more confident capital investment and helps prevent the sharp price collapses that have historically plagued the cyclical memory industry.
A Market Reshaped by AI
The push toward multi-year supply agreements reflects the extraordinary demand pressure created by AI infrastructure buildouts. Samsung posted record Q4 2025 operating profit of approximately 20 trillion won, or 13.82 billion dollars, a threefold year-over-year increase driven almost entirely by memory chip pricing. DRAM contract prices increased 313 percent in Q4 2025 compared to the same quarter the prior year, and industry analysts at Macquarie Equity Research project the global DRAM market will reach 311 billion dollars in 2026, roughly six times its 2023 level.
The supply constraint is particularly acute in high-bandwidth memory, the stacked DRAM technology that serves as the primary memory interface for AI accelerators. Samsung commenced mass production of HBM4, the fourth generation of the technology, in February 2026, with 12-layer stacks delivering speeds up to 13 Gbps and bandwidth of 3.3 TB per second. The company has already secured HBM4 supply commitments from NVIDIA for its Vera Rubin platform, from AMD for its Instinct MI455X accelerators and sixth-generation EPYC processors, and from OpenAI for its custom Titan AI chip.
Competitors Follow Suit
Samsung is not alone in pursuing longer-term arrangements. Micron Technology disclosed its first five-year strategic customer agreement during its fiscal Q2 2026 earnings call, and has stated that its entire HBM production capacity for the remainder of calendar 2026 is already fully committed under binding contracts. SK Hynix, which held approximately 53 percent of the overall HBM market in 2025, is also in discussions with hyperscale customers, though it has not publicly confirmed multi-year deal structures.
The simultaneous pursuit of long-term contracts by all three major memory suppliers suggests that both sides of the market view the current shortage as structural rather than cyclical. Google, Microsoft, Meta, and Amazon are each expanding custom AI chip programs — including Google’s tensor processing units, Microsoft’s Maia accelerators, Meta’s MTIA chips, and Amazon’s Trainium processors — all of which require substantial allocations of high-bandwidth and server-grade DRAM.
Risks and Trade-Offs
Multi-year memory contracts carry risks for both parties. Buyers face the possibility of overpaying if demand softens or alternative technologies emerge. Suppliers risk locking in prices that trail a rising market. The memory industry’s last attempt at long-term supply agreements, in 2019, faltered because contracts lacked enforceability and pricing mechanisms that could adapt to market swings.
The current generation of proposed agreements attempts to address those shortcomings through binding commitments with prepayment offsets and spot-linked pricing corridors. Whether those mechanisms prove durable across a full market cycle remains to be tested. Industry sources indicate that memory suppliers are expected to finalize long-term supply agreements with major Big Tech firms in the first half of 2026.
For the broader technology industry, the shift toward multi-year memory procurement marks a significant evolution. Memory has historically been treated as an interchangeable commodity, subject to rapid price swings and short-term negotiation. The emergence of binding, prepaid, multi-year contracts signals that AI-era hyperscalers now view memory capacity as a strategic resource worth securing years in advance — and are willing to commit billions of dollars to ensure they are not left without it.