The AI boom is impacting something deeper than memory prices: it’s transforming the balance of power between manufacturers and customers. Samsung, SK Hynix, Micron, and other major suppliers are closing multi-year agreements that ensure supply through the end of the decade. In some cases, these include deposits, financial commitments, take-or-pay clauses, and minimum prices. DigiTimes estimates that these new contracts involve approximately $38 billion in advances, deposits, and guarantees, although this privileged position could start to weaken once new production capacity begins coming online around 2028 and 2029.
The key points of the new memory market in 30 seconds
- The shortage driven by AI has given manufacturers unprecedented negotiation power.
- Micron has already signed 16 strategic agreements through 2030, with $22 billion in deposits and commitments planned.
- Contracts include guaranteed volumes, minimum prices, and take-or-pay obligations.
- Samsung, SK Hynix, and other producers are also securing long-term orders.
- The balance could shift by 2029 when new factories start adding actual capacity to the market.
This shift is significant because the memory industry has historically been subject to volatile, cycle-driven swings. When production exceeds demand, prices fall, margins tighten, and manufacturers reduce investments. Artificial intelligence is temporarily enabling a more predictable model, replacing some of that volatility.
Micron demonstrates how much the business has changed
The clearest example is Micron.
In June, the company confirmed it had signed 16 Strategic Customer Agreements (SCAs) with clients in data centers, consumer electronics, and automotive sectors. Most are approximately five-year contracts, from 2026 through late 2030.
These are not mere letters of intent.
The contracts include binding commitments to specified volumes, and a large portion operate under a take-or-pay model: the customer commits to purchasing the agreed quantity even if their needs change later.
Micron states that these 16 agreements account for about 20% of its DRAM volume and a third of NAND volume during the contracted period. Fourteen of them also include contractual minimum revenues approaching $100 billion over their remaining lifespan.
Pricing structures are also remarkable for an industry known for fluctuations.
Major agreements often establish:
- a maximum price for certain products;
- a minimum price during the contract term;
- guaranteed quantities;
- deposits or financial commitments from the customer.
Micron expects to receive around $22 billion in deposits and other financial commitments, with roughly $18 billion in cash deposits.
This is a clear indicator of shifting power: instead of competing solely to sell memory, manufacturers are getting buyers to advance billions to secure future capacity.
It’s no longer just memory; it’s about guaranteeing supply
To grasp this, consider what’s happening with AI.
New systems require extraordinary amounts of:
HBM + DDR5 + LPDDR + enterprise NAND + high-speed storage
AI-focused GPUs depend heavily on HBM memory, but servers also need large quantities of traditional DRAM.
For example, Samsung has confirmed strategic agreements to supply HBM4 to AMD and DDR5 solutions for upcoming EPYC processors and Helios systems.
Additionally, Samsung collaborates with OpenAI, and the potential demand for Stargate-related memory could reach up to 900,000 DRAM wafers per month, according to the company.
Micron has also signed multiyear agreements for memory and storage with Anthropic.
In this landscape, for hyperscalers or model developers running out of memory, delays of months in data center deployment—costing billions—are a real risk.
Hence, the product now isn’t just the chip; it’s security of supply.
The $38 billion figure needs context
DigiTimes estimates that memory manufacturers collectively hold around $38 billion in advances, deposits, and guarantees tied to supply contracts.
This figure should be interpreted cautiously since it combines different financial mechanisms and not all relate exclusively to DRAM.
For example, SanDisk, primarily a NAND manufacturer, has its own multi-year contract model called New Business Model (NBM).
SanDisk announced in August that it had signed ten such agreements after adding five new contracts in its fiscal Q4. Their financial statements also record advances and deposits linked to these contracts.
Therefore, it’s more accurate to refer to the group as memory manufacturers rather than just DRAM producers when citing the DigiTimes aggregate.
What is clear is the market’s direction: providers are replacing some traditional sales with multi-year commitments that reduce exposure to future price drops.
The memory cycle begins to look different
For decades, DRAM and NAND have been highly cyclical businesses.
The pattern was well known:
- Demand increases.
- Manufacturers invest in more capacity.
- Supply surpasses demand.
- Prices plummet.
- Production and investment are cut.
- The cycle repeats.
New agreements are trying to partially break this cycle.
If a producer has a minimum price and a customer committed to certain volumes, the impact of spot market price drops is lessened.
Micron states that the minimum prices in some SCAs could help maintain high gross margins even during downturns.
This doesn’t mean memory stops being cyclical, but a growing portion of supply might become more resilient to extreme fluctuations.
The duration of manufacturers’ power is under review
The big question is: how long will this last?
Currently, supply remains tight. Micron expects conditions to stay tight beyond 2027 and lacks clear visibility on when capacity will match demand growth.
Nevertheless, industry investments are massive.
SK Hynix just approved 54.3 trillion won (about $38 billion) for new DRAM, HBM, and NAND factories in Yongin and Cheongju. They plan to open M17 cleanrooms by late 2028 and Y2 by June 2029.
Micron is also expanding in the U.S., with a new DRAM plant in Idaho expected to start wafer production by mid-2027, and a second facility targeted for late 2028.
The clear takeaway is: 2028 and 2029 could be decisive years for balancing supply and demand.
If these new factories come online as planned and AI growth slows down, customers may regain bargaining power.
If demand continues soaring faster than capacity, the current model of pre-booked contracts could persist longer.
It’s too early to say 2029 will be the exact moment manufacturers lose their power, but it’s a likely period when this exceptional situation will start to be tested.
Customers are indirectly financing expansion
Another notable financial aspect is the impact of advance payments.
When a customer pays billions upfront to secure capacity, it provides liquidity to the manufacturer before the product is delivered.
Micron clarifies that deposits are gradually returned during the second half of the contracts and are recorded as financing flows, not immediate revenue.
Still, these advances give manufacturers a significant advantage.
They need enormous investments—building new factories, acquiring equipment, developing advanced nodes—at a time when demand is booming.
Customers guarantee future demand and provide liquidity, while manufacturers gain revenue visibility and mitigate the risk of underutilized capacity.
Both sides benefit from a sound economic rationale.
What if capacity exceeds demand in 2029?
This will be the true test for the new model.
Semiconductor factories take years to build. Decisions made during shortages might only pay off once the market has shifted.
If Samsung, SK Hynix, and Micron all add substantial capacity simultaneously, and demand grows less than expected, the industry’s old problem—the glut—could reemerge.
Minimum price contracts would help protect revenues, but not all future capacity might be covered under the same terms.
And customers will learn too.
If multiple suppliers are capable of competing for large orders, it becomes harder for manufacturers to demand multibillion-dollar deposits or favorable conditions.
Here’s the paradox: the extraordinary profits driven by shortages are funding the very factories that could eventually eliminate those shortages.
Artificial intelligence has made memory one of the most strategic assets in data centers and given manufacturers unprecedented commercial power.
But industry history advises caution.
Today, Samsung, SK Hynix, and Micron have clients willing to advance billions to secure chips through 2030. When these new plants begin production later this decade, we will see if this marks a permanent shift in the memory business or if it’s simply the most extraordinary cycle of scarcity in its history.
Frequently Asked Questions
How much money have customers prepaid to memory manufacturers?
DigiTimes estimates around $38 billion in advances, deposits, and guarantees tied to supply contracts. The figure combines different companies and financial mechanisms, so it shouldn’t be read as exclusively DRAM-related.
What agreements has Micron signed?
Micron announced 16 Strategic Customer Agreements, generally extending through 2030. They account for about 20% of its DRAM volume and one-third of NAND during that period.
Why do customers pay upfront?
Because securing memory has become essential for deploying AI servers and data centers. Deposits reduce the risk of capacity shortages when supply remains tight.
Will memory prices fall in 2029?
That’s uncertain. New capacity from Samsung, SK Hynix, and Micron will start entering the market between 2028 and 2029, but how prices evolve depends on whether production outpaces or lags behind AI demand growth.
via: Digitimes
