The AI supply chain is being repriced by long-term agreements. Micron Technology's surge is just the most visible sign; the entire industry chain—from SK Hynix and Samsung to Nvidia and OpenAI—is entering a new valuation phase driven by "prepayment, volume lock-ins, and price lock-ins."
On May 26, UBS analyst Timothy Arcuri significantly raised Micron's target price from $535 to $1,625, maintaining a Buy rating. Micron's pre-market surge exceeded 19%, adding over $140 billion in market value in a single day and pushing its total market capitalization above $1 trillion for the first time.
The report notes that up to 30% of DDR capacity will soon be locked under long-term agreements at prices slightly below current levels. Hyperscale computing companies have already secured about 60-70% of DDR5 chips through "enhanced" LTAs, ensuring procurement volumes for manufacturers like Micron. On the demand side, DRAM supply shortages are expected to persist until Q2 2028, while NAND shortages are projected to last until Q4 2027.
What truly captured attention was the shift in valuation methodology. UBS abandoned traditional sum-of-the-parts valuation in favor of a discounted future earnings framework using a P/E multiple. The underlying assumption is that long-term agreements are significantly reducing profit volatility in the memory industry.
This represents the biggest valuation gap in the AI supply chain today: the market still prices memory manufacturers as cyclical stocks, while buyers are locking in supply with long-term contracts, and sellers are using these contracts to reshape their profit floors. If this mechanism holds, Micron, SK Hynix, and Samsung are not merely experiencing a cyclical upturn but undergoing a structural revaluation of their entire valuation framework.
**The Achilles' Heel of Traditional Memory: Unpredictable Profits**
To understand long-term agreements, one must first grasp why memory companies have historically been valued so low.
The business model of memory manufacturers is essentially high fixed costs combined with spot pricing. During an upturn, products sell out, prices skyrocket, and profits multiply. During a downturn, customers destock, prices collapse, and a year's profits can be wiped out by six months of losses.
This isn't a management issue but an industry structural problem. Buyers delay purchases when prices are high and restock heavily when prices are low, amplifying the natural volatility of memory prices through their behavior.
The market's pricing logic for such companies isn't "how much will this company earn next year" but "where are we in the cycle?"—low multiples at the peak due to unsustainable profits, and low multiples at the trough due to anticipated losses.
Regardless of the cycle stage, the market habitually discounts these stocks.
This is why memory stocks have long been undervalued, and it's the core assumption that this UBS report seeks to overturn.
**LTAs: Transforming "Chasing the Cycle" into "Locking the Cycle"**
UBS's report provides specific figures: by 2027, approximately 20%-30% of industry DDR bit shipments will be covered by "enhanced" long-term agreements. This includes about 20% for Micron, 18% for SK Hynix, and 30% for Samsung.
More importantly, about 60%-70% of the industry's Server DDR5 volume has already been locked in by hyperscale cloud providers through enhanced LTAs.
These are not the symbolic "framework agreements" of the past—vague commitments on volume with prices floating with the market. The new LTAs differ in three key ways: First, they have longer terms, typically 3 to 5 years. Second, they lock in both volume and price, with a portion fixed slightly below current market levels and the remainder floating. Third, they carry higher breach costs, with buyers committing to prepayments and capital expenditure support, making contract cancellation significantly more difficult.
What does this mean for Micron?
It means that even when the next memory downturn arrives, revenue from the fixed-price portion won't collapse with spot prices. The floating portion may decline, but the fixed portion establishes a profit floor.
UBS estimates that such agreements could reduce the peak-to-trough price volatility of DDR by about half.
A 50% reduction might not sound dramatic, but for an industry where prices have historically dropped 50%-60% within a year, it means the volatility center of EPS is significantly elevated, and profit predictability undergoes a qualitative change.
LTAs are transforming the memory business from "chasing the cycle" to "locking the cycle."
**2029: The Crucial Test for This Logic**
UBS's EPS projections for Micron are $155, $167, and $117 for calendar years 2027, 2028, and 2029, respectively.
The first two years are straightforward—explosive AI demand and high memory prices naturally lead to strong earnings. Upward revisions for these years are unlikely to spark much debate.
The real divergence lies in 2029.
UBS's model does not assume "perpetual prosperity" for 2029. It incorporates a significant DRAM price correction, with floating-price DDR declining by about 50%—a genuine downturn scenario.
The result: even in this pessimistic scenario, Micron's 2029 EPS remains above $100.
This is the foundational basis for the valuation framework shift.
The new $1,625 target price is calculated as follows: 2029 non-GAAP EPS of approximately $117, discounted to 2028 at about $105, multiplied by a 15x P/E multiple.
A 15x multiple isn't software-stock territory, and it's even lower than TSMC's current ~20x valuation. However, it's significantly higher than the 5-8x multiples traditional cyclical stocks typically command during "downturn expectations."
The core bet of this valuation logic is: if profits can remain at $100 during a downturn, Micron is no longer just a company with "high peak-cycle earnings" but one with a robust profit base even during cyclical declines.
**SK Hynix's Signal: This Isn't Just Micron's Story**
If UBS's logic applied only to Micron, it would merely be a company-specific narrative. However, this trend is unfolding across the entire industry.
SK Hynix's LTA coverage is around 18%, and Samsung's is about 30%. The fact that 60%-70% of industry Server DDR5 is locked by cloud providers reflects simultaneous agreements with Microsoft, Google, Amazon, and Meta.
On May 15, Nomura Securities more than doubled its target prices for Samsung and SK Hynix, using logic highly consistent with UBS's—the current ~6x forward P/E for both companies implies the high-risk premium of a traditional cyclical stock, which no longer aligns with reality. Nomura's view is that the valuation risk premium for Samsung and Hynix should converge toward TSMC's, rather than remaining discounted as historical cyclical stocks.
From the demand side, the "FOMO paradox" emerges: LTAs lock in shipment volumes, meaning less supply remains for the spot market. Buyers without LTAs are forced either to sign agreements (locking in volume at higher prices) or to compete for limited supply in the spot market at a significant premium. Both outcomes maximize profits for memory manufacturers.
Companies with LTAs earn fixed prices, while those without are forced to pay premiums—a self-reinforcing cycle.
UBS expects DRAM supply shortages to persist at least until Q2 2028 (previously Q4 2027), while NAND shortages are projected to last until Q4 2027 (previously Q3 2027).
**Industry-Wide "LTA-ization": A Systemic Restructuring from Chips to Compute**
Zooming out, the LTA trend is not confined to the memory industry.
Nvidia's purchase commitments to suppliers have reached $95.2 billion, a staggering 89% increase from three months ago, with its CFO explicitly stating the company is "strategically locking in inventory and capacity." Broadcom announced it has "locked in the required supply chain" to support its $100 billion target. AMD's purchase commitments exceed $21 billion, doubling quarter-over-quarter.
Recently, OpenAI launched its "Guaranteed Capacity" product, allowing enterprise customers to sign 1-3 year compute usage agreements to lock in access rights, with larger discounts for longer terms. This marks the first time a major AI company has introduced such a product on the demand side—while its upstream partner Nvidia is already using the same logic to lock in TSMC's capacity.
The entire AI supply chain, from wafer capacity to memory shipments to inference compute, is undergoing the same transformation: prepayment for volume lock-ins, covered by long-term agreements.
Buyers fear not just price hikes but also the inability to secure supply. Sellers seek not only higher prices but also order visibility ahead of massive capital expenditures. LTAs codify both parties' anxieties into contracts, transforming uncertainty into calculable profit curves.
JPMorgan analyst Jay Kwon accurately summarized this trend: "LTAs are paving the way for memory manufacturers toward a new valuation framework."
**The Valuation Framework Shift Window**
Returning to the initial question: Why did UBS raise Micron's target price by 200%?
It's not because memory prices will double again, but because the nature of memory companies' profits has changed.
The shift from "chasing the cycle" to "locking the cycle," from "unpredictable profits" to "EPS still exceeding $100 even in a downturn," is driven by the expanding coverage of LTAs.
Once the profit curve transitions from cyclical to relatively stable, the valuation framework should shift—no longer pricing based on "where we are in the cycle," but applying a P/E based on "how much this company will stably earn over the next three years."
This is precisely the "biggest valuation gap in the AI supply chain" revealed by this report: the market continues to price these companies using the old cyclical stock framework, while their profit structures have quietly transformed.
Nomura's view on Samsung and Hynix, and UBS's perspective on Micron, point to the same conclusion: once long-term agreements cover a sufficient portion of shipments, the rationale for cyclical stock discounts disappears, and valuation convergence toward normal growth-type semiconductors is only a matter of time.
