Driving Micron's $100 billion surge of 19%—This UBS research report identifies "the biggest cognitive gap in AI supply chain valuation"

The AI supply chain is being repriced by long-term agreements. The surge in Micron is just the most visible manifestation; the entire industry chain from SK Hynix and Samsung to Nvidia and OpenAI is entering a new valuation stage driven by "prepayment, locked volume, and locked price".

On May 26, UBS Securities analyst Timothy Arcuri sharply raised Micron’s target price from $535 to $1625, maintaining a buy rating. Micron surged over 19% in pre-market trading, with its market cap increasing by more than $140 billion in one day, reaching over $1 trillion for the first time.

The report points out that up to 30% of DDR capacity will soon be locked into long-term agreements at slightly below current prices, and hyperscale computing companies have already locked in about 60-70% of DDR5 chips via "enhanced" LTAs, ensuring procurement volumes for Micron and others. On the demand side, DRAM shortages are expected to last until Q2 2028, and NAND shortages until Q4 2027.

What truly drew attention in this research report is the switch in valuation methodology. UBS is abandoning the traditional segmented valuation approach and instead adopting a future profit discount P/E framework, based on the assumption that the profit volatility in the memory industry is being significantly diminished by long-term agreements.

This is also the biggest cognitive gap in the AI supply chain: the market still prices memory manufacturers as cyclical stocks, but buyers are locking in supply with long-term contracts, and sellers are reshaping profit floors using these contracts. If this mechanism holds, Micron, SK Hynix, and Samsung will not just experience an upcycle, but face a structural revaluation in their valuation systems.

The Fatal Flaw of Traditional Memory: Unpredictable Profits

Before understanding long-term agreements (LTA), first understand why memory companies have historically been so undervalued.

The core of memory manufacturers’ business model is high fixed costs + spot price pricing. In upcycles, products sell, prices soar, and profits multiply; in downcycles, customers destock, prices collapse, and a year's profit can be wiped out by half a year of losses.

This isn’t a management problem—it’s industry structure. Buyers defer purchases when prices are high and restock when prices are low, which naturally amplifies price volatility.

The market logic for pricing such companies isn’t “how much will this company make next year,” but “where are we in the cycle now”—low multiples at the top because profits aren’t sustainable, low multiples at the bottom because profits turn negative.

No matter the cycle stage, the market habitually applies a discount.

This is why memory stocks have been long undervalued, and the core assumption UBS is trying to overturn in this report.

Long-term Agreements: Turning “Chasing the Cycle” Into “Locking the Cycle”

UBS puts forth a concrete number: by 2027, about 20-30% of DDR bit shipments in the industry will be covered by "enhanced" long-term agreements—20% for Micron, 18% for SK Hynix, and 30% for Samsung.

More importantly, about 60%-70% of industry Server DDR5 capacity is already locked in by hyperscale cloud providers via enhanced LTAs.

This isn’t the symbolic "framework agreement" of the past—simply agreeing on volumes while prices float with markets. This time, LTAs are different: First, terms are longer, usually 3-5 years. Second, both volume and price are locked; some prices are set slightly below market, while the rest float. Third, breach penalties are higher, buyers bear prepayment arrangements and capital expenditure commitments, making contract termination harder.

What does this mean for Micron?

It means that even when the next memory downcycle arrives, revenue from the fixed-price portion won’t collapse with spot prices. The floating price portion may fall, but the fixed portion props up the profit baseline.

UBS calculations show these agreements can reduce DDR price swings from peak to trough by about half.

Halving may not sound big, but in an industry where prices can drop 50%-60% in a year, this means EPS volatility is significantly reduced, and profit predictability is transformed.

LTAs turn the memory business from "chasing cycles" into "locking cycles".

2029: The Real Test of This Logic

UBS’s EPS forecasts: Micron’s 2027/2028/2029 calendar-year EPS will be $155/$167/$117, respectively.

The first two years are easy to understand—AI demand explodes, memory prices are high, and the company earns a lot. Upward revisions for these years are uncontroversial.

The real divergence is in 2029.

In UBS’s model, 2029 isn’t based on a “perpetual boom” premise. The model includes a notable DRAM price correction, with floating price DDR dropping about 50%—a real downcycle.

The result: Even under this pessimistic scenario, Micron’s 2029 EPS stays above $100.

This is the underlying basis for changing the valuation framework.

New $1625 target price calculation: Non-GAAP 2029 EPS about $117, discounted to 2028 about $105, multiplied by a 15x P/E.

15x isn’t a software stock multiple, and is even lower than TSMC’s current ~20x. But it’s much higher than the 5-8x cyclical stocks get during expected downcycles.

The core bet of this valuation logic: If profits in a downcycle still reach $100, Micron is no longer just a company with high peak-cycle earnings, but one with a high-profit floor even in downturns.

SK Hynix’s Signal: This Isn’t Just Micron

If UBS’s logic only applies to Micron, it would merely be a company-level story. But this trend is happening across the industry.

SK Hynix’s LTA coverage is about 18%, Samsung’s about 30%. 60%-70% of Server DDR5 is locked in by cloud players, including Microsoft, Google, Amazon, Meta signing agreements.

Nomura Securities on May 15 sharply raised Samsung and SK Hynix target prices more than 2x, using logic closely parallel to UBS—the current ~6x forward P/E for these two companies reflects high cyclical risk premiums, which doesn’t match reality. Nomura holds that Samsung and Hynix’s valuation risk premium should converge toward TSMC, instead of continued cyclical discounts.

On the demand side, the "FOMO paradox" in the post: LTAs locking in shipment volume means less supply left for the spot market. Buyers without LTAs must either sign (at higher prices) or scramble in spot markets with clear premiums for limited supply. Both outcomes maximize memory manufacturers’ profits.

Those who sign LTAs earn fixed prices, those who don’t must pay premiums—this is a self-reinforcing loop.

UBS expects DRAM shortages to persist until at least Q2 2028 (previously Q4 2027), while NAND shortages to last through Q4 2027 (previously Q3 2027).

"LTA-ization" of the Industry Chain: A Systemic Reshaping from Chips to Computing Power

Zooming out, the LTA trend isn’t just a memory industry story.

Nvidia's supply commitments to suppliers reach $95.2 billion, up 89% from three months ago, with the CFO explicitly stating they “strategically lock inventory and capacity”; Broadcom announced its supply chain is locked to support its $100 billion goal; AMD supply commitments exceed $21 billion, doubling quarter-on-quarter.

Recently, OpenAI launched its "Guaranteed Capacity" product; enterprise customers can sign 1-3 year compute usage agreements to lock in access, with longer terms yielding bigger discounts. For the first time, a large model company offers such a product on the demand side—and its upstream Nvidia is using the same logic to lock TSMC capacity.

The entire AI supply chain, from wafer capacity to memory shipments to inference compute, is undergoing the same pattern: prepayment, volume lock, and LTA coverage.

Buyers not only fear price hikes, but also fear shortages; sellers not only want price hikes, but want order visibility before massive capex. LTAs write both parties’ anxieties into contracts, turning uncertainty into a calculable profit curve.

JPMorgan analyst Jay Kwon summed up this trend: "LTAs pave the way for memory manufacturers to adopt new valuation frameworks."

The Window for Switching Valuation Frameworks

Back to the original question: Why did UBS raise Micron’s target price by 200%?

Not because memory prices will double again, but because the nature of profitability has changed for memory companies.

From "chasing cycles" to "locking cycles," from "unpredictable profits" to "EPS still above $100 even in downturns," all of this is driven by the expanding scope of LTA coverage.

Once the profit curve transforms from cyclical to relatively stable, the valuation framework should switch—not based on “where are we in the cycle,” but on “how much stable earnings in the next three years” to assign a P/E.

This is the “biggest cognitive gap in AI supply chain valuations” revealed by this report: The market still uses old cyclical stock frameworks to price these companies, while their profit structures have quietly changed.

Nomura’s call on Samsung and Hynix and UBS’s call on Micron both point to the same conclusion: once LTAs cover enough shipments, cyclical stock discounts lose their basis, and revaluation toward growth semiconductor stocks is only a matter of time.

Risk warnings and disclaimersThe market involves risk; investment requires caution. This article does not constitute personal investment advice, nor does it consider individual users' specific investment objectives, financial condition, or needs. Users should assess whether any opinions, views, or conclusions in this article suit their particular circumstances. Investing based on this is at your own risk.