Bernstein goes against the grain to turn bullish on chip stocks: Market concerns over slowing memory ASP growth are overdone, with Seagate (STX.US) as top pick and SanDisk (SNDK.US) also favored.

Bernstein releases its latest research report, maintaining an overall constructive view on memory and HDD, arguing that the recent pullback has already over-discounted concerns about slowing price increases. On the U.S. stock side, Seagate (STX.US) remains the top pick, while Western Digital Corporation (WDC.US) and SanDisk (SNDK.US) are also rated "Outperform." Bernstein's FY2027 forecasts for all three are roughly 12% above market consensus. Memory and hard disk drive (HDD) stocks have fallen significantly from their June peak and have underperformed the broader market over the past month. In its latest report, Bernstein directly addresses the most mainstream bear argumentthat the slowing pace of DRAM and NAND price increases means the cycle has peaked or is near a top, to be followed by a sharp downturn. Its conclusion: it disagrees, for three reasons. The core pillar of the bear argument is the deceleration in sequential average selling price (ASP) growth. After unprecedented sequential increases in DRAM and NAND in the first half of calendar 2026, price gains in the third and fourth quarters are bound to slow significantly. This leaves investors seeing fewer earnings-upgrade "surprises," and based on a misreading of history, they judge this as a signal that a peak or near-peak has been reached, to be followed by a sharp recession. Bernstein rebuts this point by point, offering three reasons. Reason one: Double-digit sequential gains remain extreme by historical standards By any historical standard, a 15%20% sequential increase represents an extreme pricing environment. According to the report's calculations, when NAND industry ASP growth is at this level, it sits around the 90th95th percentile over the past 15 years; DRAM's forecast 10%15% growth in Q4 sits at the 76th89th percentile. In Bernstein's view, a price increase above 90% of historical samples signals strong demand, not weak demand. On the specific trajectory, TrendForce estimates and forecasts show: NAND flash contract prices rose 85%90% sequentially in Q1, 55%60% in Q2, 18%23% in Q3, and are expected to rise 15%20% in Q4; conventional DRAM contract prices rose 93%98% in Q1, 58%63% in Q2, 13%18% in Q3, and are expected to rise 10%15% in Q4. In other words, the deceleration is relative to the unprecedented jump in the first half; absolute pricing power remains very strong. The report also acknowledges that part of the slowdown stems from consumer price tolerance hitting its limit: smartphone brands collectively raised retail prices starting in the first half of 2026 to offset high memory costs, and some manufacturers have become more conservative in production plans; room for vendors to continue passing costs on to consumers is limited. But Bernstein believes that once memory prices return to more normal levels, pent-up demand is likely to recover, and AI-enabled high-end models will provide a solid demand floor for the high end of the market. An additional buffer for SanDisk is its customer mix: the share of bits the company directs to the data centerthe fastest-growing end marketis expanding rapidly, while the consumer and edge share has fallen markedlySanDisk's data center bits as a share of total shipments have risen from 12% in fiscal Q4 2025 to 38% in fiscal Q4 2026. Reason two: In absolute dollar terms, the increases remain at historical highs A 15%20% sequential increase is not just "a strong pricing environment"on a high base, the arithmetic also means it is extremely profitable for the entire memory industry. First, the absolute step size has barely slowed. NAND contract prices have risen more than fivefold over the past year; on this base, the 18%23% increase in Q3 equates to roughly 5 cents per GB, and the further 15%20% increase in Q4 equates to another roughly 5 cents. Measured in cents per GB, this absolute increment in Q4 remains at the 95th percentile of the past 15 yearsthe highest level in history prior to 2026the rate of increase has slowed, but the absolute quarterly price contribution remains at a historical peak. Second, relative to unit costs, this step size sits at the 95th97th percentile, pointing to considerable profitability. At the industry level, the forecast absolute ASP increments in Q3 and Q4 equate to roughly 97% and 104% of unit costs, respectively, both above the 95th percentile of the past 15 years. Since unit costs do not move with prices in the short term, nearly all of the incremental ASP flows directly to gross profit. DRAM shows the same dynamics. Bernstein's phrasing: the operating leverage across the memory industry is at a level unseen in more than a decade. Reason three: Price caps in long-term agreements are diluting headline increases Most global memory suppliers already have a substantial proportion of sales locked under long-term agreements (LTAs) extending through 2028. More than half of SanDisk's FY2027 bits and about two-thirds of FY2028 bits have been sold to eight data center and edge customers under a new business model (NBM) with price floors and ceilings; Micron has more than one-third of revenue locked through 2030 under supply assurance agreements (SCAs); Kioxia aims to cover half of its sales volume by 2028. The key point: what the market reads as "weakening demand" is precisely the purpose of such contract designconverting spot prices into long-term contract prices, rather than a genuine demand slowdown. Bernstein uses sensitivity analysis to illustrate this mechanism: assuming market prices have reached the price ceiling of each contract, at a 40% LTA penetration rate, a 30% rise in market prices translates into only an 18% actual ASP increase; at 50% and 60% penetration, this falls to 15% and 12%, respectively. In other words, as more and more industry bits shift to contract prices, headline price growth will mechanically slow even if underlying demand is unchanged. Company-level commentary supports this interpretation: Micron disclosed on its fiscal Q3 2026 earnings call that the price ceilings on its largest contracts are set near the level of market prices in the second quarter of calendar 2026; TrendForce likewise notes that LTAs signed with U.S. cloud customers are limiting the magnitude of server DRAM price increases. It should be noted that Bernstein itself retains some caution about such contracts: the report points out that deposits or financial guarantees remain small relative to contract value, so it questions the enforceability of LTAs; but it also acknowledges that this generation of LTAs is indeed more protective than in the past. Three implications for SanDisk Notably, the three industry-level developments above are particularly favorable for SanDisk, because its LTA coverage is above the industry average, its product mix is tilting toward data centers, and it is a rare pure-play NAND name among global memory suppliers. First, Bernstein's forecast for SanDisk's actual ASPs already incorporates the LTA drag: it expects sequential increases of 9% and 9.5% in fiscal Q1 and Q2 2027, respectively, both still close to SanDisk's own 89th percentile over the past 15 years; the absolute step of about 3 cents/GB is above its 90th percentile. In other words, even after LTA dilution, SanDisk's actual ASP increases remain at historical highs on an absolute cents basis. Second, operating leverage means the increases flow directly to earnings: the ratio of absolute ASP increments to unit costs in F1Q27 and F2Q27 sits at the 94th and 96th percentiles of the past 15 years. Bernstein acknowledges that as gross margin expansion approaches its ceiling, SanDisk's sequential EPS growth will slow from above 60% in the prior quarter to below 20%this change in growth profile itself explains part of the recent valuation de-rating; but the dollar contribution from pricing remains at historical highs, and from here on, absolute increases matter more than headline percentage increases. Third, LTAs are downside protection rather than a growth headwind. Price floors can anchor the downside near current levelsBernstein estimates that on average 53% of SanDisk's shipped bits in fiscal 2027 to 2030 are under LTAs, implying a floor ASP of $0.26/GB, only 19% below its most recent quarter's blended ASP. Customers' willingness to pay is also a signal: data center customers have committed $16.5 billion in deposits and guarantees against SanDisk's minimum LTA revenue of $93.9 billion (17.6% coverage); for Micron, $32 billion against $100 billion (32% coverage, with 35% of revenue covered by SCAs through 2030 and management targeting about 50% thereafter). Customers' willingness to lock in price floors for three to five years at levels near or slightly below current prices, and to pledge high-coverage cash for it, indicates that under the base case they do not believe a price cliff will occur in 2027 or 2028. Micron and other suppliers also mention that newly signed agreements carry higher pricing and terms extending to 2031. On valuation, Bernstein believes SanDisk's multiple should reflect a structurally higher earnings floor. Historically, memory stocks at cycle peaks have typically traded at forward multiples ranging from the low to high single digits (reflecting the market's assumption that peak earnings are unsustainable); SanDisk currently trades at 5.2x FY2028 earnings, at the low end of this historical cycle-peak range, indicating the market is betting the cycle is about to turn. But the report argues this cycle deserves a multiple above that range, because price floors close to current prices substantially weaken the assumption of a "very low earnings floor" and extend the durability of the cycle peak. HDD simply doesn't have this problem While memory stocks are troubled by "slowing increases," HDD stocks do not have this problemin fact, HDD prices could accelerate further from here. Bernstein's judgment: even for those still skeptical of the foregoing arguments, HDD is at an earlier stage of the upcycle relative to memory. Nearline HDD average price per EB is up only 13.6% year over year (NAND and DRAM prices rose about 400% over the same period), so it has more room to run "higher for longer," and could accelerate from the pace of the past two quarters.