Global memory revenues hit a new monthly peak of $74.6 billion in July, an increase of 31.7% from the prior month and 2.8 percentage points above the 10-year seasonal average, according to UBS's July Memory Monthly report. The surge has prompted both UBS and Bernstein to anticipate steep contract price increases in coming quarters, even as the two houses disagree on how sustained the price rally will be.
The immediate beneficiaries of the stronger memory market include Micron Technology, Samsung Electronics and SK Hynix, which together supply the majority of the world's DRAM and NAND. SanDisk is also singled out as a major beneficiary because of its primary focus on NAND flash memory. UBS and Bernstein expect these producers to capture most of the contract price gains the firms now forecast will accelerate through the second half of 2026.
Breaking the results down by segment, DRAM recorded monthly sales of roughly $48.0 billion, up 27.7% month-on-month. Despite that increase, DRAM sales were about 8.0 percentage points below the 10-year seasonal average, indicating that demand is not uniformly strong across all end markets.
NAND told a different story. Sales in NAND rose to a record $25.8 billion, a 40.7% month-on-month jump and roughly 16.9 percentage points above the 10-year seasonal average. UBS interpreted the outperformance in NAND as evidence that AI-related storage demand is materially outpacing historical norms for the segment.
UBS framed the July data as a confirmation that the memory upcycle is strengthening amid accelerating AI-driven demand and continuing long-term agreement - LTA - negotiations. On contract pricing, UBS raised its forecast for DDR contract prices to climb 32% and 18% quarter-on-quarter in the third and fourth quarters of 2026, respectively. The firm also expects NAND contract pricing to increase 30% and 12% quarter-on-quarter across the same two quarters.
UBS's outlook for structural undersupply extends through at least the second quarter of 2028. That view is anchored in a projected 36.2% year-on-year increase in bit demand for 2027, which UBS expects will substantially outstrip 2027 supply growth of 19.3% year-on-year.
High-bandwidth memory - HBM - is a central pillar of UBS's demand thesis. The firm projects HBM demand to grow 90% year-on-year to roughly 33.1 billion gigabytes in 2026, followed by an additional 77% year-on-year increase to about 58.7 billion gigabytes in 2027, driven by continued AI accelerator deployments among hyperscalers. UBS's aggregate projections are sizable: total memory industry revenues of $992 billion in 2026 and $1.76 trillion in 2027.
Bernstein shares UBS's near-term bullishness on price direction, calling for a major contract price increase in the second calendar quarter of 2026 for both DRAM and NAND. However, Bernstein introduces a more cautionary view about the durability of the rally. The brokerage expects that demand destruction in the consumer segment will eventually occur and that the pace of price appreciation should narrow notably into the third quarter of calendar 2026.
Bernstein notes two temporary cushions that have so far limited downside pressure: pulled-forward consumer purchases and server demand that continues to absorb incremental supply allocated by producers. The firm treats both as short-term supports rather than permanent demand bases.
Both UBS and Bernstein emphasize the significance of Long-Term Agreement negotiations to current market dynamics. Bernstein characterizes LTAs as a structural buffer that can reduce the impact of a market correction and expects more clarity on these agreements as additional LTAs are signed. UBS likewise cites ongoing LTA negotiations as a contributor to the momentum behind the present upcycle.
The most pronounced divergence between the two firms concerns the expected duration of the cycle. UBS's structural undersupply thesis extends to at least mid-2028 and is rooted in the AI capital expenditure build-out. By contrast, Bernstein's price model envisions memory prices gradually peaking and beginning to normalize from the second half of 2027 into calendar 2028. While the two timelines overlap geographically, Bernstein's framework implies a nearer-term peak and a less elevated plateau than UBS's more aggressive scenario.
UBS is explicit about the risks that could undermine its bullish case. The firm highlights customer affordability and the sustainability of AI-related capital expenditure as the key uncertainties. In other words, hyperscaler spending discipline - rather than supply dynamics alone - is the major swing variable that is difficult to model from the memory industry side.
For market participants tracking the sector, upcoming quarterly contract price negotiations and any public commentary from major memory producers about LTA terms will provide the next important signals. Bernstein's forecast that the pace of price gains will narrow in the third quarter of 2026 establishes that period as a critical test of whether UBS's steeper pricing trajectory materializes.
Looking further out, the balance between 2027 DRAM bit demand and supply growth will be decisive. UBS projects bit demand growth of 36.2% year-on-year in 2027 against supply growth of 19.3% year-on-year. That disparity is the central variable in determining whether UBS's $1.76 trillion revenue projection for 2027 proves conservative or aspirational.
Sector context - The dynamics described in UBS's and Bernstein's reports affect not only memory manufacturers but also hyperscalers and enterprise customers that are driving AI infrastructure build-outs, plus consumer markets where substitution and pulled-forward purchases can temporarily distort seasonality. The negotiation outcomes on LTAs and the sustainability of hyperscaler capex plans will influence pricing, supplier margins and investment decisions across the memory ecosystem.