Memory chips are currently the biggest battleground for market bulls and bears, and Goldman Sachs' European TMT expert, Sean Johnstone, has shared his assessment.

In a recent report, Johnstone noted that orders for HBM (High Bandwidth Memory) and DRAM are sold out through 2027, with long-term agreements (LTAs) featuring price floors. Gross margins remain in the mid-to-high 70% range, with structural AI demand expected to keep supply tight until 2028. This forms the foundation for the bullish argument.

However, bears are focused on a different dimension: the "second derivative" of price increases. Johnstone pointed out that the month-over-month price growth rate is decelerating, with a peak expected between the second and third quarters of 2027, followed by a mild and modest decline. Additionally, once gross margins hit the mid-80% range, customers begin redesigning their systems to reduce dependency on memory—a natural market feedback mechanism.

Nvidia's actions confirm this logic. Johnstone mentioned that Nvidia is evaluating a solution using fewer HBM stacks for its next-generation "Vera Rubin" superchip. The reason is straightforward: memory costs now account for approximately 62% of Vera Rubin's total bill of materials (BOM), with the CPU-side SOCAMM2 memory module's cost share even exceeding that of the GPU-side HBM4. The significant cost pressure is forcing Nvidia to actively seek alternative paths.

For conservative investors, the strategy involves compressing valuation multiples—from 5x down to 2-3x. Johnstone emphasized that this is not a judgment that the cycle is over, but rather a belief that "the easiest scarcity beta has already been captured." He also posed a counter-question: bears need to consider whether they are shorting the fundamentals or merely shorting the momentum.

Additionally, Johnstone noted that the Wall Street Journal has reported Apple is testing memory chips from Chinese manufacturer ChangXin Memory Technologies (CXMT) for multiple product lines, including iPhones and MacBooks, in an effort to alleviate AI-driven memory supply pressure.

Absolute prices are still rising, but remaining upside depends on two variables

Johnstone stated that absolute memory prices will continue to rise until mid-2027. However, the subsequent potential depends on two factors:

- The durability of LTA price floors—whether long-term agreements can truly support the price bottom.
- Whether the HBM mix can offset the normalization pressure on traditional DRAM/NAND—i.e., whether the premium from high-end products can compensate for price declines in commodity memory.

Notably, Elon Musk has recently stated that AI demand is growing at about 200%, while supply growth is only about 20%. This data reinforces the structural bullish argument. However, Johnstone pointed out that the market's current pricing focus has shifted—investors are now pricing in "slowing price growth," with actual capacity releases not expected until late 2027 to 2028, and concentrated mainly in HBM rather than traditional memory.

Software and other tech sectors: Goldman Sachs' latest views

Beyond memory, Johnstone also reviewed several other tech topics currently in focus.

The software sector is showing clear divergence. Goldman Sachs believes that software is no longer trading solely as an "AI loser" factor. Data infrastructure and developer tools platforms (e.g., NET, PLTR, TEAM, TWLO) have rebounded from their lows, as the market is willing to give these companies credit for incremental AI-driven consumption and new workloads. In contrast, pure application software vendors (e.g., HUBS) remain under pressure, as investors need hard evidence that AI is expanding, rather than replacing, their core revenues.

Goldman Sachs software analyst Gabriela holds an incrementally positive view on SNOW and PANW, but an incrementally negative view on ADBE, INTU, and WDAY, citing potential pressure on top-of-funnel demand and core workflows for the latter group.

Regarding AI model usage (open-source vs. closed-source), earnings calls from Pinterest and Duolingo both signaled a clear trend: open-source models are accelerating their replacement of closed-source models. Pinterest management stated that the per-transaction cost of open-source models is less than 8% of comparable closed-source models. Duolingo CEO Luis von Ahn remarked, "As long as the quality is roughly comparable, we switch to open-source models because they are much cheaper," and predicted that the model mix will continue to shift toward open-source.

On AI financing, AI-related equity financing has accounted for about 40% of total U.S. stock issuance this year. Goldman Sachs estimates that capital expenditures for the hyperscale cloud providers will reach $1.1 trillion by 2027, exceeding their operating cash flow by approximately $150 billion, with free cash flow not turning positive until 2028. Goldman Sachs credit strategists estimate that hyperscalers could cover about 35% of their 2027 capital expenditures through debt financing, corresponding to a global bond issuance of roughly $400 billion. Meanwhile, this year's S&P 500 buyback growth is around +11% year-over-year, with new authorizations nearing a record $1 trillion year-to-date. An estimated $1.4 trillion in total buybacks for the year is expected to offset roughly $700 billion in primary equity supply pressure.