Why is Goldman Sachs betting on a 44% further rise in HBM next year? The price increase of traditional DRAM is rewriting the pricing logic.

On the first trading day of July, the crypto market continued its weak consolidation pattern. According to Gate market data, Bitcoin fell below the psychological barrier of $60,000, trading at $58,554.7; Ethereum simultaneously broke below the $1,600 mark, settling at $1,574.07. The total crypto market cap remained at approximately $2 trillion, with Bitcoin's market dominance exceeding 57%. Over the past 24 hours, total liquidations across the market amounted to approximately $249 million, with long positions facing significant pressure. The Fear and Greed Index dropped to a range of 11-16, remaining in the extreme fear territory for consecutive periods.

In contrast to the chill in the crypto market, the semiconductor memory market is experiencing a wave of concentrated upward revisions. In its DRAM sentiment indicator released on June 30, Goldman Sachs significantly raised its forecast for Samsung Electronics' 2027 HBM price growth from the previous 14% to 44%. This adjustment is substantial—from 14% to 44%, a full 30 percentage points higher. More notably, Goldman Sachs also pointed out that this upwardly revised forecast "may still not fully reflect upside risks," implying that the 44% increase estimate "may not be the endpoint."

Where does the 44% come from? Why did Goldman Sachs make such a significant upward revision at this juncture? Answering this requires dissecting three core logical chains in Goldman Sachs' report.

Logic 1: Spot price increases in traditional DRAM are rewriting the pricing anchor for HBM

The most direct trigger for Goldman Sachs' upward revision this time comes from the strong performance of spot prices for traditional DRAM.

According to Goldman Sachs' tracking metrics, as of June 26, the spot price of DDR5 has risen 20% since May 1, with a 25% premium over the May contract price; DDR4 spot prices rose 11% over the same period, with a premium of up to 45% over the May contract price. The latest quotes on July 1 show the average price of DDR5 16Gb stable at $46.833, rising 0.21% from the previous trading day, with spot prices still operating at high levels.

Spot prices significantly outpacing contract prices are typically seen as a leading indicator in the memory industry that subsequent contract prices will follow suit. Goldman Sachs explicitly stated in the report: "The current strong performance of traditional DRAM spot prices will certainly be taken into important consideration when discussing next year's HBM pricing."

This logical chain is not complicated. HBM contract negotiations do not occur in isolation—when memory manufacturers discuss next year's HBM supply contracts with cloud providers and AI chip companies, the market prices of traditional DRAM constitute an important reference frame. As the prices of common memory like DDR5 and DDR4 continue to strengthen, the bargaining position of memory manufacturers becomes more solid. As a higher-end, more scarce product, HBM's premium over ordinary DRAM could actually expand rather than shrink.

Goldman Sachs further extrapolated in its model: if HBM supply and demand remain tight and the price gap between traditional DRAM and HBM continues to widen, the 44% forecast could be subject to further upward revision. This is a self-reinforcing logic—the more traditional DRAM rises, the more room there is for HBM price increases.

Logic 2: Cross-validation of industry chain data shows price increases are not an isolated phenomenon

If the upward price expectation revision were only theoretical, it would be far less convincing. Another layer of support in Goldman Sachs' report comes from the comprehensive strengthening of export and revenue data in May.

According to official Korean data, South Korea's total exports in May 2026 reached $87.8 billion, a year-on-year increase of 53.2%, setting a new monthly record. Among this, DRAM exports were $18.6 billion, up 369.8% year-on-year and 21% month-on-month, also hitting a record high. NAND exports rose 206.8% year-on-year. Goldman Sachs noted that behind this performance were rising memory prices combined with strong demand from major U.S. and Chinese tech companies.

Data from the Taiwan supply chain is more intuitive. Nanya Technology's consolidated revenue in May 2026 was NT$27.67 billion, up 8.55% month-on-month and 730.14% year-on-year, maintaining triple-digit annual growth for ten consecutive months. Goldman Sachs tracked distributor Supreme Electronics, whose May revenue grew 253% year-on-year and 54% month-on-month.

The server supply chain is also booming. The combined May revenue of Taiwan's server ODM manufacturers (Inventec, Quanta, Wistron, Wiwynn) grew 53% year-on-year, driven by the rapid increase in shipments of rack-level AI servers and ASIC AI servers. Aspeed, the primary global supplier of server BMCs, saw its May revenue increase 69% year-on-year—notably, this was achieved on top of a high base of 75% growth in the same period last year.

This set of data collectively points to one conclusion: demand from AI servers and traditional servers is still absorbing higher memory prices. Price increases are no longer just numbers changing on quote screens but are genuinely reflected in manufacturers' revenue and export data. As long as downstream customers are willing to accept price increases, memory manufacturers' bargaining power for 2027 HBM contracts will not be weak.

Logic 3: Supply rigidity + demand resilience, the supply-demand gap is hard to close in the short term

The third support for the HBM price increase expectation comes from structural constraints on the supply side.

Currently, over 95% of global DRAM production capacity is held by Samsung, SK Hynix, and Micron. Since HBM gross margins exceed 70%, significantly higher than the roughly 40% for traditional DRAM, the three major manufacturers continue to prioritize limited 12-inch wafer capacity for HBM production, which in turn keeps supply tight for general-purpose memory like DDR4 and DDR5.

The supply elasticity of HBM is far lower than that of ordinary DRAM. Capacity conversion is more difficult, customer certification cycles are longer, and building additional production capacity requires massive investment and often takes years. This means that even with clear price signals, the supply side's response speed is inherently limited.

There are no signs of weakening on the demand side either. Micron's entire 2026 HBM supply is already sold out under fixed-price contracts. TrendForce expects SK Hynix to maintain about 50% of the HBM market share in 2026, Samsung about 28%, and Micron about 22%. Although the capacity expansion plans of the three giants are aggressive—Samsung plans to increase HBM capacity by 50% in 2026, targeting 250k wafers per month—the supply gap is unlikely to be fully closed before 2027 given the continued expansion of AI server demand.

Goldman Sachs estimates that Samsung's average selling price for DRAM in the second quarter of 2026 will rise about 46% quarter-on-quarter, then slow to 15% and 7% in the third and fourth quarters, respectively. Prices may continue to rise, but the fastest phase of increases likely occurred in Q2. Goldman Sachs maintains a "Buy" rating for Samsung Electronics, with a 12-month target price of 480,000 Korean won for common shares and 360,000 Korean won for preferred shares.

Risk Boundaries: 44% is a forecast, not a promise

No analysis can only talk about the positives. Goldman Sachs' report also includes clear risk warnings.

First, it's important to clarify that 44% is Goldman Sachs' forecast for Samsung's HBM price growth in 2027, not a realized quote or Samsung's own guidance. It is based on the assumptions that HBM supply and demand remain tight and traditional DRAM prices stay strong. If these assumptions change, the forecast itself will be revised.

Goldman Sachs trader Kippei Yamaura noted in another report that the recent sharp sell-off in tech stocks reflects not just a short-term correction but that AI infrastructure and memory supply chains are starting to face structural challenges. The report pointed out three major risks:

First, HBM supply and demand may turn around in 2027-2028. As Samsung, Micron, and SK Hynix continue to significantly expand production, HBM capacity will increase rapidly, and the market generally expects the premium from supply shortage to peak during this period.

Second, Chinese memory forces are rising quickly. According to the Financial Times, Apple is actively lobbying the U.S. government to approve procurement of memory from China's CXMT. If policies change in the future, it could directly impact the market position of existing DRAM suppliers.

Third, the AI investment frenzy is beginning to show signs of cooling. Qualcomm is studying ways to reduce reliance on HBM, and Nvidia is continuously optimizing memory usage efficiency. OpenAI recently lowered its service prices, raising renewed market doubts about when massive AI investments will translate into actual profits.

Additionally, smartphone demand remains a drag. Although China's smartphone shipments in May grew 19% year-on-year, cumulative shipments from 2026 to date are roughly flat with 2025, and Q2 shipments are still expected to decline 14% year-on-year. Price increases themselves can suppress some consumer electronics demand—AI servers and cloud providers have greater capacity to absorb higher memory prices, but terminals like phones and PCs may reduce purchases or delay restocking if they cannot pass on costs.

Two Markets, One Question

Let's bring the focus back to the crypto market. Bitcoin fell below $60,000 on July 1, with a cumulative decline of nearly 20% in Q2. In June, the U.S. Bitcoin spot ETF saw net outflows of about $4.06 billion, the largest monthly redemption since the product's launch.

Memory chip price increases and crypto asset declines may seem to belong to two different worlds, but in reality, they are answering the same question: Under the dual constraints of liquidity environment and end-user demand, how are asset prices being repriced? AI computing demand supports the price increase logic for HBM and DRAM, but the weakness in consumer electronics sets boundaries for that logic. The crypto market faces a similar tension—institutional funds continue to flow out via ETFs, and on-chain data shows some long-term holders are "capitulating."

Investors in both markets are waiting for the same answer: whether end-user demand can support current valuations.

Conclusion

Goldman Sachs raised its forecast for Samsung's 2027 HBM price growth from 14% to 44%, backed by three progressively layered logics: spot price increases in traditional DRAM are rewriting HBM's pricing reference; Korean export and Taiwanese manufacturer revenue data cross-validate the industry boom; supply rigidity combined with demand resilience make the supply-demand gap hard to close in the short term. Goldman Sachs also explicitly noted that the 44% forecast "may not be the endpoint."

But 44% remains a forecast, not an already realized price. The realization of this forecast ultimately depends on the sustainability of AI server demand, the pace of HBM supply release, and whether consumer electronics can bear more expensive memory. Goldman Sachs itself pointed out three structural risks: a potential turnaround in HBM supply and demand, the rise of Chinese manufacturers, and a cooling of AI investment.

For crypto market investors, the pricing logic of memory chips offers a window to observe the health of the AI industry chain. At a time when crypto assets themselves are in a weak consolidation phase, understanding the cost structure and pricing dynamics of computing infrastructure may be more valuable than simply staring at candlesticks. The logical chain for price increases is clear, but whether it ultimately materializes—the market will provide its own answer.

FAQ

Q1: Why did Goldman Sachs significantly raise its Samsung HBM price forecast from 14% to 44%?

The core reason is the continued strength of traditional DRAM spot prices. DDR5 spot prices have risen 20% since May, with a 25% premium over contract prices; DDR4 rose 11%, with a premium as high as 45%. Spot prices typically lead contract markets, and the current strong performance of traditional DRAM will be used as a reference in next year's HBM pricing negotiations.

Q2: Is the 44% growth forecast already happening?

No. 44% is Goldman Sachs' forecast for Samsung's HBM price growth in 2027, not a realized quote or Samsung's own guidance. Goldman Sachs also believes that if HBM supply and demand remain tight and the price gap between traditional DRAM and HBM continues to widen, this forecast could still be revised upward.

Q3: Besides traditional DRAM price increases, what other data supports this forecast?

South Korea's DRAM exports in May grew 369.8% year-on-year and 21% month-on-month, hitting a record high. Nanya Technology's May revenue grew 730% year-on-year, with triple-digit growth for ten consecutive months. Taiwan server ODM manufacturers' combined revenue grew 53% year-on-year, and server BMC supplier Aspeed's revenue grew 69% year-on-year. Industry chain data is cross-validating the price increase logic.

Q4: Did Goldman Sachs mention any risks?

Goldman Sachs traders pointed out three major structural risks in another report: HBM supply and demand may turn around in 2027-2028; Chinese memory manufacturers (such as CXMT) are rising; and the AI investment boom is beginning to show signs of cooling. Additionally, smartphone demand is weak, and price increases themselves may suppress consumer electronics terminal demand.

Q5: What relevance does HBM price increases have for the crypto market?

HBM is the core memory paired with GPUs in AI servers, and its price trend reflects cost changes in AI computing infrastructure. AI-related projects and mining hardware costs in the crypto market are indirectly linked to chip prices. Moreover, there is a liquidity-level linkage between memory chip industry health and global risk asset pricing, making it a useful indicator for observing sentiment in the tech industry chain.