HBM Semiconductor Rally Check: Scarcity Is Real, but Valuation Discipline Still Matters

The HBM semiconductor rally is still anchored in a real supply-demand squeeze, but the evidence no longer supports a simple “shortage equals upside” story. The better thesis is narrower: scarcity can justify premium pricing only if suppliers convert HBM4 risk production into qualified, profitable volume, while investors keep cycle valuation grounded in balance-sheet discipline rather than peak earnings excitement.

🏭 HBM Semiconductor Rally: The Shortage Is Not Imaginary

The strongest fundamental support for the rally is that HBM supply is constrained by both demand growth and production substitution. According to Yonhap’s report on TrendForce’s analysis, negotiations have already shifted toward 2027 HBM4 supply contracts, and TrendForce expects suppliers to have stronger bargaining power because AI demand is rising while HBM production crowds out conventional DRAM capacity.

That matters because HBM is not merely another DRAM product with a higher selling price. It consumes scarce advanced memory capacity and packaging resources, and each generation tends to require larger die sizes. TrendForce, as reported by Yonhap, said this “crowd-out” effect is intensifying as HBM generations advance. The investment implication is straightforward: if HBM wafers reduce the supply available for DDR5, LPDDR5X, GDDR7, and other DRAM products, pricing power may spread beyond HBM into broader memory markets.

The demand side is equally specific. TrendForce reportedly expects AI ASICs to increase HBM consumption as chip-level HBM content moves from 96GB and 192GB configurations toward 216GB and 288GB. It also expects Nvidia’s next-generation Rubin Ultra platform in 2027 to increase HBM capacity per GPU to 384GB, while Google TPU and other AI ASIC shipments add incremental demand. These are not abstract AI slogans; they are bill-of-material pressure points. More memory per accelerator means the same number of systems can require substantially more HBM supply.

Still, the rally’s quality depends on price realization, not just unit demand. TrendForce reportedly noted that annual HBM pricing structures have not fully reflected recent market price increases, and that HBM wafer revenue and profitability had fallen below DDR5 64GB RDIMM by the first quarter of 2026. If accurate, that is a crucial warning: the market may be pricing HBM as the premium product, while suppliers may still be negotiating to make the economics match the narrative.

HBM4 Risk Production Is Bullish Only If Qualification Follows

The second support for the rally is that Samsung Electronics and SK hynix are reportedly moving HBM4 into production before final customer certification. ZDNet Korea reported that both companies have been conducting “risk production” for Nvidia-related HBM4 demand, meaning wafers are committed before final qualification and formal purchase orders are fully settled.

This is rational because HBM lead times are long. ZDNet reported that it generally takes around four months to ship finished HBM products. If a supplier waits for every qualification gate before starting production, it may miss the launch schedule of the customer’s accelerator platform. Risk production is therefore a signal of urgency and confidence, but it is also a transfer of uncertainty onto the supplier’s own balance sheet.

For investors, the distinction is important. Pre-certification production can look like proof of demand, but it is not the same as accepted revenue. If a product fails a qualification test, suffers yield problems, or ships in lower-grade bins than expected, the supplier may carry inventory, rework costs, or lower realized pricing. The bullish reading is that customers need HBM4 badly enough to pull supply forward. The bearish reading is that suppliers are absorbing more operational risk to stay relevant in a narrow customer window.

ZDNet reported that Samsung said its HBM4 was already in production input after customer performance evaluation and that shipments, including 11.7Gbps top-grade products, were scheduled from February. The same report said SK hynix had built a mass-production system for HBM4 and was producing customer-requested volume. Those statements support the view that HBM4 is becoming a live commercial battleground rather than a distant technology roadmap.

But the risk is not symmetrical. Samsung entered HBM4 with more urgency because it had struggled in earlier Nvidia HBM3E qualification cycles, according to market reports summarized by Junggi Economy. SK hynix, meanwhile, had the stronger incumbent HBM position but reportedly faced questions around high-speed HBM4 implementation and ramp timing. The result is not a clean leader-follower story. It is a three-player contest in which execution windows are short and the cost of a missed qualification cycle is large.

Share Shifts Are the Cleanest Test of the Rally

The most useful way to test whether the rally is supported by fundamentals is to watch share shifts, not press-release language. Junggi Economy cited Counterpoint Research estimates that in the second quarter of 2025, SK hynix held 62% of the HBM market, Micron held 21%, and Samsung held 17%. The same article cited Daishin Securities estimates that 2025 full-year HBM share would be SK hynix 61%, Micron 22%, and Samsung 18%.

Those figures are secondary-source market estimates, not company filings, so they should be treated as directional rather than definitive. Directionally, however, they show why the rally has become more complicated. SK hynix has not merely benefited from industry growth; it has captured a leadership premium. Micron has not merely participated; it has reportedly moved into the low-20% share range. Samsung is not merely waiting for demand to arrive; it is trying to recover position in a market where qualification, yield, and customer allocation are now decisive.

This competitive structure matters because scarcity does not benefit every supplier equally. If HBM4 qualification broadens the supplier base, customers may gain leverage even in a tight market. If only one or two suppliers can deliver required speed, power, thermals, and volume, pricing power remains concentrated. The rally therefore rests less on the word “HBM” and more on who owns qualified output at the exact specification that AI accelerator customers need.

Micron’s role is particularly important because a credible third supplier changes the pricing game. Even if total HBM supply remains tight, customer procurement teams gain negotiating options when Micron can meet performance and volume requirements. That does not eliminate scarcity, but it can reduce the probability that any single supplier captures all of the economics. For a late-cycle market, that distinction can separate sustainable earnings from margin expectations that are too clean.

Why PBR Has Entered the Debate

The valuation debate has shifted because earnings-based multiples can mislead at cycle extremes. Asia Economy reported on a Kyobo Securities view that semiconductor valuation should rely more on price-to-book ratio, or PBR, than price-to-earnings ratio, or PER, because memory remains cyclical even with HBM-driven structural demand.

The logic is credible. In memory upcycles, earnings can expand rapidly as pricing improves and fixed costs are absorbed across higher-value output. PER can then look optically low exactly when profits are near unusually strong levels. Conversely, during downturns, PER can look meaningless or expensive because earnings collapse. PBR is not perfect, but it forces investors to compare market value with the asset base required to produce those earnings.

This is especially relevant for HBM because the industry is being asked to spend ahead of confirmed long-term profitability. Risk production, advanced packaging, larger die sizes, and customer-specific qualification all increase capital and operating intensity. If investors value the cycle only on near-term earnings, they may understate the capital needed to defend those earnings. A PBR lens asks a colder question: how much premium over book value is justified by returns that may normalize after the shortage eases?

The counterargument is also serious. If HBM changes memory from a commodity-like cycle into a more structurally constrained infrastructure market, historical PBR ranges may be too conservative. AI customers are not buying memory for discretionary gadget cycles; they are buying bandwidth for compute platforms. If that demand proves durable through 2027 and 2028, book value may understate the franchise value of qualified HBM supply.

The disciplined answer is to avoid treating either multiple as a shortcut. PER can understate risk when earnings are inflated by shortage pricing. PBR can understate value if returns on capital structurally improve. The right use of PBR in this rally is not to declare the stocks cheap or expensive; it is to impose a cycle check on assumptions about margins, customer concentration, and required reinvestment.

🔥 Where the Rally Could Overheat

The overheating scenario begins with extrapolation. If the market assumes that 2027 HBM contract prices rise by multiples, that HBM4 qualification proceeds smoothly, and that AI accelerator demand absorbs every available stack at premium margins, then little room remains for disappointment. Each assumption has evidence behind it, but none is guaranteed.

The first risk is price resistance. TrendForce’s reported view that suppliers may push for sharply higher 2027 contract prices is not the same as signed contracts at those levels. Customers such as Nvidia, hyperscalers, and ASIC designers have strong incentives to diversify suppliers and optimize memory configurations. If they succeed, HBM pricing may rise, but less dramatically than supplier-favorable forecasts imply.

The second risk is execution. HBM4 risk production compresses schedule risk but does not erase technology risk. Speed bins such as 11.7Gbps, thermal stability, yield, packaging capacity, and customer validation can all affect shipped volume and margin. A supplier can be directionally right on demand and still disappoint investors if qualified output arrives late or at lower profitability.

The third risk is valuation reflexivity. Strong share-price performance can encourage investors to treat market capitalization as confirmation of fundamentals. In cyclical industries, that is dangerous. A high PBR may be justified if returns on equity remain structurally elevated; it is vulnerable if earnings are being capitalized near a temporary shortage peak.

✅ What to Watch Now

The key evidence is not whether AI demand remains exciting. It is whether HBM4 supply agreements confirm the pricing power that TrendForce reportedly sees building into 2027. Contract duration, pricing mechanisms, volume commitments, and qualification milestones matter more than headline demand commentary.

Investors should also watch whether Samsung, SK hynix, and Micron converge or diverge in HBM4 execution. If all three suppliers qualify meaningful volume, scarcity may persist but customer leverage improves. If qualification remains concentrated, the leader’s margin advantage may last longer. Either outcome supports a different valuation framework.

Finally, watch PBR alongside returns on capital. A rising book multiple is not automatically excessive in a structural upcycle, but it raises the burden of proof. The market must show that HBM profits are not just high, but durable after the first wave of AI infrastructure urgency passes.

The HBM rally still has fundamental support: content per AI chip is rising, supply is technically constrained, and 2027 contract negotiations appear supplier-favorable. But the late-cycle warning signs are visible too: risk production before certification, uneven competitive share, and valuation arguments that depend on sustaining elevated returns. That combination calls for discipline, not disbelief.