Good for ↑
Nvidia ships the DGX Spark with half the memory at the same price, while the full-memory version jumped 75% in price. That points to a hard limit on memory supply. The extra profit from scarcity flows to memory makers such as Micron.
Source: The Register ↗
Nvidia would rather ship a weaker product than wait, which suggests a severe shortage. Even Nvidia cannot get enough memory soon at any price. The 75% price jump on the 128 GB model is Nvidia passing on higher memory costs, not gouging buyers. The pricing power sits with the makers of DRAM and HBM, the high-bandwidth memory used in AI chips. Buyers still pay $6,950 for the full-memory unit, nearly double the launch price. Buyers who cannot walk away let memory makers raise prices. Micron (MU) and two Korean rivals dominate, and all of them run near full capacity for HBM3e and LPDDR5X chips. A 5.28% 10-year yield usually weighs on growth stocks, but not on pricing power in a scarce input. Investors still value memory makers as a boom-and-bust business, not one with a new floor of AI demand.
Investors read the DGX Spark memory cut as a supply hiccup for Nvidia's AI PC plans. They see it as mildly bad for Nvidia's consumer and prosumer products and irrelevant to its data center chip business. Memory suppliers are not part of the discussion.
Why that may be wrong
Analysts who cover memory stocks are not linking a product page change for an AI PC to the tight supply of high-bandwidth memory for data centers. They watch DRAM spot prices and flash contract cycles, not Nvidia's DGX Spark model changes. The two facts sit with separate groups of analysts, and nobody has put them together. The information is public, but it spreads slowly.
The link fails if Nvidia cut the memory on purpose, to split its product line for profit, and not because of a shortage. Then memory makers have no capacity limit and no extra pricing power. Watch the next earnings reports from memory makers. If prices for high-bandwidth memory or DRAM contracts fail to rise, the shortage story is wrong.
Nvidia may have cut the memory on purpose, to protect margins and create a premium model, with no real shortage behind it. In that case memory makers face no supply limit and only benefit from Nvidia's pricing choices. The weakest step is reading the 75% jump on the 128 GB model as passed-on memory costs. It could reflect Nvidia's own pricing power instead. The clearest test is the next report from Micron (MU). Flat or falling average prices for its high-bandwidth memory would undo the case, and the argument for valuing memory makers more highly would reverse. A 10-year yield at 5.28% and rising also makes higher valuations harder, even after strong earnings.
The closest match is the 2017-2018 flash memory shortage. Apple's iPhone ramp created demand that would not bend, and PC and server memory demand piled on top. Samsung and SK Hynix widened gross margins by 15-20 percentage points in 12 months. A shortage that shows up at a big device maker, Apple then and Nvidia now, is a late sign. By the time it ships a cut-down model, memory prices have already turned. Investors who waited for Micron's earnings to confirm it in 2017 still saw a gain of 60% or more. In the 2021 car chip shortage, the bottleneck sat at TSMC, but analog chip makers with the scarce capacity, such as Texas Instruments and ON Semi, gained the pricing power. In both cases, the winners controlled the scarce supply, not the famous brand.
Nvidia ships a DGX Spark with half the memory at the original price, and the full 128 GB model jumps 75% in price. Both point to a memory shortage upstream that is limiting AI hardware shipments.
Nvidia chose to ship a weaker product instead of waiting. That suggests supply of high-bandwidth memory (HBM) and dense DRAM has hit a hard limit. Buyers such as Nvidia, AMD and the big cloud companies cannot switch away from HBM3e and LPDDR5X chips. That gives memory makers unusual pricing power in a business investors still treat as a boom-and-bust commodity.
As the shortage shows up across more AI hardware, analysts who cover memory stocks will likely raise their price and gross margin forecasts for Q4 2026 and 2027. Micron (MU), the US memory maker with the most high-bandwidth memory business, captures that pricing power most directly. Its shares have been rising slowly as the AI memory story builds. Investors could come to value it as AI infrastructure instead of a boom-and-bust commodity, a shift only partly reflected so far.
No comments yet.
Sign in to comment