Server DRAM spot prices diverge from contracts
Spot prices for 64-gigabyte server DRAM RDIMMs topped $3,100 on July 23, roughly 146% above the fixed transaction price of $1,260 recorded at the end of June, according to industry data cited by Chosun Biz. Meritz Securities attributed the steep uptrend that began in mid-July to tight supply as sovereign AI projects in countries including Saudi Arabia move forward and test demand for new data centers expands. Industry officials said securing even sample quantities has become difficult, signaling that AI build-outs are progressing faster than contracted volumes can accommodate.
Analysts view the divergence as a potential precursor to higher contract prices, following a pattern from January 2026 when urgent AI server orders pushed spot prices up first and fixed prices followed in February. The market sees the move as evidence that AI investment effects are spreading to server DRAM after high-bandwidth memory, with Alphabet's raised capital expenditure outlook and expanded AI data center investments by Meta, Microsoft and Amazon Web Services driving demand. Samsung Electronics and SK Hynix are positioned to benefit, as server DRAM remains a core revenue source for both companies' memory businesses alongside HBM.
Automakers absorb rising memory costs
The AI memory shortage is now rippling into vehicle pricing. General Motors Chief Financial Officer Paul Jacobson said during the company's earnings call that GM expects costs to rise by $1.5 billion to $2 billion, driven largely by rising DRAM prices, and lifted its full-year vehicle price projection to a 0.3% increase from a previously flat or slightly lower outlook, according to Tom's Hardware citing Nikkei Asia. Chinese automaker BYD has raised prices on its driver assistance features by 20%, and South Korea's Hyundai has called on domestic chipmakers to strengthen the local supply chain.
Vehicle memory content is climbing rapidly. Micron estimated in 2023 that the average vehicle used 90GB of DRAM and NAND, with expectations of 278GB by 2026 and high-end models requiring up to 2TB. A British OEM breakdown showed infotainment systems needing 4GB to 16GB, ADAS and safety systems 8GB to 32GB, and centralized systems 16GB or more, while AI assistants push requirements to at least 256GB. Micron predicts Level 4 autonomous vehicles will need at least 300GB of RAM. Automotive-grade memory requires extensive validation, meaning supply cannot ramp quickly even when fab capacity exists.
Musk moves toward vertical chipmaking
Elon Musk publicly thanked Micron during Tesla's quarterly call for providing a "significant allocation" of memory on "reasonable terms," while outlining an ambitious plan to reduce dependence on outside suppliers, The Next Web reported. Musk is building Terafab, a joint Tesla and SpaceX effort to manufacture computing and memory chips. A research fab is being constructed at Giga Texas with roughly $3 billion in initial investment and capacity for a few thousand wafers per month, while the full build-out could cost $55 billion to $119 billion, with Intel signed on to help design and fabricate.
Musk framed Terafab as insurance rather than leverage, saying the project aims to ensure Tesla and SpaceX have enough chips "down the road" because the industry is not growing fast enough in logic or memory to meet demand. Tesla and SpaceX do not rank among Micron's top 10 customers, according to Bloomberg data, underscoring the scale of the gap between the automakers' needs and supplier priorities.
SK Hynix accelerates packaging buildout
SK Hynix's board authorized approximately 7.09 trillion Korean won, or about $4.83 billion, on July 22 for its P&T7 advanced packaging plant in Cheongju, South Korea, accelerating construction without raising the company's overall capital commitment ceiling of 19 trillion won, Brief Asia reported. The facility broke ground on April 22, 2026, and is expected to begin pilot production in late 2027 with volume ramp planned for the first half of 2028. HBM4 variants and potentially co-packaged optics modules are expected to feature prominently in initial production.
Advanced packaging has become a critical bottleneck in the AI semiconductor supply chain, with modern techniques stacking memory dies vertically or arranging them side-by-side using through-silicon vias and micro-bumps to dramatically increase data throughput. SK Hynix's HBM3E products, which rely on these methods, are already shipping to customers building large language model training clusters. The Cheongju investment reflects a bet that packaging capacity, rather than raw wafer fabrication, will determine market share in the high-margin AI segment over the next three years.
Chip stocks rally on AI capex outlook
SK Hynix's U.S.-listed shares rose 4.5% on July 23 to $172.70, building on a volatile but upward run since the company's record Nasdaq debut on July 10, IBTimes reported. The rally was fueled by Alphabet's earnings report released after the prior market close, in which the Google parent raised its 2026 capital expenditure forecast to $195 billion to $205 billion from earlier guidance of $180 billion to $190 billion. Alphabet reported Google Cloud revenue jumped 82% to $24.8 billion in the second quarter, with Chief Financial Officer Anat Ashkenazi saying "the demand still outpaces that investment."
The broader semiconductor sector also rose, with Micron Technology gaining roughly 3% at the start of U.S. trading. SK Hynix raised $26.5 billion in its American depositary receipt offering on July 10, the largest first-time share sale by a foreign company in U.S. history, surpassing Alibaba's 2014 listing. The company sold 177.9 million ADRs at $149 each, with each ADR representing one-tenth of a common share traded in Seoul. SK Hynix is scheduled to report second-quarter earnings on July 29, providing investors their first detailed look at the company's financial performance since its Nasdaq listing.
Share this article
