惯性聚合 高效追踪和阅读你感兴趣的博客、新闻、科技资讯
阅读原文 在惯性聚合中打开

推荐订阅源

博客园_首页
量子位
D
DataBreaches.Net
博客园 - 司徒正美
J
Java Code Geeks
博客园 - 【当耐特】
OSCHINA 社区最新新闻
OSCHINA 社区最新新闻
aimingoo的专栏
aimingoo的专栏
B
Blog
The Cloudflare Blog
D
Docker
I
InfoQ
爱范儿
爱范儿
MongoDB | Blog
MongoDB | Blog
腾讯CDC
月光博客
月光博客
Hugging Face - Blog
Hugging Face - Blog
Microsoft Azure Blog
Microsoft Azure Blog
Vercel News
Vercel News
阮一峰的网络日志
阮一峰的网络日志
小众软件
小众软件
S
SegmentFault 最新的问题
GbyAI
GbyAI
有赞技术团队
有赞技术团队

Latest from Tom's Hardware in Artificial-intelligence

Microsoft says 'Transformation Paradox' holding back AI adoption in the workplace — 45% of respondents say it's safer to focus on current goals, rather than AI innovation Palantir co-founder Peter Thiel backs $140M wave-powered AI data center startup — Panthalassa aims to run offshore… Google, Microsoft, and xAI agree to let US government test AI models before public release — OpenAI and Anthropic also on board after renegotiating deals with Washington Nvidia CEO Jensen Huang says China should not have Blackwell or Rubin AI GPUs — firmly states US should have 'the first, the most, and the best' when it comes to AI hardware China pushes for 70% homegrown silicon wafer use as domestic firm ramps up 12-inch production — government seeking to localize critical chip supply chain amid AI boom and export restrictions Intel swipes Qualcomm veteran of 25 years to lead client computing — Alex Katouzian jumps ship to oversee consumer… Trump administration considers mandatory pre-release vetting of AI models — Anthropic's Mythos cited as… Nvidia's exposure to Asian supply chains for components hits 90% of its production costs — marked increase from 65% could intensify as physical AI adds even more exposure Anthropic in early talks to buy DRAM-less AI inference chips from UK startup — Fractile's SRAM architecture reduces need for pricey memory during extreme pricing and shortage crunch Chinese court rules companies can't fire workers just because AI is cheaper — ruling says automation alone… Jensen says Nvidia now has 'zero percent' market share in China — says US export policy 'has… US Navy signs deal with AI firm for training underwater drones to detect mines in Strait of Hormuz — $100 million would allow drone minesweepers to update their detection algorithms in days instead of months The Pentagon announces AI deals with OpenAI, Google, Microsoft, Amazon, Nvidia, and more — LLMs to be deployed on classified Department of War networks ‘for lawful operational use’ SoftBank plans robotics and AI firm in the US to build data centers — aims for $100 billion valuation and an IPO… Huawei could seize China’s AI chip crown in 2026 as Nvidia's H200 shipments stall in regulatory limbo — Beijing pushes homegrown AI hardware dominance in a market projected to hit $67 billion by 2030 Talent over tokens: AI models are becoming more expensive to run, and productivity gains are limited — efficient workers might be the solution to strained budgets Samsung and SK hynix warn AI-driven memory shortages could last until 2027 and beyond, as HBM demand explodes — customers already reserving supply years ahead, while the wider DRAM market begins to tighten Victim of AI agent that deleted company's entire database gets their data back — cloud provider recovers critical files and broadens its 48-hour delayed delete policy Exploding number of AI data center build-outs delay Texas housing projects — data centers' high demand for electricians prices out contractors, homes now take two months longer to complete Meta's multi-billion-dollar Graviton deal highlights intensifying CPU shortages in AI infrastructure — the industry signals a shift to Agentic inference workloads, pushing demand OpenAI has effectively abandoned first-party Stargate data centers in favor of more flexible deals — company now prefers to lease compute and says Stargate is an umbrella term Google signs classified Pentagon AI deal but exits $100 million drone swarm program — report claims employees revolted over ethical fears, delivered letter to CEO Pichai Nvidia exec says AI is more expensive than actual workers — yet some companies don't see the extra costs as a… Meta will beam sunlight from space to power AI data centers, solar-collecting satellites will orbit 22,000 miles above Earth — firm reserves 1 Gigawatt of orbital solar energy and 100 Gigawatt-hours of long-duration storage Market slumps as OpenAI reportedly misses internal targets for active users and revenue — Nvidia, Oracle, AMD, and CoreWeave shares all tremble on the news OpenAI and Microsoft News site linked to OpenAI super PAC sent bots posing as journalists to interview real people — site has published nearly 100 articles with real quotes gathered by fake writers Claude-powered AI coding agent deletes entire company database in 9 seconds — backups zapped, after Cursor tool… DeepSeek launches 1.6 trillion parameter V4 on Huawei chips as U.S. escalates AI theft accusations — U.S. gov't alleges IP theft by DeepSeek and other Chinese AI firms NEO Semiconductor's revolutionary 3D X-DRAM for AI processors has passed proof-of-concept validation — company secures funding to develop next-gen memory HBM alternative
Nvidia's memory costs soar 485%, latest AI systems now co...
Anton Shilov · 2026-05-21 · via Latest from Tom's Hardware in Artificial-intelligence
Nvidia
(Image credit: Nvidia)

As prices of components are increasing rapidly due to high demand from the AI sector, the cost of these machines is also increasing significantly. Morgan Stanley Research estimates that a next-generation Vera Rubin-based VR200 NVL72 rack will cost major hyperscale cloud service providers (CSPs) around $7.8 million per unit (via @Aaronwei3n), which is tangibly more than about $4 million per GB300 NVL72. Furthermore, because every VR200 NVL72 rack packs plenty of DRAM and NAND, memory now accounts for around 25% of the total cost.

Nvidia plans to charge $55,000 per Rubin GPU and $5,000 per Vera CPU when selling them in volume inside VR200 NVL72 chassis to hyperscalers, according to Morgan Stanley. Although the upcoming VR200 NVL72 racks use the already familiar Oberon chassis, they use more sophisticated switching, networking, printed circuit board (PCB), cooling, power supply, and even chip packaging components, which increases bill-of-material (BOM) costs and eventually the price of the systems. As a result, each VR200 NVL72 will cost hyperscalers around $7.8 million, according to Morgan Stanley, which is higher than around $7 million we were told by one of our sources in late March. Meanwhile, the cost of memory within a VR200 NVL72 rack will be about $2 million, up 435% from the memory cost in GB300 NVL72, according to the same figures.

Sheesh.$NVDA VR200 Bom Analysis from MS. pic.twitter.com/sutjttSkyWMay 21, 2026

There are several reasons why the cost of memory is expected to account for 25% of the cost of a VR200 NVL72 system and why the system carries $2 million worth of memory.

First up, each of such racks now contains 54 TB of LPDDR5X memory, up from 17 TB of LPDDR5X in the case of a GB200 NVL72, a threefold increase. SemiAnalysis estimates that Nvidia paid $8 per GB per GB of LPDDR5X in Q1, though that price may increase as demand rises in the coming quarters, especially if we are talking about SOCAMM2 modules that are expensive to make and test. In any case, even at $8 per GB, each GB200 NVL72 machine carries $136,000 worth of LPDDR5X memory, whereas each VR200 NVL72 system will contain $408,000 worth of LPDDR5X content. If the price rises to $10, we are talking about $540,000 for LPDDR5X alone. Note that even $10 per GB may be an underestimate* as Nvidia adds its own markup.

Secondly, each VR200 NVL72 rack carries about $1 million or more of 3D NAND storage, up from virtually zero inside GB200 NVL72.

As a result, $2 million of memory content per Vera Rubin NVL72 rack is not something completely unexpected: the system uses a lot of LPDDR5X and 3D NAND memory (not to mention HBM4 memory onboard of Rubin GPUs), and memory now comes at massive prices.

*Contract price of DDR5 memory is now between $12 and $16 per GB, depending on various factors and luck, according to Framework. Spot price for DDR5 was about $20 per GB on average at press time, according to DRAMeXchange. LPDDR5X is more expensive than DDR5. When installed on SOCAMM2 modules (which are exclusively used by Nvidia's Vera CPUs), it will get even more costly, especially when Nvidia's markup is added.

Get Tom's Hardware's best news and in-depth reviews, straight to your inbox.

Google Preferred Source

Follow Tom's Hardware on Google News, or add us as a preferred source, to get our latest news, analysis, & reviews in your feeds.

Anton Shilov is a contributing writer at Tom’s Hardware. Over the past couple of decades, he has covered everything from CPUs and GPUs to supercomputers and from modern process technologies and latest fab tools to high-tech industry trends.