Nvidia (NVDA) customers are being told to expect server price hikes of more than 15% as surging memory chip costs squeeze the AI hardware supply chain. The increases will hit systems shipped early next year, arriving just as traders reposition ahead of the company's Q2 earnings report due this week.

Steeper Prices for Next-Generation Systems

The affected systems include those built around Nvidia's Vera Rubin and Grace Blackwell chips. The size of each price increase will depend on the chip generation and memory configuration involved. Contract server builders have already notified major data center operators — including Microsoft, Google and Oracle — of the coming increases, according to Bloomberg. For those operators, servers sit at the core of the capital budgets they have been raising to record levels for AI data center construction, so component-level price increases feed directly into the cost of the buildout.

Memory Makers Gain Rare Leverage

The pricing pressure comes from memory chipmakers Samsung, SK Hynix and Micron, which together control most of the world's DRAM production. Surging demand for AI infrastructure has given them rare pricing power, and Nvidia's AI accelerators are highly dependent on DRAM capacity, which leaves the company especially exposed to swings in memory costs.

Much of that dependence runs through high-bandwidth memory, a stacked form of DRAM that only these three manufacturers produce at scale. Making it consumes far more wafer capacity per bit than standard memory, tightening supply across the broader DRAM market — an industry with a long history of boom-and-bust cycles in which shortages have produced sharp price spikes before new capacity comes online.

Nvidia currently holds a gross margin of around 75% and charges tens of thousands of dollars per chip. Supply from contract manufacturer TSMC still cannot keep up with demand. The company has also raised prices on gaming-focused PC graphics cards, according to industry site Tom's Hardware, extending the increases beyond its data center business.

Customers Building Their Own Chips

Amazon, Microsoft, Google and Meta Platforms remain heavily dependent on Nvidia for their data center buildouts, even as each develops in-house alternatives. Amazon is working to expand its Trainium chips, which power much of Anthropic's AI training. Microsoft has its Maia chip, while Google and Meta are developing TPU and MTIA chips respectively.

Last week, Google expanded its partnership with Marvell Technology to speed up production of its TPU chips. Other firms, including Cerebras, Etched and AMD, are also scaling up AI chip production. Those rivals buy memory from the same three suppliers, so the cost squeeze extends across the accelerator industry rather than hitting Nvidia alone.

The price hikes could push some customers to accelerate their shift toward in-house silicon, though meaningful independence from Nvidia remains a long-term project for all of them.

Earnings in Focus This Week

NVDA stock dropped from $227 to around $214.75 in recent sessions as traders positioned ahead of the Q2 earnings report due this week. The stock has pulled back to near its 50-period Exponential Moving Average and is sitting at a key support level of $214, which was the high from July 25.

Wall Street expects Nvidia to report revenue of around $92 billion for Q2, representing 96% growth year-over-year. Q3 guidance is expected to come in near $103 billion. Nvidia has a track record of beating forecasts, so some analysts expect the real Q2 and Q3 numbers to land closer to $96 billion and $112 billion respectively.

One potential stock catalyst beyond earnings would be an expanded share buyback program. Nvidia announced an $80 billion repurchase earlier this year, and a larger announcement could move the stock. Memory costs have been a recurring topic on Nvidia's earnings calls as Blackwell ramped, and management commentary on the latest increases will be one of the items to watch alongside the headline numbers. Further out, the pace of memory capacity expansion at Samsung, SK Hynix and Micron, and how quickly hyperscalers' in-house chips reach meaningful scale, are among the variables that will shape how long the pricing pressure persists.

The stock has technical support at $214, with resistance at $227. A break below $200 would signal a shift in the current trend.