Nvidia has spent the past two years as the company setting prices in the AI industry, not the company absorbing them. That changed this week. Bloomberg reported Saturday, August 22, that Nvidia has notified some of its largest customers that server prices containing its AI chips will rise by more than 15 percent in many configurations, with the increases taking effect on systems shipped starting early next year. Even the most dominant company in AI hardware, it turns out, cannot simply eat the cost of the memory chip shortage rippling through the entire industry.

What Nvidia Actually Told Customers

The price increases will hit systems built around Nvidia's current flagship Grace Blackwell chips and its next-generation Vera Rubin architecture, according to people familiar with the process cited in Bloomberg's original reporting. The exact size of each increase depends on the specific chip generation and memory configuration involved, meaning not every customer or every system will see an identical jump. Fortune's reporting adds a detail that clarifies who actually gets hit hardest: the companies most directly affected are the contract manufacturers that build servers on behalf of large data center operators like Microsoft, Google, and Oracle, meaning the increase moves through the supply chain before it ever reaches Nvidia's own direct enterprise customers as a finished bill.

Reuters, in its own coverage of the Bloomberg report, noted it could not independently verify the claims, a standard caveat for reporting that relies on anonymous sources describing private commercial communications. But the pricing pressure driving the story is not itself in dispute. It shows up in memory chip pricing data from multiple independent research firms, and it shows up in Nvidia's own recent public statements about the market it operates in.

The Root Cause Sitting Underneath Nvidia's Own Chips

Nvidia's GPUs do not work in isolation. Every AI server pairs Nvidia's processors with large quantities of high-bandwidth memory and server DRAM, the specialized memory chips that feed data to the processor fast enough to keep it running at full capacity. According to 24/7 Wall St's analysis, server DRAM prices roughly doubled in the first quarter of 2026 alone, and Counterpoint Research documented an 80 to 90 percent quarter-over-quarter price increase across DRAM, NAND, and HBM during that same period. Herald Business's reporting frames the mechanism plainly: a shortage of memory chips has pushed even Nvidia, a company with immense pricing power of its own, to pass costs downstream rather than absorb them, which is "significantly strengthening the pricing power of memory manufacturers such as Samsung Electronics and SK Hynix."

That shift in leverage is the real story here, more than the specific 15 percent figure. Samsung and SK Hynix control the overwhelming majority of the world's high-bandwidth memory production, and as AI infrastructure spending has scaled faster than memory fabrication capacity, those two companies have gained genuine pricing power over a company that, in almost every other part of the AI supply chain, sets the terms itself. Nvidia can dictate GPU pricing to nearly anyone. It cannot dictate memory pricing to the handful of companies that actually make the chips its own processors depend on.

Why This Shortage Isn't Ending Soon

The forecasts attached to this story are genuinely sobering for anyone hoping this is a short-term supply hiccup. TrendForce projects that demand for HBM and server memory will continue outpacing supply growth, keeping DRAM markets tight through 2027. Deloitte's own analysis, cited by 24/7 Wall St, goes further, projecting that meaningful new memory production capacity will not come online until 2029 or 2030, and separately forecasting that AI-server DRAM prices could quadruple over the course of 2026 alone. Gartner has projected the supply crunch persisting at least through the first half of 2027. Three independent research firms converging on a multi-year timeline, rather than a temporary spike correcting itself within a few quarters, should reshape how every company building AI infrastructure plans its budgets going forward.

This is not a problem confined to server hardware either. Apple has already raised product prices by as much as 20 percent, and Amazon reportedly hiked Echo Dot pricing by 60 percent, both companies explicitly attributing the increases to memory cost spikes driven by AI data center demand. The same underlying scarcity pushing up the price of a Grace Blackwell server rack is pushing up the price of a smart speaker sitting on someone's kitchen counter, because both devices ultimately compete for the same limited pool of memory chip manufacturing capacity.

The Timing That Makes This Sting More

Nvidia is scheduled to report its fiscal second-quarter earnings next week, a moment Fortune describes as one of the most closely watched corporate updates in the entire technology industry, given how much capital has flowed into AI infrastructure on the promise that the spending will eventually pay off. A price increase notification landing in customers' inboxes just days before that earnings report is unlikely to be coincidental timing from Nvidia's perspective. Companies typically prefer to have pricing changes already communicated and absorbed into customer expectations before a major earnings call, rather than announcing a margin-protecting price hike in the same breath as quarterly results, where it risks reading as a signal of cost pressure rather than a routine business update.

The stock market's own reaction this week suggests investors are already pricing in some of this pressure independently. Chip stocks broadly declined on Monday, August 24, with Micron shedding 5.8 percent and AMD and Broadcom both pulling back more than 2 percent, a selloff that reflects growing market unease about how sustainable the current pace of AI infrastructure spending actually is once rising component costs get factored into the math. Rising costs alone do not necessarily undermine the AI buildout, companies with sufficient capital can simply pay more, but they do compress margins for everyone downstream of Nvidia, from the contract manufacturers assembling the servers to the cloud providers ultimately renting out the compute to end customers.

What This Adds to an Already Complicated Buildout

Nvidia's price increase does not exist in isolation from the broader set of obstacles already complicating the AI data center buildout globally. Fortune's own reporting lists project delays, labor shortages, tightening capital markets, and local community resistance to new data center construction as existing headwinds facing the industry, well before this specific memory-driven cost increase enters the equation. Layering a double-digit percentage increase on top of already-strained project economics does not derail any single planned data center on its own, but it does compound the total cost of an infrastructure buildout that was already measured in hundreds of billions of dollars before this latest increase.

Who Actually Benefits From All of This

The clearest winners in this entire dynamic are the memory manufacturers themselves. Samsung and SK Hynix's growing pricing power is not an abstract market observation, it directly explains why SK Hynix has been willing to commit up to $720 billion toward new AI memory fabrication capacity, a bet that only makes financial sense if elevated memory prices persist long enough to justify that scale of capital investment. Every company downstream of that memory supply chain, from Nvidia itself down through server manufacturers, cloud providers, and ultimately consumers buying smart home devices, is currently subsidizing that capacity expansion through higher prices, whether or not any individual buyer realizes that's what their higher bill actually represents.

What to Watch Next

Nvidia's earnings report next week will be the first real test of how the company frames this price increase publicly, and whether it treats the hike as a temporary, cost-driven adjustment or signals a more permanent recalibration of how AI hardware gets priced going forward. The more consequential question, though, sits well beyond any single earnings call: whether the current wave of AI infrastructure spending can continue at its present pace once the actual cost of building that infrastructure keeps climbing, quarter after quarter, with no credible relief in memory supply expected for at least another two to three years. Nvidia passing along a 15 percent increase is not evidence the AI boom is ending. It is evidence that even its most dominant company no longer controls every input required to keep building it.