Nvidia has told its largest customers to expect list prices for servers built on its upcoming Vera Rubin platform and its current Grace Blackwell systems to rise more than 15% — in some cases as much as 17% — for units shipping in early 2027, according to a Bloomberg report that cited people familiar with the matter. The notice reportedly reached buyers through Nvidia's contract server manufacturers rather than directly from the company, and Nvidia has not publicly confirmed the figures.
The driver, according to the report and corroborating trade coverage, is the price and availability of memory: DRAM, LPDDR and high-bandwidth memory have all become scarcer and more expensive this year, and memory now accounts for a far larger share of a finished AI server's cost than it used to. An analysis cited in reporting from Wccftech put memory's share of the bill of materials for a Vera Rubin VR200 rack — a roughly $2.1 million system — at about 29%, versus the 20% share Nvidia has historically targeted. Samsung and SK Hynix warned as early as April that memory shortages could persist through 2027.
Timing collides with earnings
The reported price increases surfaced days before Nvidia is due to report fiscal second-quarter results on Wednesday, August 26 — a report investors will parse closely for any signal about how rising input costs are affecting margins on its data-center business, still by far the company's largest revenue driver. Nvidia shares have fallen for multiple consecutive sessions heading into the print, with chip peers Intel, AMD, Micron and Taiwan Semiconductor also sliding as investors weigh memory-cost pressure across the sector, according to Tom's Hardware.
The price increases would land at a delicate moment for the broader AI infrastructure buildout, which has drawn growing scrutiny over whether spending on data centers is outpacing near-term returns. Rising component costs give further ammunition to skeptics who argue the economics of large AI models remain fragile, even as Nvidia's revenue and order backlog continue to grow. For Nvidia's hyperscaler customers — who are collectively spending hundreds of billions of dollars a year building out AI data centers — a 15%-plus jump in server costs would add meaningfully to already enormous capital budgets, even as it likely boosts near-term revenue for memory makers supplying the components.
Neither Nvidia nor its named contract manufacturers have issued an on-record statement confirming the specific percentages reported by Bloomberg. The company is expected to face questions on the topic from analysts during Wednesday's earnings call.
