NEW YORK, August 23, 2026, 02:10 EDT Nvidia server prices are up, according to reports, and big memory names are already talking about 2027 cost outlooks as a result.
- Bloomberg said a lot of big Nvidia server buyers saw increases of more than 15%.
- The reported updates affect early-2027 deliveries. That includes Vera Rubin and Grace Blackwell systems.
- Nvidia hasn’t confirmed the price hikes. Reuters said it couldn’t verify the changes on its own.
Nvidia (NASDAQ:NVDA) customers are looking at AI-server price hikes of more than 15% in several cases, Bloomberg said Saturday. The price changes are expected to hit systems delivered from early 2027. Systems with Vera Rubin and Grace Blackwell setups are covered in the new pricing.
The report is relevant as memory now takes up a big chunk of rack space. Nvidia’s Vera Rubin NVL72 has as much as 20.7 terabytes of HBM4. Its chips also run with 54 terabytes of LPDDR5X memory.
A flat 15% budget hike would be overly simplistic. The reported price jumps change with chip generation and memory setup. Nvidia had no comment outside normal hours. Reuters could not independently confirm Bloomberg’s report.
| Reported pricing point | Current information | Status |
|---|---|---|
| Increase | Over 15% in a lot of cases | Bloomberg report; company hasn’t confirmed |
| Effective shipments | Starting in early 2027 | Reported |
| Named platforms | Vera Rubin and Grace Blackwell | Reported |
| Variation | Depends on chip generation and memory config | Reported |
| Public list prices | Still not out | Unavailable |
Server contractors told customers about the price hikes, Bloomberg reported. Their customers are Microsoft (NASDAQ:MSFT), Alphabet’s Google (NASDAQ:GOOGL) and Oracle (NYSE:ORCL). This puts the cost jump between Nvidia’s parts and cloud firms.
The hardware specs make configuration look key. Vera Rubin has the same 72-GPU rack as GB200, but packs more HBM memory and more NVLink bandwidth. Nvidia says Vera Rubin’s specs are still preliminary.
| Rack specification | GB200 NVL72 | Vera Rubin NVL72 |
|---|---|---|
| GPU generation | Blackwell | Rubin |
| GPU count | 72 | 72 |
| CPU count | 36 Grace | 36 Vera |
| GPU memory | 13.4 TB HBM3E | 20.7 TB HBM4, preliminary |
| GPU-memory bandwidth | 576 TB/s | 1,580 TB/s, preliminary |
| NVLink bandwidth | 130 TB/s | 260 TB/s, preliminary |
| CPU memory | 17 TB LPDDR5X | 54 TB LPDDR5X, preliminary |
Nvidia’s shares moved before its August 26 earnings. In the first quarter, the company posted $75.2 billion in data-center sales, making up about 92% of revenue.
| GAAP measure | Q1 FY2027 | Q4 FY2026 | Q1 FY2026 |
|---|---|---|---|
| Total revenue | $81.615 billion | $68.127 billion | $44.062 billion |
| Data-center revenue | $75.2 billion | $62.3 billion | Not restated in the Q1 FY2027 release |
| Gross margin | 74.9% | 75.0% | 60.5% |
| Operating expenses | $7.621 billion | $6.794 billion | $5.030 billion |
Memory makers are ramping up dense chips for the new platform. Micron Technology (NASDAQ:MU) started volume shipments of 36GB, 12-layer HBM4 in Q1. The company says stacks can deliver bandwidth above 2.8TB per second.
Nvidia said Vera Rubin hit full production in May, with shipments to start this fall. CEO Jensen Huang called it “built for this moment.” Nvidia announcement
The schedule matches up with the early-2027 pricing shift that’s been reported. OEM buyers should request quotes based on specific configurations. Factors like memory, networking and delivery timing could all affect the quote.
Risks: Nvidia hasn’t confirmed these hikes, which customers are affected, or the final deal terms. Memory prices could move before early-2027 shipments. Contract discounts might mean true increases differ from what’s been reported.
The headline figure is broader than what’s actually happening. Nvidia’s upcoming racks pack extra memory and bandwidth into every order. That means buyers set aside a bigger reserve for those parts, but they don’t just use a standard 15% across the board.
