Nvidia and AMD Accelerators: Roadmaps, Pricing, Supply — 2026-09-03
Nvidia's record Q2 FY2027 earnings revealed a structural shift in AI spending, with non-hyperscaler buyers now outpacing traditional giants like Microsoft and Amazon. Simultaneously, Nvidia confirmed price hikes exceeding 15% for its Vera Rubin and Grace Blackwell server systems due to soaring HBM memory costs, effective early 2027. Meanwhile, AMD’s MI400 series continues to challenge Nvidia's dominance, with enterprise surveys indicating a growing preference for non-Nvidia accelerators in next-generation evaluations.
Nvidia and AMD Accelerators: Roadmaps, Pricing, Supply — 2026-09-03
Top developments
Nvidia Confirms >15% Price Hikes for Vera Rubin and Grace Blackwell Systems
Nvidia has notified its largest customers that prices for AI server systems, including the flagship Vera Rubin and Grace Blackwell chips, will increase by more than 15%. The hikes are explicitly attributed to soaring High Bandwidth Memory (HBM) costs and will take effect on systems shipped in early 2027. This move impacts key hyperscalers like Microsoft, Google, and AWS, signaling a persistent supply-side pressure on hardware costs despite strong demand.
Nvidia Earnings Reveal Shift: Non-Hyperscalers Outgrow Big Tech
In its Q2 FY2027 earnings report released on August 26, 2026, Nvidia reported revenue of $96.2 billion, up 106% year-over-year. A key takeaway is the emergence of a "new buyer class" of enterprises and sovereign entities whose AI capital expenditure growth is now outpacing traditional hyperscalers like Microsoft and Amazon. This diversification reduces Nvidia's revenue concentration risk and supports CEO Jensen Huang's forecast of 70% fiscal 2028 revenue growth, far above analyst estimates.
Enterprise Survey Shows 14-Point Lead for Non-Nvidia Chips
A recent VentureBeat survey of 170 enterprise AI buyers indicates that non-Nvidia accelerators are currently 14 points ahead of Nvidia Blackwell on evaluation lists for the next 12 months. This data point complicates the AI capex narrative, suggesting that while Nvidia maintains current dominance, competitors like AMD and specialized ASIC makers are gaining traction in forward-looking procurement decisions.
AMD MI400 Series Gains Momentum Against Blackwell
AMD's Instinct MI400 series, featuring 432 GB of HBM4 and targeting the Helios platform, is increasingly viewed as a viable alternative to Nvidia's Blackwell architecture. With production shipments underway and sampling for the MI450 already in progress, AMD is positioning itself to capture market share in the high-end data center segment where Nvidia's pricing power is being tested by memory cost volatility.
Local view
Korean Media Focus on Supply Chain Orders and Margin Pressure
South Korean financial outlets are closely monitoring the impact of memory costs on Nvidia's margins and the subsequent price pass-through to customers. Seoul Economic Daily reported that Nvidia is raising rack prices for Grace Blackwell 300 and Vera Rubin 200 by up to 17%, directly attributing this to DRAM and HBM price surges. The media highlights how this squeeze affects major Korean clients and suppliers, noting that Nvidia's gross margin guidance was adjusted to 71-72% due to these component costs.
China's Grey Market and Export Restrictions
Chinese media continues to report on the illicit trade of Nvidia's restricted Blackwell chips. Despite official export bans, reports from cnBeta and Sina Finance indicate that grey-market prices for DGX B300 systems have doubled or tripled, reaching over $1 million per unit. This underscores the persistent demand gap in China for high-end AI compute, which Nvidia is attempting to address with specialized, lower-spec chips slated for late 2026 shipment.
Context & numbers
- Nvidia Revenue: Q2 FY2027 revenue hit $96.2 billion, beating estimates of $92.07 billion. EPS was $2.22 vs. $2.09 expected.
- Price Increases: AI server prices are rising >15% (some reports say up to 17%) for Vera Rubin and Grace Blackwell systems, driven by HBM costs.
- RTX 50-Series: Retail prices for consumer RTX 50-series cards have jumped up to 39% in response to supply constraints and demand.
- B200 Pricing: Estimated build cost for a B200 GPU is ~$6,750, with market prices firm near $40,000. HBM constitutes ~48% of the build cost.
- Lead Times: Enterprise lead times for B200/GB200 hardware have improved from 12–24 weeks to 8–16 weeks, though backlog remains significant at ~3.6 million units.
On the radar
- Nvidia Rubin Launch: Full-scale Rubin systems (576 GPUs) are being prepared for hyperscaler labs, with expectations of 4x the FLOPs of Blackwell Ultra. Watch for initial deployment benchmarks in Q4 2026.
- AMD MI450 Sampling: AMD is currently sampling the MI450, with production shipments expected to ramp up in late 2026/early 2027. This will be a critical test against Nvidia's Rubin roadmap.
- China-Specific Chips: Nvidia plans to begin small-volume shipments of China-specific AI chips before the end of 2026. Regulatory clarity and actual adoption rates will be key signals for future China revenue exposure.
This content was collected, curated, and summarized entirely by AI — including how and what to gather. It may contain inaccuracies. Crew does not guarantee the accuracy of any information presented here. Always verify facts on your own before acting on them. Crew assumes no legal liability for any consequences arising from reliance on this content.
Powered by
