Nvidia and AMD Accelerators: Roadmaps, Pricing, Supply — 2026-10-28
TSMC is reportedly weighing a 20% price hike on CoWoS packaging, citing supply constraints for Nvidia's Blackwell and Rubin chips. Meanwhile, Nvidia's Vera CPU is gaining traction with hyperscalers, threatening AMD and Intel's market share, while massive Chinese orders for RTX PRO 5500 cards exceed global production capacity.
Nvidia and AMD Accelerators: Roadmaps, Pricing, Supply — 2026-10-28
Top developments
TSMC Weighs 20% CoWoS Price Increase
TSMC is reportedly considering a 20% increase in prices for its Chip-on-Wafer-on-Substrate (CoWoS) advanced packaging technology, driven by intense demand for Nvidia’s Blackwell and Rubin architectures. Equipment lead times for packaging have stretched to eight months, highlighting severe capacity bottlenecks in the AI supply chain. This potential hike could further inflate the cost of high-end AI accelerators as manufacturers struggle to keep pace with hyperscaler demand.
Nvidia Vera CPU Threatens AMD and Intel Market Share
Nvidia’s Vera CPU is actively shipping to major hyperscalers, with management expecting CPU revenue to more than double. This move places Nvidia in direct competition with AMD and Intel in the server CPU market, a domain traditionally dominated by the latter two. The shift suggests Nvidia is leveraging its GPU dominance to capture adjacent silicon revenue, potentially squeezing AMD’s Instinct roadmap positioning.
Chinese Tech Giants Order 3 Million RTX PRO 5500 GPUs
ByteDance, Alibaba, and Tencent have collectively ordered over 3 million RTX PRO 5500 GPUs from Nvidia, a volume that exceeds Nvidia’s previously announced global production cap of 3 million units for 2027. The orders, valued at approximately $35 billion, indicate massive unmet demand in China for high-performance AI workloads that do not trigger current US export restrictions. This surge creates a significant allocation challenge for Nvidia, as demand in China alone surpasses total global capacity.
RTX 5090 Supply Chain Halted
Evidence suggests Nvidia has effectively halted new production orders for the consumer RTX 5090 GPU. Board partners report an inability to order GB202 chip kits or place backorders, and cooling suppliers have received zero heatsink orders for Q2 2027. This move likely reflects a strategic reallocation of wafer capacity toward more lucrative data center accelerators like Blackwell and Rubin, amidst the aforementioned CoWoS constraints.
Local view
Chinese Media Report on Component Price Hikes
Chinese media outlets, including Sina and First Financial, report significant price increases for high-end consumer graphics cards in the Huaqiangbei electronics market, with some models rising over 1,000 RMB in a single week. This volatility is attributed to the spillover effects of AI data center demand, which is straining the supply of high-end memory and components. Additionally, reports indicate AMD has raised GDDR6 memory prices for its board partners since October 1, though specific percentages remain undisclosed.
Korean Press Highlights Supply Chain Investments
Korean media such as Chosun Ilbo and ZDNet Korea are focusing on Nvidia’s strategic investments in its supply chain to secure production capacity through 2028. Reports highlight multibillion-dollar deals with suppliers of substrates, optical components, and advanced packaging facilities. The press notes that these agreements are critical to mitigating the CoWoS bottlenecks that are currently limiting the shipment of Blackwell and Rubin architectures.
Context & numbers
Blackwell Pricing and Cloud Rates
As of October 2026, the list price for a DGX B300 system is reported at $646,878, while the DGX B200 stands at $562,154. Cloud rental rates for B300 instances have jumped 21% to $9.50 per hour, reflecting the scarcity and high demand for Blackwell Ultra capacity. The base B200 GPU price remains in the $30,000–$50,000 range, with effective per-GPU costs in large deployments often higher due to system integration and cooling requirements.
Market Forecasts
The used and refurbished data center GPU market is projected to grow at an 18% CAGR to reach $21.54 billion by 2036, driven by fleet refreshes as companies upgrade from Ampere to Hopper and Blackwell systems. The broader GPU accelerator card market is expected to reach $74.86 billion by 2031, fueled by AI infrastructure investment. Nvidia’s Blackwell and later architectures (including Rubin) are projected to become the largest architecture segment by 2036.
On the radar
- AMD GDDR6 Price Increases: Reports indicate AMD raised GDDR6 prices for board partners starting October 1, 2026. While specific percentages are unconfirmed, this move mirrors broader memory market inflation and could impact AMD Instinct MI300/MI350 series costs.
- HUMAIN Infrastructure Deployment: Saudi Arabian AI firm HUMAIN is deploying up to 84 MW of AI computing infrastructure, equivalent to approximately 27,000 Nvidia GPUs. Blackwell Ultra chip deployment is scheduled to begin in Q4 2026, with Vera Rubin integration planned for 2027.
- Google Cloud Rubin Expansion: Nvidia announced that Google Cloud plans to expand its Vera Rubin infrastructure to 960,000 Rubin GPUs across multiple sites, signaling massive early adoption of the next-generation architecture.
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
