NVIDIA Rubin GPU Shipment Forecast Lowered: Market research firm TrendForce predicts that NVIDIA's next-generation Rubin GPU's shipment share in 2026 has been revised down from the previous 29% to 22%. Supply chain challenges (such as HBM4 verification, network interconnect upgrades, and liquid cooling optimization) are the main reasons, which may lead enterprises to extend the use cycle of the Blackwell architecture and accelerate diversification toward AMD and custom chips.
Blackwell Architecture Shipment Share Rises: Due to the Rubin delay, NVIDIA's Blackwell platform is expected to see its shipment share increase significantly to over 71% in 2026, further consolidating its market dominance. TrendForce points out that geopolitical risks and extended supply chain tuning cycles are key factors affecting the shipment structure.
H100 GPU Rental Price Surges Nearly 40%: According to SemiAnalysis data, the one‑year rental price of NVIDIA H100 GPUs soared from $1.70 per hour in October 2025 to $2.35 per hour in March 2026. Strong market demand for GPU computing power has led to a sharp jump in AI chip rental prices, further exacerbating the computing power shortage.
US Proposes Comprehensive Tightening of AI Chip Export Licenses: The US Department of Commerce is drafting new rules that would require NVIDIA, AMD, and others to obtain a license to export AI accelerators to any country in the world. This move would make the US government the de facto gatekeeper of global AI computing flows, whereas previously only about 40 countries were restricted.
CoreWeave Secures $8.5 Billion GPU‑Backed Loan: AI cloud service provider CoreWeave obtained $8.5 billion in debt financing from a consortium of lenders, using its NVIDIA GPUs as collateral, marking the rise of “ComputeFi” (computing‑power‑backed finance). In addition, Meta recently made additional investments in CoreWeave's AI infrastructure.
NVIDIA Leads AI Data Center Financing: NVIDIA recently co‑led a $505 million financing round for data center developer Firmus Technologies, planning to deploy next‑generation AI data centers in the Asia‑Pacific region, further solidifying its AI infrastructure ecosystem position.
Chinese Cloud Providers Including Tencent Cloud Raise AI Computing Prices: On April 9, Tencent Cloud announced a 5% uniform increase in the list prices of AI computing, container services, and EMR products effective May 9. Alibaba Cloud and other vendors had previously raised prices, signaling the end of the “cheap AI computing” era and highlighting the growing supply‑demand gap.
Domestic AI Accelerator Card Market Share Exceeds 40%: IDC data shows that total AI GPU shipments in the Chinese market in 2025 reached 4 million units, of which domestic semiconductor vendors shipped 1.65 million units, capturing 41% of the domestic AI server market. NVIDIA's share of the Chinese AI GPU market has dropped from 95% before sanctions to 55%.
Shanghai 10,000‑GPU Cluster Achieves Ultra‑High Availability: Shanghai INESA's Smart Computing Technology team successfully achieved 99.99% ultra‑high availability for a 10,000‑GPU cluster, reducing the total annual failure time to less than one hour. This marks a significant advancement in China's ultra‑large‑scale GPU cluster coordination technology.
Global AI Computing Power Concentrated in US Cloud Providers: An Epoch AI research report points out that more than 60% of global AI computing power is controlled by US hyperscale cloud providers. Among them, Google, with approximately 4 million H100‑equivalent self‑developed TPU chips, has become the world's largest single holder of AI computing power, with far less dependence on NVIDIA than Microsoft, Amazon, and other peers.