Nvidia 財報面臨 AI 審視

🔑 增強重點摘要

• •Wall Street anticipates Nvidia's Q4 FY2026 revenue at $65.6-65.91 billion with EPS of $1.52, driven primarily by $60 billion from data center business amid ongoing AI capex surge exceeding $650 billion in 2026.[2][3][4]
• •Nvidia's previous Q2 FY2026 results showed $46.7 billion revenue, up 56% YoY, with Blackwell Data Center revenue up 17% sequentially and no H20 sales to China but benefits from $830 million in H20-related gains.[1]
• •Rising competition includes Meta's multi-year deal for billions in AMD chips, Anthropic's $10 billion Broadcom/Google TPUs purchase, and U.S. stake in Intel, prompting AI firms to diversify from Nvidia reliance.[4]
• •Analysts highlight higher bandwidth memory costs due to shortages potentially pressuring margins, despite strong hyperscaler demand and no signs of AI spending slowdown.[3]
• •Updates expected on Vera Rubin next-gen rack-scale AI system, following Meta's February deal for millions of Nvidia chips including Rubin, valued in tens of billions.[5]

🛠️ 技術深入

• •Blackwell platform delivers generational leap in AI performance, with production of Blackwell Ultra ramping at full speed and extraordinary demand.[1]
• •NVIDIA NVLink enables rack-scale computing, supporting orders-of-magnitude increases in training and inference for reasoning AI models.[1]
• •Vera Rubin is Nvidia's next-generation rack-scale AI system, with six million Blackwell GPUs shipped in prior four quarters and $500 billion projected sales across Blackwell/Rubin.[5]
• •Announced RTX PRO 4000 SFF Edition and RTX PRO 2000 Blackwell GPUs for professional visualization.[1]
• •Full-stack NVIDIA DRIVE AV software platform now in full production for intelligent transportation.[1]

🔮 前景展望AI analysis grounded in cited sources

⏳ 時間線