💰 Financial nuclear explosion

-Single-quarter revenue captured US$39.3 billion (about RMB 284 billion), a year-on-year increase of 78%, surpassing Meta's annual revenue

-Single-quarter data center revenue almost doubled (93%) to US$35.6 billion (about RMB 250 billion), setting a historical record

-Data center business has increased by 10 times in two years, accounting for more than 90% of revenue (US 35.6 billion)

-Spend US$33.7 billion to repurchase stocks, creating a rare transaction in the history of technology

🚀 Technology hegemony

-Blackwell architecture chips exploded before mass production: they won 110 in the fourth quarter USD 100 million orders (Super Xiaopeng's annual revenue), Microsoft and Google take over half of them

-Inference AI computing power black hole: Demand for models such as DeepSeek R1, xAI Grok3, GPT-4 surged by a hundred times, and the efficiency of kilocard cluster training is 5 times higher than that of previous generations

-Technical generation difference crushed: The AI ​​generation efficiency of the US version of GB200 chip is 60 times higher than that of the Chinese special version

⚠️ Hidden worries emerge

-China's battlefield is in a hurry: revenue is impacted by the chip ban, and the performance of Huawei Ascend 910B has reached 80% of the H100

-Self-developed chip divergence: Microsoft Maia, Google TPU v5 erode 15% of the order, Blackwell Gross profit margin declined by 3% due to yield problems

-Consumer market stalled: gaming graphics card business unexpectedly declined by 11%, network equipment sales declined instead of increasing

🛠️ Huang's rule of breaking the deadlock

-"Software breaks through hardware shackles": Developers can bypass export restrictions through algorithm optimization, and the CUDA ecology may become the key to breaking the deadlock

-"Inference model is a demand catalyst": DeepSeek R1's long-thinking chain reasoning requires 100,000 servers to operate at full speed, forcing companies to invest in additional computing power

-Geo-game trump card: Global technology giants have hundreds of billions of dollars in AI infrastructure investment every year, and are still Nvidia's "stabilizing needle"

(Note: January 27 The "computing power efficiency revolution" in the day-long search for the "computing power efficiency revolution" triggered stock price fluctuations, and Huang Renxun urgently extinguished the fire. "This type of model will give rise to a hundred times the demand for chips"; Blackwell's single-card computing power has reached 5 times the previous generation, but the potential for heat dissipation may become a bottleneck for mass production. )

This financial report call will be like the real verification of Huang's law - when AI enters the "deliberate" era, Nvidia's computing power empire is still swallowing up global capital. But Huawei Ascend's attack and the wave of self-developed chips are causing the crown of the king of chips to crack for the first time.