Nvidia Reports Record Profit, Stock Stays Flat
Nvidia reported a record $43 billion profit and a 75% surge in Q4 datacenter revenue, outperforming analyst expectations. Despite the strong results driven by high demand for AI chips, the company's stock traded flat as investors signaled a desire for greater capital returns and questioned the sustainability of its growth.
- During the earnings call, CEO Jensen Huang announced that the company returned $41.1 billion to shareholders through repurchases and dividends in fiscal year 2026 and has $58.5 billion remaining in its buyback authorization. The company also projected revenues of $78 billion for the next quarter, significantly exceeding analyst expectations. - Competitor AMD is gaining traction with its MI300X accelerator, which offers 192GB of HBM3 memory, compared to the 80GB in Nvidia's H100, a key selling point for customers like Microsoft Azure and Meta. While Nvidia's CUDA software ecosystem remains a significant advantage, AMD's ROCm platform is improving its compatibility. - Intel launched its Gaudi 3 AI accelerator, claiming it offers 50% better inference and 40% better power efficiency on average than Nvidia's H100 for certain models. Intel is making Gaudi 3 available to OEMs like Dell, HPE, Lenovo, and Supermicro in various form factors. - Cerebras Systems unveiled its third-generation Wafer Scale Engine, the WSE-3, a 5nm chip with 4 trillion transistors and 900,000 AI cores. The WSE-3 delivers 125 petaflops of peak AI performance and is designed to train models up to 10 times larger than GPT-4. - Hyperscalers are increasingly designing their own custom silicon to optimize for specific workloads and reduce costs, a trend that poses a long-term challenge to Nvidia's dominance. Google's TPU v5p, for instance, can train large language models 2.8 times faster than its previous generation, the TPU v4. - Amazon has consolidated its custom AI chip efforts by halting the development of its Inferentia chip to focus on its Trainium line, which now handles both training and inference. The upcoming Trainium 3 chip, built on a 3nm process, is expected to double the performance of Trainium 2. - The market for Application-Specific Integrated Circuits (ASICs) in AI servers is projected to grow from 20.9% in 2025 to 27.8% this year, as their power efficiency and performance for specific tasks attract more customers. ASICs can deliver 30-40% better power efficiency compared to general-purpose GPUs for their intended workloads. - Despite the intense competition, Nvidia still holds over 80% of the AI chip market share. The company's upcoming Vera Rubin platform is expected to further reduce the costs associated with AI model training and inference.
