强于大市(维持评级):传媒英伟达:Groq赋能推理算力

Investment Rating - The industry rating is "Outperform the Market," indicating that the overall return of the industry is expected to exceed the market benchmark index by more than 5% in the next 6 months [14]. Core Insights - Nvidia has formed a strategic partnership with AI chip startup Groq to enhance inference technology, with Groq's core team joining Nvidia while continuing to operate independently [3]. - The inference market is experiencing rapid expansion, with a significant increase in demand for inference computing power as the global AI industry transitions from model training to large-scale inference applications [4]. - Groq's technology, featuring LPU and SRAM architecture, optimizes inference performance, allowing for predictable, low-latency execution of large language models [5]. Summary by Sections Event Background - Nvidia and Groq have established a non-exclusive licensing agreement for inference technology, with Groq's leadership team integrating into Nvidia to advance technology upgrades and applications [3]. Industry Trends - The demand for inference computing power is growing rapidly, with Google processing 980 trillion tokens monthly as of July, doubling since May, and domestic models exceeding 50 trillion daily calls, a tenfold increase year-on-year [4]. Groq's Technical Value - Groq's LPU design focuses on speed and accuracy, utilizing static scheduling and deterministic execution to enhance performance in inference tasks. The second-generation LPU is manufactured using Samsung's 4nm process technology, improving speed and efficiency [5]. Investment Recommendations - The report suggests a positive outlook on the growth of inference computing demand and related supply chains, recommending attention to wafer fabrication and upstream equipment expansion in the domestic market, as well as growth in optical modules and cabinet assembly outsourcing in the overseas market [6].