Analysis: Huang Renxun needs to prove Groq's integration roadmap today, otherwise the custom chip narrative will prevail

According to 1M AI News's monitoring, prominent industry analyst Patrick Moorhead released an in-depth analysis before the opening of GTC 2026, with the key conclusion being: The true test of today's Huang Renxun's speech is whether he can demonstrate a complete roadmap for the collaborative operation of Training GPUs, Inference Accelerators, Groq Decoding Processors, and CPUs under a unified software layer. Success would mark GTC 2026 as NVIDIA's completion of platform transformation; failure would shift the narrative towards hyperscale cloud providers' in-house chip development.

The confirmed facts he outlined are as follows: The Vera Rubin NVL72 rack (72 Rubin GPUs + 36 Vera CPUs, NVLink 6 at 3.6TB/s per GPU interconnect) has been deployed in AWS, Google Cloud, Microsoft, and Oracle's four cloud factories, with mass production ramping up in the second half of the year; The Rubin GPU, with 1.6 times the transistor count of Blackwell, achieves 5 times the inference performance, reaching 50 Petaflops for inference and 35 Petaflops for training; The $20 billion Groq acquisition has been completed, utilizing a non-exclusive licensing framework, concurrently bringing in founder Jonathan Ross and around 80% of the engineering team, surpassing the scale of the $7 billion Mellanox acquisition in 2019.

Moorhead's predictions for today's speech: The official launch of NemoClaw (NVIDIA's open-source platform for enterprise AI agents); showcasing the roadmap for the 2028 Feynman architecture, which, according to analyst reports, will adopt TSMC's A16 1.6nm process; Ross is also expected to take the stage.

He also pointed out three risks: Groq's integration at hyperscale cloud levels has not yet been validated, and the $20 billion investment in untested technology is costly; Energy constraints are the biggest variable for 2027, with nearly 40% of new data centers concentrated in power-abundant Texas, but coastal regions face real bottlenecks; NVIDIA's data center AI market share is projected to compress from over 90% to around 70% in the next two years.