Analysis: Jensen Huang must demonstrate the Groq integration plan today; otherwise, the custom chip narrative will dominate.

According to CoinWorld, based on monitoring by 1M AI News, renowned industry analyst Patrick Moorhead released a lengthy analysis before the opening of GTC 2026. The core judgment is: the true test of today’s Jensen Huang speech is whether he can demonstrate a complete roadmap for the coordinated operation of training GPUs, pre-fill accelerators, Groq decoding processors, and CPUs under a unified software layer. Achieving this would mark NVIDIA’s platform transformation at GTC 2026; failing to do so, the narrative will shift toward large cloud providers developing their own chips.

The confirmed facts Moorhead outlined include: the Vera Rubin NVL72 rack (72 Rubin GPUs + 36 Vera CPUs, interconnected via NVLink 6 at 3.6TB/s per GPU) is already deployed across AWS, Google Cloud, Microsoft, and Oracle cloud providers, with mass production ramping up in the second half of the year; Rubin GPUs achieve 5 times the inference performance of Blackwell with 1.6 times the transistor count, reaching 50 Petaflops for inference and 35 Petaflops for training; the $20 billion Groq acquisition has been completed, under a non-exclusive licensing framework, with founders Jonathan Ross and about 80% of the engineering team introduced simultaneously, surpassing the scale of the 2019 $7 billion Mellanox acquisition.

Moorhead’s forecast for today’s speech includes the official launch of NemoClaw (NVIDIA’s open-source platform for enterprise AI agents) and a roadmap for the 2028 architecture Feynman, which reports suggest will use TSMC’s A16 1.6nm process; Ross is also expected to appear on stage. He also pointed out three risks: Groq’s integration at the scale of large cloud providers has not yet been validated, and the $20 billion investment in unproven technology is costly; energy constraints will be the biggest variable in 2027, with nearly 40% of new data centers concentrated in the energy-rich Texas, while coastal regions face real bottlenecks; NVIDIA’s data center AI market share is expected to shrink from over 90% in the next two years to about 70%.