JPMorgan expects that NVIDIA will launch the Blackwell Ultra chip (GB300) at the conference and may disclose some details of the Rubin platform, as well as focus on the comprehensive upgrade of AI Hardware, including higher performance GPUs, HBM memory, stronger cooling and power management, and the CPO (co-packaged optics) technology roadmap. In addition, JPMorgan anticipates that humanoid robots and physical AI will also be highlights of the conference.
Blackwell Ultra leads the way, the Rubin platform shows its first signs, and AI Hardware undergoes a comprehensive upgrade……$NVIDIA (NVDA.US)$What surprises will be revealed at GTC 2025.
On March 18, NVIDIA is about to hold the 2025 Global AI Conference for Developers. Yesterday, JPMorgan Analysts Gokul Hariharan, Albert Hung, and others released a report for the forecast.
JPMorgan expects that NVIDIA will launch the Blackwell Ultra chip (GB300) at the conference and may disclose some details of the Rubin platform. This conference will also focus on the comprehensive upgrade of AI Hardware, including higher performance GPUs, HBM memory, stronger thermal and power management, as well as the CPO (Co-Packaged Optics) technology roadmap.
Moreover, JPMorgan expects humanoid robots and physical AI to also be highlights of the conference - for investors, this means NVIDIA will continue to lead in AI Hardware innovation, and relevant industry chain companies are likely to benefit. However, investors also need to pay attention to the risk of slowing growth in Datacenter AI spending.
Currently, the overall sentiment in the market about AI remains pessimistic, with main concerns including the peak of Datacenter AI spending in 2025, competition between GPUs and ASICs, and the downward adjustment of Taiwan Semiconductor CoWoS Orders. But JPMorgan believes that the GTC conference should help revive the market's positive sentiment towards AI Stocks.
In the long term, global AI capital expenditure still has room for growth in 2026, mainly driven by increased spending from USA cloud vendors, a recovery in China's CSP capital expenditure, and rising enterprise-level AI demand.
Blackwell Ultra: a performance monster, ready to burst forth.
JPMorgan expects that Blackwell Ultra (GB300) will be the highlight of this GTC conference. This chip will use a Logic Chip similar to the B200 but will have significant improvements in HBM memory capacity and power consumption. JPMorgan anticipates:
HBM capacity skyrockets: 288GB, using HBM3e 12 high-stacking technology;
Power consumption surges: thermal design power (TDP) reaches 1.4kW;
Significant performance increase: expected to be 50% higher than the B200 in FP4 computing performance;
Shipping time: expected to start shipping in the third quarter of 2025.
JPMorgan also mentioned that the changes to the Blackwell Ultra system will benefit suppliers in areas such as power supply, Battery, cooling, connectors, ODM, and HBM.
Rubin platform: 2026 new engine for AI computing.
Although the Rubin platform is expected to be in mass production by 2026, JPMorgan believes that NVIDIA may share some details about the Rubin platform at this GTC conference. The Rubin GPU is expected to feature the following specifications:
Dual Logic Chip structure: similar to Blackwell, but equipped with two Taiwan Semiconductor N3 process chips;
HBM4 memory: 8 HBM4 cubes, total capacity of 384GB, an increase of 33% over Blackwell Ultra;
Power consumption continues to rise: is expected to reach around 1.8kW TDP;
Vera ARM CPU upgrade: expected to migrate to N3 process and may adopt a 2.5D packaging structure;
1.6T network: Rubin platform may adopt a 1.6T network architecture, equipped with two Connect X9 NICs;
NVL144/NVL288: NVL144 and NVL288 rack structures may be launched to improve GPU density;
Mass production time: the Rubin platform may have the possibility of an earlier mass production time, starting mass production as early as the end of 2025 or the beginning of 2026, with large-scale shipments expected to occur in the second quarter of 2026.
CPO technology: the future of Datacenter interconnection.
JPMorgan believes that NVIDIA's CPO (Co-Packaged Optics) technology roadmap will be another highlight of this GTC conference.
CPO helps improve bandwidth, reduce latency, and lower power consumption, but currently, GPU-level CPO still has significant technological challenges, such as heat dissipation issues (the optical engine generates a lot of heat), reliability, and the risk of deformation of the packaging substrate.
JPMorgan expects that CPO will initially be applied to Switches, specifically the Quantum (InfiniBand) and Spectrum (Ethernet) series, as an optional solution for the Blackwell Ultra platform, with widespread application of CPO on the GPU side expected to be realized at the earliest in 2027 during the Rubin Ultra era.
Physical AI and Humanoid Robots: Growing Attention.
JPMorgan stated that NVIDIA showcased progress in physical AI at the previous GTC conference, and given the acceleration in the development of humanoid robots and physical AI, market attention in this area may significantly increase during this GTC conference.
Currently, NVIDIA has released Cosmos (an AI foundational model platform) and GR00T (a humanoid robot development platform), and more announcements regarding multimodal AI, robotics, and Digital Twin are expected at this GTC conference.
JPMorgan also mentioned that the rising sentiment in the robotics field will benefit supply chain manufacturers such as BizLink and Sinbon.
Editor/ping
