Given the ongoing AI-enabled GPU gold rush, it is hardly a surprise that NVIDIA's GPU Technology Conference (GTC) has become an eyeball magnet on Wall Street, with serious analysts publishing note after note on what to expect during NVIDIA's highlight event of the year.
JPMorgan’s Outlook on NVIDIA’s GTC
Blackwell Ultra: A Performance Beast on the Horizon
JPMorgan expects the Blackwell Ultra (GB300) to be the highlight of this year’s GTC conference. This chip will feature a logic structure similar to the B200 but with significant improvements…
— Jukanlosreve (@Jukanlosreve) March 14, 2025
Now, JP Morgan is out with its take on what to expect during NVIDIA's GTC 2025.
Of course, the Blackwell Ultra GPU (GB300) is expected to be the proverbial star of the show. The Wall Street titan expects the new chip to feature a logic structure similar to the one found in the B200 chip, but with a High Bandwidth Memory (HBM) capacity of 288GB and leveraging the HBM3e 12-high stacking technology, Thermal Design Power (TDP) of up to 1.4kW, and a 50 percent performance boost in FP4 computing. Moreover, JP Morgan expects NVIDIA to start shipping the GB300 chips in Q3 2025.
Additionally, even though NVIDIA's next-gen Rubin architecture will enter the production phase only in 2026, JP Morgan does expect the GPU giant to reveal incremental details. As such, the Wall Street titan expects Rubin GPUs to feature:
- A logic structure similar to Blackwell's but equipped with 2 TSMC N3 process chips in what is essentially a dual logic chip structure.
- Eight HBM4 stacks with a total capacity of 384GB.
- A TDP of ~1.8kW
- Vera ARM CPU upgraded to TSMC's N3 process and leveraging a 2.5D packaging structure.
- 1.6T Network featuring two ConnectX-9 NICs.
- Possible introduction of NVL144 and NVL288 rack structures.
- Initial production to commence in late 2025 or early 2026, with mass shipments expected to commence by Q2 2026.
Moreover, JP Morgan believes that NVIDIA will reveal additional details about its Co-Packaged Optics (CPO) technology roadmap, which is intended to increase bandwidth, reduce latency, and lower power consumption. However, GPU-level CPO continues to face significant technical challenges, including heat dissipation issues (due to high optical engine heat generation), reliability concerns, and substrate deformation risks. Therefore, the Wall Street titan only expects a broad adoption of CPO in 2027, with the 2026 Rubin iterations likely to only see this technology incorporated as an optional feature within switches such as the Quantum (InfiniBand) and Spectrum (Ethernet).
Meanwhile, as we noted recently, NVIDIA filed for a new patent - bearing the publication number US20250078199A1 - on the 06th of March, 2025. The patent envisions discrete sections of a GPU working within local confines to store and access data, and perform computations, thereby reducing the delays that are inherent in accessing distant computational resources. It is possible that NVIDIA might reveal new details in relation to its ongoing sub-GPU localization efforts at the GTC.
About the author: Writing is my one incontrovertible passion. Over the past six years, he has authored over 2,200 distinct articles on financial and tech-related topics, spanning nearly 1 million words. And he has been a member of Wcctech mobile team since 2025. As an alumnus of the University of Toronto, Rotman Commerce Program, I bring nuance, in-depth knowledge, and a unique perspective to every topic that I cover. When I'm not writing, I'm traveling the world, exploring hidden confectionaries and restaurants as an aspiring food connoisseur.
Follow Wccftech on Google to get more of our news coverage in your feeds.
