To The Moon

Nvidia's GTC Technology Conference (GTC 2024) kicked off yesterday, highlighted by a keynote from Jensen Hwang spanning roughly two hours. With over 300,000 registrations (including online) and 10,000 attendees in the audience, this highly anticipated event rivaled the excitement of a Taylor Swift concert in Singapore!

Ironically, Jensen Hwang began his address with a humorous remark, quipping, "Today is not a concert," despite his rock star appearance in a leather jacket.

In the keynote speech, Nvidia unveiled its next-generation Blackwell accelerated GPU computing architecture and the upcoming B200 chipset, specifically tailored for AI and data-center acceleration purposes.

However, aside from the debut of the Blackwell chip (which is already widely anticipated), there were no major revelations. Some topics had been previously addressed in past GTC events or earnings conference calls.

Perhaps due to this disappointment, Nvidia's stock price is currently declining in the after-market session. Oh well.

Key takeaway of the Keynote speech

1. Hwang underscored the significance of AI and accelerated computing as akin to an industrial revolution, with NVDA's GPU and simulation platform, Omnibus, playing a pivotal role.

2. The newly unveiled Blackwell GPU boasts a performance enhancement of 5x compared to the previous Hopper model. It is designed with two GPU dies connected and integrated with HBM3e memory totaling 192GB.

3. Numerous system-level solutions based on the Blackwell architecture (including HGX B200, HGX B100, GB200 NVL72) and Grace ARM-based CPU solutions (GB200, featuring 2 B200 chips and 1 Grace CPU) have been introduced to expedite time-to-market for customers and enhance product segmentation over time. These products are expected to become accessible to Nvidia’s partners later this year, and it’s anticipated that they will fuel continued revenue growth through 2025.

4. The rollout of the next-generation GPU copper interconnect, NVLink 5, across its platforms (reminiscent of the H100 model), coupled with the integration of 800GB InfiniBand and Ethernet optical networking connectivity (leveraging the 1.6Tbps PAM4 DSP chipset from Marvell), facilitates speeds of 200Gbps/channel.

5. NVIDIA Inference Microservice (NIM) offers pre-trained models or software solutions to customers seeking AI integration.

6. The AI Foundry service caters directly to customers looking to develop new applications.

7. Omnibus is deemed essential for integrating AI into industrial environments, including robotics.

8. The team is utilizing its Project GR00T foundation model (software designed for training and instructing robots) along with the Jetson Thor robotics computer (built upon the Blackwell architecture). In the automotive sector, the team is actively growing its automotive pipeline and has disclosed that BYD will be implementing DRIVE Thor.

Lacklustre market reaction

Despite the seemingly exciting keynote speech, Nvidia share price lost steam and seems like Nvidia’s latest unveiling failed to measure up to investors' high expectations.

Nevertheless, in my opinion, it's premature to feel disappointed because Nvidia is expected to host an Analyst Day during the GTC tonight, where they may provide insights on market size and earnings growth.

That said, Nvidia is still maintaining its momentum across all market segments. PC gaming is expected to be robust in 1H24, serving as a significant revenue catalyst for the company. Additionally, Nvidia is expected to see substantial growth in the data-center segment, driven by the increasing adoption of GPU-accelerated deep learning among hyperscale customers for handling extensive datasets. There is also strong optimism in the automotive and enterprise segment as well, although the extent of autonomous driving adoption in the market remains uncertain.

@MillionaireTiger @Daily_Discussion @TigerStars @CaptainTiger