To The Moon

$NVIDIA Corp(NVDA)$ launched the world's strongest GPU H200 yesterday, driving the stock price to achieve 9 consecutive rises. It is the longest consecutive rise record in nearly 7 years, approaching the all-time high!

According to reports, H200 has 141GB of memory, which directly increases 76% compared with the 80GB of H100, and the memory bandwidth also increased from 3.35TB/s to 4.8TB/s, an increase of 43%. It will be compatible with the H100, and the reasoning speed is almost twice that of latter:

H200 is also the first GPU to offer HBM3e, a faster, larger memory that accelerates generative AI and large language models while advancing scientific computing for HPC workloads. With the HBM3e, the H200 delivers 141GB of memory at 4.8TB/s, nearly doubling the capacity and 2.4 times more bandwidth than its predecessor, the A100!

As the leader of AI chip, H200 is still not the limit. According to Nvidia's research and development progress, B100 will be launched in 2024, with more powerful performance:

H200 will be delivered in the second quarter of next year, $Alphabet(GOOG)$ $Alphabet(GOOGL)$ $Microsoft(MSFT)$ and other giants have confirmed the purchase!

At present, the GPU market mainstream players only Nvidia, $Advanced Micro Devices(AMD)$ and $Intel(INTC)$, of which, Nvidia is the undisputed overlord. AMD followed, it will ship MI300 in the fourth quarter, and expected to generate more than $2 billion in revenue next year. Intel is also preparing, plans to enhance the HBM capacity of Gaudi AI chips, but as Intel's business is more complex, investors are not optimistic.

As the large model continues to iterate, the demand for computing power is almost unlimited, and the GPU used will also multiply. According to market, the GPT-4 is about 10,000-25,000 A100 training. And Musk said, the GPT-5 may require 30,000-50,000 H100.

Currently, AI applications have not yet broken out, and large companies are still in the stage of research models. If more AI explosive products are launched in the future, the downstream demand for computing power will be higher, or it will push Nvidia to a higher position.

According to media reports, Nvidia has confirmed the addition of $Taiwan Semiconductor Manufacturing(TSM)$ CoWoS advanced packaging orders in October, which may indicate that AI chips are still exploding.

After hours next Tuesday, Nvidia will report results for the quarter to the end of October. Analysts expect revenue of $16 billion, up 170% from a year earlier, and revenue for the next quarter is expected to reach 17.7 billion:

It is expected that Nvidia will reveal the demand for AI next year on the earnings call. Nvidia could create a miracle again if there is any optimistic information!