Asia and Global Semiconductor and Tech Research

Andrew Lu

Equity Bottom-Up

<ul>
<li>Nvidia reports/guides a better than expected 3Q/4Q23 sales, margin, and EPS on stronger AI GPU sales growth of nearly 3x.</li>
<li>Nvidia reports a healthy 3.04 MOI, down 5% q/q and down 37% y/y and contributes nicely to account for 9% of TSMC sales.</li>
<li>In spite of concerns on good news priced in, seasonal weaker 1Q24, and MI300X/ASIC alternative AI solutions, we expect more raise to come in 2024-2025E.</li>
</ul>


Nvidia reports/guides a better than expected 3Q/4Q23 sales, margin, and EPS on stronger AI GPU sales growth of nearly 3x.
Nvidia reports a healthy 3.04 MOI, down 5% q/q and down 37% y/y and contributes nicely to account for 9% of TSMC sales.
In spite of concerns on good news priced in, seasonal weaker 1Q24, and MI300X/ASIC alternative AI solutions, we expect more raise to come in 2024-2025E.


As Expected from Our Earlier BOM/CoWoS Analysis, Consensus to Raise Nvidia Estimates Inevitably

Equity Long-Short, Semiconductors, Fabricated Knowledge

Douglas O'Laughlin

Equity Long-Short, Semiconductors

Fabricated Knowledge

Learn more about the world's most important manufactured product. Meaningful insight, timely analysis, and an occasional investment idea.

<ul><li>Nvidia, in characteristic fashion, murdered EPS estimates.</li><li>Revenue was $2 billion ahead of the guide, and the guide was another $4 billion ahead of consensus for the next quarter.</li><li>Going into the print, many people were concerned that positioning was bad, i.e., everyone was bullish and expectations were too high. Well, they did not disappoint.</li></ul>

Nvidia, in characteristic fashion, murdered EPS estimates.Revenue was $2 billion ahead of the guide, and the guide was another $4 billion ahead of consensus for the next quarter.Going into the print, many people were concerned that positioning was bad, i.e., everyone was bullish and expectations were too high. Well, they did not disappoint.

<ul>
<li>Nvidia CEO Jensen Huang has repeatedly emphasized that the data center is the new unit of compute.</li>
<li>While this concept seemed straightforward at first, the deeper implications became clearer after Nvidia's presentations at GTC and OFC 2024.</li>
<li>I only recently grasped exactly what is happening, and a simple reframing of the underlying principles that drove Moore’s Law makes the entire picture clearer.</li>
</ul>


Nvidia CEO Jensen Huang has repeatedly emphasized that the data center is the new unit of compute.
While this concept seemed straightforward at first, the deeper implications became clearer after Nvidia's presentations at GTC and OFC 2024.
I only recently grasped exactly what is happening, and a simple reframing of the underlying principles that drove Moore’s Law makes the entire picture clearer.


The Data Center is the New Compute Unit: Nvidia's Vision for System-Level Scaling

<ul>
<li>"It is so much cheaper to do simulations than real experiments, so much more flexible in testing...we can even do things which cannot be done in any lab… So it’s inevitable this trend will continue for some time." - Richard Hamming, <a href="https://www.amazon.com/Art-Doing-Science-Engineering-Learning/dp/1732265178" target="_blank" rel="noopener">The Art of Doing Science and Engineering</a>.</li>
<li>Richard Hamming said this to the President of Bell Labs in the 1950s. And I think this insight rings more accurate than ever. Simulations of reality are much more flexible and cheaper than observing and interacting with reality.</li>
<li>I will assert that LLMs and “AI” are effectively the logical conclusion of this trend. A large language model is trained on the largest amount of data possible, often the entire crawlable Internet<a href="https://kili-technology.com/large-language-models-llms/9-open-sourced-datasets-for-training-large-language-models" target="_blank" rel="noopener">, all the books ever written, and so much more</a>.</li>
</ul>


"It is so much cheaper to do simulations than real experiments, so much more flexible in testing...we can even do things which cannot be done in any lab… So it’s inevitable this trend will continue for some time." - Richard Hamming, The Art of Doing Science and Engineering.
Richard Hamming said this to the President of Bell Labs in the 1950s. And I think this insight rings more accurate than ever. Simulations of reality are much more flexible and cheaper than observing and interacting with reality.
I will assert that LLMs and “AI” are effectively the logical conclusion of this trend. A large language model is trained on the largest amount of data possible, often the entire crawlable Internet, all the books ever written, and so much more.


China New Economy Equity Analyst, Blue Lotus Research Institute

Robert McKay

China New Economy Equity Analyst

China growth and global non-US Internet equities

Blue Lotus Research Institute

Blue Lotus Research is a Hong Kong Type 4 Licensed investment advisory firm focusing on Chinese growth equities and global non-US Internet equities

<ul>
<li>Nvidia's H20 and upcoming B20 were launched to comply with China export controls, and have remained successful due to deficiencies and supply issues for Huawei's Ascend AI accelerators.</li>
<li>However, Nvidia's China success will be short-lived due to increasing LLM compute requirements and no method for it to offer a more performance dense offering due to export controls.</li>
<li>Not only will Huawei's upcoming Ascend AI accelerator mitigate this issue, but it will likely also aim to address some of the other deficiencies of the Ascend platform too.</li>
</ul>


Nvidia's H20 and upcoming B20 were launched to comply with China export controls, and have remained successful due to deficiencies and supply issues for Huawei's Ascend AI accelerators.
However, Nvidia's China success will be short-lived due to increasing LLM compute requirements and no method for it to offer a more performance dense offering due to export controls.
Not only will Huawei's upcoming Ascend AI accelerator mitigate this issue, but it will likely also aim to address some of the other deficiencies of the Ascend platform too.


Nvidia's China Dominance in AI Accelerator Has Serious Deficiencies

United States

NVIDIA Corporation designs, develops, and markets three dimensional (3D) graphics processors and related software.  The Company's products provide interactive 3D graphics to the mainstream personal computer market.

Price Gauge
52-week Range	0.00 - 0.00
Market Cap (USDm)	0.00
P/E	0.00
Dividend (Yield)	0.00

NVIDIA Corp (NVDA US)

As Expected from Our Earlier BOM/CoWoS Analysis, Consensus to Raise Nvidia Estimates Inevitably

As Expected from Our Earlier BOM/CoWoS Analysis, Consensus to Raise Nvidia Estimates Inevitably. However, investors see good news priced in in a...

Andrew Lu

Nvidia: Better than Dreamed

Think about how good Nvidia's earnings are. It's even better than that. Think about it again. It's better still. It's that good. Jokes aside,...

Fabricated Knowledge

The Data Center is the New Compute Unit: Nvidia's Vision for System-Level Scaling

Copper, Cooling, and Compute Density: The Pillars of Nvidia's Data Center Vision. Nvidia CEO Jensen Huang has repeatedly emphasized that the data...

Fabricated Knowledge

Jensen's World: Compressing Reality

Nvidia is pursuing a strategy of broadening its software distribution on one of the largest installed bases in the world.

Fabricated Knowledge

Nvidia's China Dominance in AI Accelerator Has Serious Deficiencies

Nvidia's success in China may be short-lived due to the increasing number of failure points as model compute scales. Meanwhile, Huawei's Ascend...

Blue Lotus Research Institute

Price Gauge

52-week Range

Market Cap (USDm)

P/E

Dividend (Yield)

As Expected from Our Earlier BOM/CoWoS Analysis, Consensus to Raise Nvidia Estimates Inevitably

As Expected from Our Earlier BOM/CoWoS Analysis, Consensus to Raise Nvidia Estimates Inevitably. However, investors see good news priced in in a...

Andrew Lu

Nvidia: Better than Dreamed

Think about how good Nvidia's earnings are. It's even better than that. Think about it again. It's better still. It's that good. Jokes aside,...

Fabricated Knowledge

The Data Center is the New Compute Unit: Nvidia's Vision for System-Level Scaling

Copper, Cooling, and Compute Density: The Pillars of Nvidia's Data Center Vision. Nvidia CEO Jensen Huang has repeatedly emphasized that the data...

Fabricated Knowledge

Jensen's World: Compressing Reality

Nvidia is pursuing a strategy of broadening its software distribution on one of the largest installed bases in the world.

Fabricated Knowledge

Nvidia's China Dominance in AI Accelerator Has Serious Deficiencies

​Nvidia's success in China may be short-lived due to the increasing number of failure points as model compute scales. Meanwhile, Huawei's Ascend...

Blue Lotus Research Institute

Nvidia's success in China may be short-lived due to the increasing number of failure points as model compute scales. Meanwhile, Huawei's Ascend...