&#34;Meta Platforms' new goal is AGI. In order to achieve this goal, we will need approximately 350,000 NVIDIA chips by the end of this year. If we include other GPUs, the total computing power will be equivalent to nearly 600,000 H100 chips.&#34;
<img src="https://imageproxy.pbkrs.com/https://wpimg-wscn.awtmt.com/1871cdba-be7b-47f6-b57c-21ce87ac8f26.png?x-oss-process=image/auto-orient,1/interlace,1/resize,w_1440,h_1440/quality,q_95/format,jpg" alt="" original-src="https://imageproxy.pbkrs.com/https://wpimg-wscn.awtmt.com/1871cdba-be7b-47f6-b57c-21ce87ac8f26.png"/>
In the early morning today, Zuckerberg announced this news in his latest Instagram post, demonstrating his determination to develop AI.
&#34;600,000 is an astonishing number, and the cost of purchasing the chips may be even more astonishing, conservatively estimated to be around 9 billion US dollars.&#34;
Raymond James analyst pointed out that the price of each NVIDIA chip is between $25,000 and $30,000. Although Huang Renxun has said &#34;the more you buy, the more you save,&#34; even with large-scale procurement, it may still cost over 9 billion US dollars.
The AI gold rush continues to heat up, and &#34;shovel sellers&#34; like NVIDIA have become the biggest beneficiaries. And this is just Meta Platforms alone. As tech giants invest heavily to compete for the leading position in AI, a flood of wealth is pouring towards the &#34;shovel sellers.&#34;
Facing the money that is being handed to them, no one would refuse. The key is whether they have the ability to catch it, and it seems that NVIDIA alone may have a hard time swallowing it all.
<h2>NVIDIA may not be able to swallow all the GPU wealth</h2>
Since the AI boom has driven the demand for chips, NVIDIA has been plagued by &#34;capacity constraints.&#34;
Taking Meta Platforms as an example, it seems that NVIDIA is struggling to meet its demand.
According to previous reports, NVIDIA sold nearly 500,000 A100 and H100 GPUs in the third quarter of last year. Due to the high demand, the delivery cycle for H100 is as long as 36 to 52 weeks.
Research firm Omdia pointed out that Meta Platforms and Microsoft are the largest buyers of NVIDIA GPUs, with each purchasing about 150,000 H100 chips in Q3.
However, H100 was not released until the end of 2022, and it is worth noting that Omdia estimated that by the end of last year, the delivery cycle for H100 had already reached 52 weeks. This means that, according to Meta Platforms' demand, NVIDIA's GPUs may have a hard time meeting its computing power requirements.
In order to meet the planned demand, the computing power needs to be filled either by self-developed chips or by purchasing from NVIDIA's competitors, AMD/Intel.
Some netizens commented:
<blockquote>
It's not surprising that the demand may come from two companies. On the one hand, NVIDIA has limited chip production capacity, and on the other hand, as a buyer, they will definitely have a second supplier to negotiate or ensure supply security.
But ultimately, the bottleneck lies with TSMC. Currently, all cutting-edge GPUs are produced by TSMC, including Intel's GPUs.
<img src="https://imageproxy.pbkrs.com/https://wpimg-wscn.awtmt.com/16bce1ea-3964-4c85-a6db-8ab733ed4a78.png?x-oss-process=image/auto-orient,1/interlace,1/resize,w_1440,h_1440/quality,q_95/format,jpg" alt="" original-src="https://imageproxy.pbkrs.com/https://wpimg-wscn.awtmt.com/16bce1ea-3964-4c85-a6db-8ab733ed4a78.png"/>
</blockquote>
<h2>AMD may also benefit from this</h2>
From the perspective of NVIDIA's competitors, AMD is most likely to take a share of Meta Platforms' remaining demand.
AMD is directly competing with NVIDIA and has released the &#34;most advanced AI accelerator&#34; MI300, which runs on Meta Platforms' Llama 2.
Lisa Su, CEO of AMD, demonstrated the performance of the newly released AMD Instinct MI300X chip at an AI conference, stating that when running the Meta Platforms LLAMA 2-70B model, the performance of MI300X is 1.2 times that of NVIDIA.
Su also pointed out in the demonstration that when inferring Llama 2, a server composed of 8 MI300X chips runs 1.4 times faster than H100, clearly demonstrating the superior performance of MI300X in Llama 2.
<img src="https://imageproxy.pbkrs.com/https://wpimg-wscn.awtmt.com/cdb14d5e-6b25-40a5-ba34-1f424780ca31.png?x-oss-process=image/auto-orient,1/interlace,1/resize,w_1440,h_1440/quality,q_95/format,jpg" alt="" original-src="https://imageproxy.pbkrs.com/https://wpimg-wscn.awtmt.com/cdb14d5e-6b25-40a5-ba34-1f424780ca31.png"/>
The most groundbreaking news is that Meta Platforms will collaborate with AMD to build data centers using MI300X.
The reason why the two parties can come together is due to their love for open source and long-term cooperation.
Meta Platforms' Llama 2 is an open-source large model, and AMD's computing platform Radeon is also an open-source project.
In addition, AMD and Meta Platforms have had long-term cooperation in the past. Since 2019, Meta Platforms has been working with AMD to develop EPYC CPUs. Meta Platforms has recently deployed servers based on AMD Genoa and Bergamo in its infrastructure on a large scale.
In the future, AMD may get a share of the &#34;chip king&#34; NVIDIA's market, which may be the reason for AMD's skyrocketing this week and the analysts raising their target price.
This Tuesday, AMD's stock price surged, rising 8% to reach its highest closing price since November 2021, only about 2% lower than its historical peak.
Analysts have been raising their target prices for AMD, with Barclays raising its target price by $80. They predict that the company's AI chip sales may reach $4 billion this year and exceed $7 billion by 2025. O'Malley has given AMD a &#34;buy&#34; rating, citing strong demand for AMD's highest-end AI chip, the MI300.
At the same time, TSMC's latest earnings report brought good news and also drove AMD and NVIDIA's stock prices to new highs.
In terms of performance guidance, TSMC expects its first-quarter revenue to decline by around $18-18.8 billion, but it will grow in each subsequent quarter, with HPC being one of the key driving factors.
As the demand for chips remains high, major chip giants are expected to benefit further in the gold rush of AI.

NVDA

As AI continues to thrive and demand remains high, major chip giants have the potential to share in NVIDIA's success.

The "prosperity beyond measure" of GPUs, NVIDIA alone cannot handle