AI investment exceeding expectations triggers Meta's sharp decline? This blame should not fall on GPUs

Friends who are interested in AI must be paying close attention to Meta's performance today. Although the company's guidance is lower than expected, it announced during the earnings call that it will continue to increase its investment in AI.

The company continued its strong growth momentum in the first quarter, with overall revenue and profits exceeding analysts' expectations and reaching the upper limit of the company's Q1 guidance. However, due to the guidance falling short of expectations, it shattered the confidence of the bulls, leading to a significant drop in after-hours trading.

In terms of AI, which is the most closely watched aspect by the market, the company once again announced an increase in capital expenditure for AI data centers, achieving growth for two consecutive quarters.

1. Increase Investment in AI, Significantly Raise 2024 Full-Year Capital Expenditure Forecast

During the earnings call, the company guided that the full-year capital expenditure for 2024 will be in the range of $35 billion to $40 billion, higher than the previous range of $30 billion to $37 billion, far exceeding market expectations. The increase in capital expenditure this time is still to support its ambitious plans in the field of AI.

In addition, capital expenditure will also be used for:

• Reality Labs —— R&D for VR/AR;

• Chip development;

From a financial perspective, the company currently has nearly $32.3 billion in cash and cash equivalents, with a cash flow of approximately $12.531 billion in Q1 2024. This is sufficient to support the company's planned capital expenditure.

According to the company's early-year plan, by the end of 2024, the company will have 350,000 NVIDIA H100 GPUs. Including other GPUs it already owns, its total computing power is equivalent to 600,000 H100s. For comparison, training the 1.8 trillion parameter GPT-4 requires 8,192 H100 GPUs, and 600,000 H100s can support training nearly 74 models of the same size.

In 2023, Meta and Microsoft were tied for first place in purchasing 150,000 H100 GPUs. In 2024, Meta proposed a plan to "add 350,000 H100s to build a large-scale computing infrastructure," greatly enhancing the company's AI competitiveness.

The increase in capital expenditure this time is nearly $3 billion. If it is all used for the construction of AI data centers, factors such as energy supply, cooling systems, depreciation, and operating expenses need to be considered. According to calculations provided by industry experts, setting up 350,000 H100s would require an investment of nearly $8.75 billion (for cards only)

2. The Llama 3 model has been applied to various business platforms of the company

The capital expenditure on GPUs ultimately needs to be implemented on the disruptive player in the industry, the Meta open-source model Llama series.

On April 20th, the company released the 8B and 70B Llama 3 models, while also indicating the training of a 400B+ parameter Llama 3 model.

During the performance meeting, Meta mentioned that large AI models have gradually been applied to the company's products, achieving many remarkable results, specifically:

• Approximately 30% of Facebook's News Feed is published through an AI recommendation system;

• Over 50% of the content seen on Instagram is now recommended by artificial intelligence;

• After introducing AI model recommendations to Reels, user viewing time increased by 8% to 10%;

Currently, the Llama 3 model is also being applied to various business platforms of the company.

Clearly, the power of Llama 3 is beginning to show, posing a threat to closed-source models such as Open AI and Claude.

3. Chinese e-commerce's overseas expansion trend continues, AI empowers precise ad placements

First-quarter revenue saw a significant year-on-year increase, mainly due to the unrelenting demand of Chinese manufacturers for overseas expansion, the year-on-year increase in total ad impressions, and the rise in ad prices.

Overseas Expansion ——

In 2023, the trend of Chinese e-commerce companies expanding overseas surged again, with the demand for product expansion driving the need for overseas advertising. Throughout 2023, ad revenue from China accounted for 10% of Meta's total revenue, contributing to a 5-percentage-point growth. Data shows that in the first quarter of 2024, Chinese companies' enthusiasm for overseas advertising placements remains strong.

It is worth noting that the possibility of marketing budget reductions for Chinese e-commerce and gaming companies is mainly due to geopolitical factors (such as increased tariffs) and corporate budget cuts resulting from macroeconomic downturns.

Year-on-Year Increase in Ad Impressions ——

During the performance meeting, the company mentioned that AI models have been applied to the ad delivery system, achieving precise targeting, resulting in a 20% year-on-year increase in total ad impressions in Q1.

Of course, the increase in total ad impressions is also related to the continuous optimization of the company's business:

• AI models empower precise ad targeting

• Expansion of Threads user base • Increase in Reels user engagement • Whatsup launches Click-to-Messenger feature

Increase in advertising unit price ——

According to the company's disclosed financial reports, in the past five quarters, the advertising unit price has rebounded after a decline, and has increased year-on-year for two consecutive quarters.

It is speculated that this is due to the improvement of Meta's AI technology and infrastructure, which has enhanced the precision and effectiveness of advertising placements, thereby increasing the value of advertising.

4. Net profit doubled year-on-year

In the first quarter, revenue increased by 27%, but net profit increased by 116%, far exceeding the revenue growth rate. The main reasons are the growth in revenue, reduction in operating expense ratio, and decrease in tax rate.

From the financial data perspective, in Q1 2024:

• Revenue increased by 27%: driven by Chinese companies' overseas demand, increase in total ad impressions, and rise in advertising unit price; • Operating expense ratio decreased by 11%: due to the increase in revenue base and the large-scale restructuring and layoffs in 2023; • Tax rate decreased by 9%: the financial report did not specify the exact reason for the decrease in tax rate, and the company guided that the annual tax rate will remain at a moderate level.

5. Reality continues to incur losses, but within market expectations, not a negative factor

The company continues its dual strategy of focusing on AI and metaverse (VR/AR), planning to invest heavily in Reality Labs. The business generated 440 million in revenue in Q1 24, lower than the market's expected 496 million, and the department continues to incur losses. However, as early as the Q4 2023 earnings call, the company hinted that this business would increase losses in 2024, which is within market expectations and does not constitute a negative factor. During this communication meeting, the company reiterated its expectation that the operating losses of Reality Labs in 2024 will significantly increase year-on-year.

6. Increase in legal costs due to privacy issues and antitrust investigations

In the first quarter, the company's General and Administrative (G&A) expenses increased by 20%, mainly due to higher legal expenses. This increase is related to the recent antitrust lawsuits and investigations the company faces globally, as well as legal issues related to user privacy regarding the company's products

7. Post-market Stock Price Plunge

The strong performance in 2023 has made the investment market optimistic, with market expectations that Meta will continue to maintain a considerable growth rate. This means that Meta may need to deliver continuous above-expectation results and improve performance guidance to sustain its current stock price level. Therefore, despite Q1 revenue of $36.46 billion, higher than analysts' expectations of $36.12 billion, the performance guidance for Q2 was lower than expected, leading to a 16% plunge in the post-market.

Overall, the company's first-quarter performance data is still quite impressive. Although the Q2 guidance fell short of expectations, considering the company's consistently conservative guidance style, the market's reaction may have been somewhat exaggerated. However, the significant increase in capital expenditure on AI still reflects the industry's high prosperity.

Next, let's look at Microsoft and Google's capital expenditures