Chip giants face a big test

Latest update time：2024-08-12

Reads：

????If you hope to meet more often, please mark the star ?????? and add it to your collection~

With all the second quarter 2024 earnings coming out, it’s time to dispel the clouds of excitement and panic, ignore performance claims and errors, and analyze the data center business, including examining the upstream and downstream supply chains. It’s time to see if the AI boom in the semiconductor industry is still going strong.

We begin our analysis with two major categories: processing and networking, and the top five semiconductor companies supplying data centers.

The top five semiconductor companies that provide products for data centers account for almost 100% of networking and processing. Overall growth in the second quarter of 2024 was again a healthy 15.3%, all of which came from processing. Of this, the networking business shrank slightly (-2.5%), while the processing business grew 20.3%. Since Nvidia said the company's decline in networking business was due to shipment adjustments, the growth figures may not represent a major shift in the underlying business.

From a year-on-year perspective, the overall growth was rapid, reaching 167%, with processing volume increasing by 211% and network volume increasing by 66%.

As can be seen from the operating profit graph, operating profit growth was even stronger, highlighting the huge demand for Nvidia in the data center.

The comprehensive operating profit growth rate for the whole year was 522%, of which the processing business increased by 859% and the network business increased by 211%.

The quarterly operating profit growth rate was consistent with the revenue growth rate, indicating that operating profits have stabilized and slightly favored the processing business, as shown below.

Companies and Market Share

While Nvidia is far ahead, market share doesn't matter to the GPU giant, but it does to other competitors. Every percentage point counts.

The total data center processing revenue and market share are as follows:

Nvidia has a large revenue share in the overall market, it has complete control over the margins, and the ability to command a higher premium is an important indicator of the width of Nvidia’s moat. Nvidia’s competitors are trying to push the performance/price metric to convince customers to switch, but at the same time, they are praising Nvidia AI GPUs because Nvidia’s margins are higher.

The changes in market share are shown below:

While it’s a case of “all of a sudden, nothing happened,” the key point is that despite the displeasure of other AI pursuers, Nvidia stood firm and slightly tightened its grip on profits.

The Blackwell delay has yet to impact results, and it's unlikely to hurt Nvidia's bottom line, as the H100 remains the de facto choice for data centers.

Data Center Supply Chain

The semiconductor market’s shift toward AI GPUs has dramatically changed the semiconductor supply chain. AI companies are now transforming into system companies that control other parts of the supply chain, such as memory supply.

Supply conditions were roughly flat from the previous quarter, with strong demand from cloud computing companies and limited supply of CoWoS packaging and HBM memory. The memory situation is improving, although not all suppliers have been approved by Nvidia.

As can be seen, memory companies have been the clear winners in terms of revenue growth since the last cycle low.

There is no doubt that SK Hynix has been a major supplier to Nvidia because Samsung has had problems with verification. Samsung's latest operating profit data suggests that the company is currently supplying HBM to Nvidia or other companies, and the HMB supply situation may be more relaxed.

GPU/CPU Supply

TSMC makes nearly all processors and networking chips. The company recently reported record revenue in the second quarter of 2024, but has not yet reached maximum production capacity. CoWoS is the only area where capacity is still limited, but TSMC is adding a lot of capacity every quarter, which will not affect key players in the data center supply chain.

In addition, monthly revenue in July also hit a new high.

Genju TSMC announced on Friday that its net revenue in July was NT$256.95 billion (US$7.93 billion), setting a record for the month.

TSMC said in a press release that its revenue in July increased 23.6% from the previous month and 44.7% from the same period last year.

The company said that in the first seven months of this year, total revenue reached NT$1.52 trillion, an increase of 30.5% from the same period in 2023.

TSMC forecast third-quarter sales of between $22.4 billion and $23.2 billion at an investor conference in mid-July, a 9.5% increase from the previous quarter at the midpoint, due to strong demand for advanced process technology used in smartphones and artificial intelligence (AI) applications.

Some market analysts pointed out that TSMC's revenue increased in July thanks to increased orders from Apple and US artificial intelligence chip design company Nvidia.

TSMC said that for the full year 2024, its sales could grow 24% to 26% in dollar terms, higher than its earlier forecast of 21% to 26%.

HPC business increased by $3 billion, but no customer or company background was disclosed. Since Apple used to be the only 3nm customer and usually buys less in the second quarter, it looks like it is a new 3nm customer and most likely a data center supplier.

It's probably one of the cloud companies that is trying to leverage its own architecture. Amazon is very active with Trainium, Inferential and Graviton, while Google has TPU.

In addition, Intel's Lunar Lake and AMD's MI series may also be candidates. Due to Nvidia's Blackwell problems, the company will stay on the 4nm (5nm) process until Rubin is ready for release.

Apple is also likely to start using M-series processors in its data centers.

TSMC's revenue growth is undoubtedly good news for the data center market. Regardless of what investment banks think about the return on investment in AI, the data center market will continue to grow in the third quarter.

The Demand Side of the Equation

The AI revolution has triggered an explosion in data center computing. Analyzing Nvidia’s current customer base provides insight into the different demand channels driving growth.

Two-thirds of demand is driven by large tech companies in the cloud and consumer space, while the final third is more fragmented across enterprise, sovereign, and supercomputing. The latter two aren’t really driven from a short-term ROI perspective, nor are they going to suddenly disappear.

Recently, several banks and financial institutions have questioned large technology companies' investments in artificial intelligence, leading to the recent bear market in the stock market.

I’m not one to believe in conspiracy theories, but we all know that volatility is good for banking. I also know that banks are just as clueless about the long-term returns of AI as I am, so I’ll continue to focus on the facts whether the market is up or down.

The main source of funding for the AI boom will continue to be big tech companies

Technology capital expenditure

These five companies account for the majority of capital spending flowing into the data center processing market.

It’s almost as if the financial world views all of the capital expenditures of large cloud customers as brand new investments in questionable AI business models. The reality is that data center investments are not new, and they are increasing AI investments while creating tangible revenue streams.

From a growth perspective, looking at the starting point before the AI boom, it’s clear that data center investment growth actually tracked cloud revenue growth.

While I'll let others decide if this is a good return on investment, the CapEx growth doesn't look crazy compared to cloud revenue growth. That may happen later, but it's certainly defensible now.

The next question is how much processing profit can the large cloud computing companies generate from their capital expenditures?

While capital expenditures have also increased significantly since the start of the AI boom, the share of processing in total capital expenditures has certainly increased. It is worth noting that new AI servers have significantly higher performance than early CPU servers used in traditional data centers.

The increase in capital expenditures in the second quarter bodes well for data center processing companies. This means that the top five companies saw capital expenditures increase by $8.3 billion. In contrast, semiconductor companies saw processing and networking revenue increase by $4.3 billion.

Even better, the capex commitments of the big cloud companies will continue for the foreseeable future. Alphabet, Meta, and Amazon will have higher capex budgets in the second half of the year, with Meta's capex set to increase significantly in 2025.

Microsoft revealed that while almost all of its capital expenditures are related to AI and data centers, about half of its capital expenditures are currently spent on land and buildings. These boxes will later need to be filled with a large number of expensive AI GPU servers and require a firm commitment to long-term capital expenditures.

in conclusion

While current valuations and stock price volatility may seem crazy, the semiconductor industry, while growing rapidly, is not crazy. It is alive and well.

Nvidia may have problems with Blackwell, but can continue to sell H100. AMD and Intel will start to eat into Nvidia, but it hasn't happened yet. Cloud computing companies will also start to quietly adopt their architectures.

While advanced packaging may remain tight, the supply chain looks better positioned to serve new AI-driven businesses with improved memory availability.

TSMC's HPC revenue is growing rapidly, which bodes well for the next earnings season.

Capital expenditures by the large cloud companies have grown in tandem with their cloud revenues, and all have committed to strong capital expenditure budgets over the next 2 to 6 quarters.

In a few weeks, Nvidia will start the data center processing earnings drama again. I'll have my popcorn ready.

In the Meta call, the ROI of AI is divided into two categories: Core AI (related to the ROI view) and Gen AI (as a long-term bet, there is no point in talking about ROI at the moment).

Reference Links

https://semiwiki.com/semiconductor-services/348333-a-post-ai-roi-panic-overview-of-the-data-center-processing-market/

Click here???? to follow us and lock in more original content

END

*Disclaimer: This article is originally written by the author. The content of the article is the author's personal opinion. Semiconductor Industry Observer reprints it only to convey a different point of view. It does not mean that Semiconductor Industry Observer agrees or supports this point of view. If you have any objections, please contact Semiconductor Industry Observer.

Today is the 3855th content shared by "Semiconductor Industry Observer" for you, welcome to follow.