DELL's lower-than-expected quarterly report is just to grab orders, actual orders will far exceed expectations | AI Dehydration

Author: Zhang Yifan

Editor: Shen Siqi

Source: Hard AI

A few weeks ago, there were rumors in the market that Musk's xAI would build a supercomputer.

This model data center, serving Tesla's autopilot and the next generation Grok, is another giant computing factory after the Meta supercomputing center. The supplier of xAI has always been highly anticipated in the market.

Last night, Musk publicly revealed on his Twitter the rack suppliers for xAI - Dell (DELL) and Super Micro Computer Inc. (SMCI), with each holding half of the shares.

It is worth noting that this data center has purchased 100,000 H100 computing cards, making it at least four times the size of the current largest AI cluster.

Subsequently, Dell rose for three consecutive days, with an increase of over 13%. Super Micro also rose by 4.5% in two days.

I. Why did Dell rise?

Why did Dell's increase exceed Super Micro's by nearly twice?

Mainly because of the cooperation with xAI, breaking the market's previous low expectations for Dell.

Previously, Hard AI discussed in "AI Server Shipment Surge, Why Dell's Profit Unexpectedly 'Slipped'" that Dell's latest quarter financial report showed that its storage business and operating profit were below market expectations.

Due to Dell's low profit margin in the AI server field, it raised concerns in the market, weakening the market's profit expectations for Dell's future AI servers.

However, the cooperation with xAI gave the market a shot in the arm, once again raising expectations for Dell.

Three days ago, Morgan Stanley released a report on Dell, stating that based on feedback from the supply chain, Dell's orders are stronger than initially expected, especially receiving a large number of orders from secondary cloud service providers.

Morgan Stanley further stated that the assembly of AI servers is not simple, with high technological barriers, and there are not many manufacturers that can mass-produce on the production line.

At the same time, Dell has always been in a leading position in end-to-end computing solutions for enterprises. NVIDIA CEO Jensen Huang has also emphasized that Dell is the preferred company for any enterprise computing needs.

In addition, in the face of Dell's low profit margin in AI servers, industry experts speculate that Dell is currently using a strategy of seizing market share at low rates, and in the long run, the AI business is still promising.

Morgan Stanley pointed out that the current market demand for storage is also rebounding, which will contribute to Dell's overall positive outlook.

II. TSMC's Price Increase May Bring 3% to 6% Revenue Upside

1. TSMC's Price Increase

In early June, rumors of TSMC's 3nm process price increase surfaced at COMPUTEX 2024 in Taipei.

There are three main reasons for the price increase.

1) Chip prices are high, while TSMC's supply costs are low

The H100 is priced at $25,000 to $30,000, while TSMC's supply cost is only $666.

Specifically, TSMC's 3nm process wafer price is $20,000, and one wafer can yield 30 H100 chips, equivalent to each H100 chip selling for only $666 After NVIDIA received H100, the selling price reached $25,000 to $30,000.

Following the rumor of TSMC's price increase, NVDA CEO Huang Renxun expressed his agreement, stating that "TSMC deserves a higher price."

Citi commented that TSMC should be able to reflect its value to customers in the context of the surge in demand for AIGPU/accelerators and edge AI devices.

2) Increase in material, R&D costs, and production factors such as electricity and water costs

The rise in material, R&D costs, and production factors such as electricity and water costs prompted TSMC to raise prices.

TSMC's management pointed out that the company is facing cost increases from some overseas wafer fabs, inflation, and higher electricity costs.

They also mentioned that they will charge higher prices to overseas wafer fabs compared to Taiwanese ones.

Previously, many reports indicated that TSMC's factory costs in the United States are much higher than those in Taiwan, with one of the main reasons being that labor costs in the United States are much higher than in Taiwan.

On one hand, facing the increase in material, R&D costs, and production factors such as electricity and water costs; on the other hand, TSMC needs to maintain the company's profitability and growth.

Therefore, the price increase is justifiable.

3) Demand exceeds supply

The demand for the N3 process is far greater than the supply. From an economic perspective, when demand exceeds supply, price increases naturally follow, and the highest bidder wins.

Citi pointed out that apart from Apple, the second half of 2024 will see a large number of N3 orders from Intel, Qualcomm, and MediaTek, and more N3 AI GPU/accelerator orders are expected in 2025.

Currently, TSMC and Samsung are the only foundries globally capable of supplying 3nm. However, Samsung's 3nm production is lagging, so Samsung's high-end SoCs can only be produced by TSMC.

TSMC will truly be the exclusive supplier in 2024-2025, and the price increase is a natural progression.

2. Impact of price increase on revenue

On June 17th, according to Commercial Times, TSMC may raise the wafer prices for N5 and N3.

As of 2024 Q1, N5 and N3 processes accounted for 37% and 9% of TSMC's revenue, respectively.

Currently, due to the limited capacity of TSMC's N3 process, most AIGPU/accelerators are on N7/5/4.

Citi predicts that with the ramp-up of TSMC's N3 capacity and the improvement in capacity utilization, the revenue share of N3 will increase.

Furthermore, with the emergence of AI PCs/AI phones, advanced processes (N5/4/3) will continue to be in high demand Citi predicts that the advanced nodes (N5/4/3) will account for 53%/61% of TSMC's revenue in 2024E/25E, and expects the average selling prices of N5 and N3 processes to potentially increase by 5-10%, providing TSMC with approximately 3% to 6% revenue upside