Nov 25, 2022
Which semiconductor companies can still grow in a downturn? (Part 1)
Ryunsu Sung
Let’s revisit a passage from “Semiconductors: Is the rebound without impact coming to an end?”:
For reference, Micron has already cut its 2023 CAPEX (capital expenditures) by 50%. The press release implies there will be further CAPEX reductions. Micron now expects next year’s DRAM output to be negative versus this year. This tells us that the current DRAM inventory problem is far more serious than most people think.
As mentioned there, the DRAM inventory issue is severe, and the main reason is the decline in demand for consumer electronics such as PCs and smartphones.
That doesn’t mean enterprise semiconductor demand for servers and cloud is holding up well either.
Take a look at the letter Meta CEO Mark Zuckerberg sent to employees as he carried out a large-scale restructuring at Meta, one of the major data center customers for Nvidia chips.
I’m currently in the middle of a thorough review of our infrastructure spending. As we build our AI infrastructure, we’re focused on becoming even more efficient with our capacity. Our infrastructure will continue to be an important advantage for Meta, and I believe we can achieve this while spending less.
“We are also conducting a thorough review of our infrastructure spending. As we build out our AI infrastructure, we plan to use AI more efficiently relative to scale. Meta’s infrastructure will continue to provide a competitive edge, and I believe we can maintain that edge while spending less.”
Tech companies, including Meta (META), which laid off more than 10,000 employees at once, are trying to cut spending across various areas as part of their restructuring efforts, and hardware such as data centers and servers inevitably falls within that scope.
In 2019, the accounting useful life of AWS servers was three years, but by 2022 it had been extended to six years.
There are thought to be two reasons for extending the accounting useful life.
First, the pace of hardware advancement has not been very fast.
When building a data center, companies consider a wide range of variables, including CPUs, memory semiconductors such as DRAM and NAND, cooling systems, land costs, and electricity bills. Because CPU performance was not improving rapidly, many companies did not feel a strong need to replace their CPUs with newer models.
Second, it helped reduce costs.
Data centers and servers are classic examples of capital expenditures. Although cash goes out upfront for capital expenditures, they are expensed over multiple years in accounting terms. If you spread the cost over six years instead of three, the apparent annual expense is cut in half.
In this environment, the company that deserves attention is AMD.
More details in Part 2…
Newsletter
Be the first to get news about original content, newsletters, and special events.