A Semiconductor Company That Can Grow Even in a Downturn: Why AMD (Part 2)

To understand the problems now facing the semiconductor market, let’s briefly revisit the key points from Part 1.

Tech companies including Meta (META), which laid off more than 10,000 employees at once, are cutting spending across a wide range of areas, and hardware such as data centers and servers inevitably falls into that category. In 2019, the accounting useful life of AWS servers was three years, but by 2022 it had stretched to six. There are thought to be two main reasons for extending the accounting useful life.

First, hardware was not improving that quickly. When building a data center, companies have to consider not only CPUs but also memory chips such as DRAM and NAND, as well as cooling systems, land costs, electricity bills, and many other variables. Because CPU performance was not advancing rapidly, many companies simply did not feel a strong need to replace their existing chips with new ones.

Saying that hardware was not improving quickly is only half true.

More precisely, the pace of improvement for Intel’s (INTC) x86-based CPUs was extremely slow, while AMD’s progress was very rapid. Yet for various reasons, AMD failed to win over data centers as a viable alternative. Until now.

Second, it helped cut costs. Data centers and servers are classic examples of capital expenditures. Cash goes out the door up front, but on the books the cost is expensed over multiple years. If you spread that cost over six years instead of three, the apparent annual expense is cut in half.

So why have AMD’s chips until now failed to gain traction in data centers and servers?

Although both are based on the x86 architecture, there are subtle design-level differences in how AMD and Intel CPUs operate. Ultimately, it may be more accurate to say that because AMD’s market share was low, software developers had little incentive to optimize their products for AMD chips.

Performance and power consumption matter, but for server CPUs, where stability is the top priority, it is easy to see why enterprises were reluctant to switch to AMD too quickly.

Unlike Intel, which has its own fabs, AMD outsources all of its production to TSMC. TSMC’s top-priority customer, however, is Apple (AAPL). AMD is an important client, but it has not been able to secure as much cutting-edge capacity as it would like—and that remains true today. That said, as TSMC ramps up investment in advanced process nodes, the volume available to AMD is steadily increasing.

The issue now is that in the high-end segment—in other words, the performance-driven market—there is no longer any compelling reason to choose Intel’s server CPUs.

According to AMD’s own materials, even if you combine two of Intel’s top-end CPUs, the 3rd Gen Xeon Platinum 8380, they still cannot outperform AMD’s mid-range Genoa EPYC 64-core server processor.

In terms of raw performance, AMD delivers roughly twice the performance per dollar, and in terms of energy efficiency, it delivers the same performance while consuming 25% less power.

A gap of this magnitude effectively means that AMD will monopolize the high-end server CPU market (at least within the x86 ecosystem).

According to data compiled by SemiAnalysis from Mercury Research, AMD’s share of the x86 server CPU market rose from 6.3% in 2020 to 13.9% in the second quarter of 2022—more than doubling in just two years.

What deserves even closer attention is ASP, or average selling price.

Since the first quarter of 2021, AMD’s ASP has been on a steady upward trajectory. Intel’s ASP, by contrast, had been roughly flat before plunging sharply in the second quarter of 2022.

Unable to compete on performance or power efficiency, Intel has chosen to compete on price. I encourage subscribers to think through for themselves what happens to margins at a company that competes primarily on price.

As a result, AMD’s revenue share is nearly double its unit share, reaching 22.9%.

These figures are from before the launch of Genoa, AMD’s 4th-generation server CPU, which means AMD’s ASP is highly likely to continue rising from here.

This is why I believe that, on a revenue basis, AMD can reach a 40% share of the x86 server CPU market within the next two years.