Semiconductor world in for a rough ride as chip bubble bursts at the high end
Just don't expect foundry spending to stop anytime soon
Analysis The semiconductor gold rush is all but over at the high end, and we've had our fill. Or so the past month of dismal earnings might have you believe.
Electronics giant Samsung saw its profits contract 69 percent during the fourth quarter, while revenues slumped eight percent overall. South Korean memory manufacturer SK Hynix, meanwhile, followed a few days later with an equally bleak report. Both companies told a story of macroeconomic forces that were suppressing consumer spending and driving DRAM and NAND flash inventories to unprecedented levels.
Put simply, where there was once a chip shortage there is now a glut. Well, of memory, anyway — more on that later.
Intel, AMD, and Qualcomm, whose top-end chips depend on DRAM and NAND flash and are thus inexorably intertwined, saw declines across key markets including PCs, smartphones, servers, and game consoles. If customers aren't buying memory, it makes sense they wouldn't be buying PCs and servers to put it in.
While the rapid deterioration of the semiconductor market may have come as a surprise to some, the writing has been on the wall for months.
And bear in mind, we're talking about memory and high-end components here in this piece – there are still ongoing supply issues and delays with microcontrollers, glue electronics, and similar low-level components.
Micron was among the first semis to succumb to market forces. After riding strong demand for months and promising tens of billions in new fabs, the company decimated its workforce, laying off 4,800 after its Q1, 2023 earnings tumbled 88 percent from the year prior on a nearly $200 million loss.
But even this wasn't the first sign that the pandemic bubble had burst. More than six months prior, the industry watchers at TrendForce offered a stark outlook for the memory market, warning of growing inventories of yet unsold product. The memory vendors were headed for economic turmoil long before this week.
How did this happen?
The semiconductor industry has, for the past three years, been caught in a perfect storm that has fueled steady demand.
In the wake of the COVID-19 outbreak, an entire remote-working economy was born over the course of a few weeks. Every tech and software company worth its salt rushed to capitalize on the shift out of the office.
Security vendors rushed products tailored to home offices out the door; notebook vendors crammed higher resolution cameras and microphones into their wares; and software vendors like Teams and Zoom scrambled to keep their services online. Whether directly or indirectly, all of these dynamics fueled semiconductor demand in one capacity or another.
Before long, existing inventories emptied, and with factories closed due to COVID-19 lockdowns, it seemed everything was in short supply. Within a year, the semiconductor supply chain was stretched beyond its limits, and we were in the full grip of the chip shortage.
To make matters worse — for consumers anyway — the then-current generation of graphics cards proved particularly efficient at mining cryptocurrencies. And while Nvidia and AMD pretended to care about how difficult it was to buy a GPU, they were only too happy to collect the massive profits the cards brought in every quarter.
In mid 2022, the bubble burst. It was already getting easier to find components — lead times for basic, but essential, parts were down below 26 weeks and headed in the right direction — just as the Ethereum merge came. Within a few weeks, cryptocurrency valuations collapsed, taking GPU prices down with it.
- China doesn't need to take Taiwan's fabs to escape US trade bans
- As wafer demand dries up, foundry revenues head for a cliff, we all celebrate
- Intel inside a world of pain as revenue plunges by a third
- The great semiconductor drought may be about to break
This would have been good news for consumers, but the economy had already started to cool. Interest rates were rising; the cost of living was going up; in Europe, the war in Ukraine had driven fuel and energy prices to all-time highs. Suddenly, folks weren't so keen on buying another new PC or smartphone.
Over the course of a few months, demand for "high-margin" chips — think CPUs and GPUs — evaporated. To be clear, the chip shortage isn't over. It's just that demand — particularly for processors used in consumer hardware — is gone. Supply of many components, especially those used for power delivery and automotive environments, remain heavily constrained.
The worst has yet to come
Memory fabs' options, given the current predicament, are limited. These companies have fairly long and complex supply chains and largely compete on pricing and process tech, CCS Insights analyst Wayne Lam previously told The Register.
Until things improve, Samsung and SK Hynix are focusing on products like LPDDR5, DDR5, and HBM used in mobile, next-gen server, and high-performance computing products, which it anticipates will drive future demand.
By comparison, conventional fabs, like those operated by TSMC, and those not dedicated to memory at Samsung, have relatively short, straightforward supply chains. This, Lam explained, gives them greater flexibility to reallocate capacity to chips that are in higher demand.
Despite all of this, the next few quarters are looking pretty grim. Over the past few weeks, chipmakers and foundry operators have offered a near universally grisly outlook on the quarters to come.
Intel expects its revenues to fall to $11 billion and potentially lower in the first quarter of 2023. Meanwhile, AMD is predicting flat revenues for Q1 and poor PC and gaming sales throughout the 2023 fiscal year. And, this week, Qualcomm again blamed the fact nobody is buying phones for its shoddy performance. Even TSMC forecast a revenue drop, its first in four years.
The general consensus among these companies seems to be that things should start picking up again in the second half of 2023. But not everyone is so sure. In a recent report, Dylan Patel at SemiAnalysis predicted that inventory levels wouldn't return to normal in Q2, and that the bear market for semiconductor companies would stretch on longer than anyone anticipated.
"We believe there is around a 15 percent downside for semiconductor companies' valuations or significant sideways movement before the real bull run can start," he wrote. "Days of inventory are currently at all-time highs. Even higher than the dot-com bubble and the 2008 financial crisis. This inventory will take a lot longer than two quarters to digest."
Patel isn't the only one. In a recent TrendForce report, analysts predicted foundry revenues would contract 4 percent in 2023, as wafer demand continues to dry up and inventory consumption slows.
When the semiconductor market does recover, the next question will be how quickly. Short of another civilization-altering event, it seems we're unlikely to see the kind of pandemic-level investment in IT that fueled so much of the semiconductor demand over the past three years.
Why all the fabs then?
Given the perilous state the semiconductor space appears to be in, it may surprise many to see chipmakers like Samsung, TSMC, and Intel pushing ahead on fab projects.
These facilities aren't cheap by any stretch of the imagination — prices are often measured in the tens of billions of dollars just to build even one fab. So, you'd think these would have been the first on the chopping block. If nobody is buying chips, why are you building more fabs? But that's exactly what they're doing.
Intel may be trying to get more cash out of the German government, but they are — for the moment anyway — still committed to building a new fab there. And despite the poor quarter, Samsung said it's pushing ahead with its foundry expansions, including its new operation in Texas. SK Hynix has committed to building a US packaging facility. TSMC has scaled back its capex plans for the 2023 fiscal year, but it still plans to spend between $32 billion and $36 billion on new infrastructure. If you're wondering, that's between $4 billion and $8 billion less than in 2022.
While it might sound like poor business sense, it's important to remember just how large and complex these facilities are. They're not just buildings. They're filled with sophisticated, highly-regulated equipment that takes time to acquire, install, and validate. Most of the 20-or-so fabs announced to date won't come online before 2025.
It certainly doesn't hurt that the US and Europe are funneling tens of billions into domestic chip production in a bid to reduce their reliance on the Asia Pacific fabs before tensions with China boil over. ®