The last time the SOX rose 40% in a month, Bill Clinton was president and the Twin Towers were still standing.
That’s some historical context for what’s shaping up to be a monumental month for semis.
On April 24, the SOX capped an 18-session (!) run of gains with a 4.3%, Intel-inspired barnburner, pushing the gauge’s 14-day RSI to a nosebleed 85.
The figure above gives you some context for the ferocity of the rally. An 85 14-day RSI is rarefied air. There are only a handful of comparable episodes looking back across more than three decades of index data.
Nearly everything’s working in the space. Consensus EPS estimates for Micron, which is up some 50% in April, have almost doubled since late-February as analysts incorporate ravenous demand for high-bandwidth memory.
Texas Instruments, not exactly an exciting name, but a bellwether both for chips and the macro environment more generally, capped a five-week run of gains with its best weekly showing since the dot-com boom. The rally’s courtesy of accelerating industrial demand and 90% growth in its (still small) data center business.
The list goes on. ARM’s up 55% this month (after a 19% rally in March and a 21% gain in February) and AMD an absurd 71%.
YTD, ARM’s more than doubled. The 2026 story for AMD’s in some sense even more remarkable: It was down for the year as of April 6. It’s now up 56%.
And then there’s Intel, which I bought at $20.96 on August 2, 2024 and, unfortunately, gave up on just a month later, selling my entire position at $19.56 for a small loss. Oops. (“Just get me outta this P.O.S.,” he said. “It’s dead money,” he said.)
Intel at a record high is big news. Once up a time, it was the world’s most important chip maker. It was one part of a de facto PC duopoly (“Wintel”) and data centers ran on its chips. That was back when CPUs were synonymous with compute and Theodore Roosevelt was president.
Times change. By and by, Intel fell from grace, becoming a perennial underperformer and, eventually, a semi also-ran. Now, by appearances, it’s back.
The company’s report this week — and particularly a current-quarter sales guide which was $800 million above consensus at the low-end of the range — served notice that what Bloomberg aptly described as a “once-improbable turnaround” is materializing.
Friday’s record-high for the shares was the first in almost 25 years, and it’s good news for The White House. Donald Trump’s statist push has netted the US government a $25 billion paper gain.
So, if you missed the rally — or if you’re like me and you second-guessed yourself for indulging 90s nostalgia in a cheap bet on a left-for-dead relic — don’t fret: As a taxpayer, you own the stock.
Is all (any) of this sustainable? It’s tempting to say “no,” but as a Nomura tech vol trader explained this week, the market’s signaling something important about the evolution of the AI narrative.
“In the last few months, it’s been increasingly obvious that inference demand, driven primarily by agentic-type workflow is going to be what drives the intense demand for compute,” he said. “The corollary is that CPU demand is skyrocketing as it’s not just GPU/TPU/ASICs that are needed, but also CPUs to actually coordinate and orchestrate these tasks.”
As a self-confessed layperson when it comes to the inner-workings of technology (any technology), I’m not in a position to challenge that assessment even if I were inclined, and I’m not.
But (and that probably needs to be in all-caps), the semi rally’s almost surely overcooked in the near-term.
The figures above, from BofA’s Michael Hartnett, underscore the point. On the left, you can see how far afield the SOX is from its 200-day. On the right, you can get a sense of what the rally means for the macro if it’s “the real deal,” so to speak.
“The SOX is the most bubbly since June of 2000,” Hartnett wrote, noting that the upside breakout in so-called “blue-collar semis” like Texas Instruments “implies a surge in ISM manufacturing above 60.” The “up-in-chips,” he went on, is “saying up-in-cyclicals.”
For his part, Nomura’s Charlie McElligott wrote Friday that the nature of the semi move suggests the broader stock rally is at least in part a function of “a true fundamental input.” “This is more than ‘hype’ alone,” he said.
“BUT, the semi rally’s almost surely overcooked in the near-term” – good, because I have a few more tech companies on my “wish list” that I can’t bring myself to pay the entry price, after this week’s performance.
I’m bored with SPY, so reallocating some of that investment (still the vast majority of my investments) into individual tech names. I managed to buy Micron and TSM on Thursday. What could go wrong! 🙂
Turns out H is human. Thanks for sharing your Intel journey.
As you hinted at, the expansion of the rally from GPUs for LLM training to to chips better suited and, perhaps, less expensive chips used in inference should not shock anyone. Nor should Intel’s resurgence – rather than trying to be an also-ran late comer in GPUs, they stood aside and focused on the inevitable uptake in inference demand. At that time, the genius stock analysts decried this strategy. (And cheered Oracle’s latecomer entry into datacenters.)
But the tide lifting TXN, ADI along with stocks like ON and GFS is really not an AI trade. Rather it may simply reflect their inclusion in some indices and ETFs. Not that I am complaining about it.
re Intel: Or as Tom’s Hardware succinctly put it, “CPU requirements for AI workloads are multiplying, driving intensifying shortages and price hikes — Intel already shifting production from consumer chips to Xeon as inference workloads drive server CPU ratios back toward parity with GPUs”
Meanwhile, speaking of left-for-dead relics, nvda looks like it might wanna go blue sky
I bought Amazon at $216 before the war started, and sold it about two-weeks later at $209. It is now at $263. Doh! I hate being wrong about being right.
Being a touch more than a layman, I just had a medium dive into McElligott’s claims (and boy howdy, can that man grasp the core of an industry with frightening clarity!) and it seems that this flight back toward traditional semis (CPUs) is causing rebalancing woes: everyone’s tooled up to produce discrete system components which need to be combined to do AI, and the clearly-superior architecture is unified memory systems which marry CPU, “GPU” and memory on the same board and provide a fat bus to connect them.
Moreso than a sudden surge in demand, it’s this rebalancing that’s causing semis to jump: suddenly, NVDA alone can’t power AI and everyone else needs to pump out sophisticated new hardware designs which are in short supply — even as GPU is hitting a bit of a supply glut, since NVDA or AMD GPUs are falling out of favor as a discrte component.
What does that say about share prices? Hell if I know, other than “the mix will be different tomorrow.” Producers of memory — MU, and so forth — will see consistently high demand, as it’s the common factor limiting both the old architecture and the new architecture. NVDA is adapting to the times with their H200 architecture and beyond, so they will do fine. The zugswang does give other players an opportunity to make some bucks, even as the tensor processing market is still largely controlled by NVDA.