Opinion · August 2026
AI Hardware Bounced Tuesday. Security Software Kept Falling. That’s Not the Same Story.
Monday was an ugly day for both AI hardware and security software: Super Micro fell 5.56%, Nebius fell 3.75%, AMD fell 3.49%, and Marvell fell 3.27%, while Palo Alto Networks fell 1.95%, Datadog fell 4.18%, CrowdStrike fell 0.66%, and ServiceNow fell 0.33%. Eight names, one shared Monday. Tuesday split them cleanly: the four hardware names bounced 4.84% to 9.35%, each on a dated, company-specific catalyst covered in today’s News article. The four software names kept falling, 0.82% to 3.13%, with no dated catalyst behind any of them. That is not one story with two chapters. It is two different mechanisms that happened to collide on the same Monday.
The same eight names, two very different Tuesdays
Figure 1 · Monday’s move (x-axis) vs. Tuesday’s move (y-axis)
A shared Monday selloff split into a hardware bounce and a software continuation
All eight points sit left of center (a Monday decline). The four green hardware points sit in the upper-left quadrant (Monday down, Tuesday up); the four red software points sit in the lower-left quadrant (down both days). Computed directly from Massive Market Data closes, August 21, 24, and 25, 2026.
The distinction matters because a chart like this one can be read two ways, and only one of them is right. The lazy read is “the market rotated out of software and into hardware” — a single narrative, a see-saw, money moving from one side to the other. The read that survives contact with the News article’s sourcing is different: Super Micro bounced because Cisco named it in a partnership announcement; AMD bounced because an analyst doubled a price target; Nebius bounced on continued Nvidia/Goldman Sachs coverage; Marvell bounced because Alphabet expanded a custom-chip relationship with it. Four separate, dated, company-specific events. Meanwhile, I could not find a single dated headline for why Palo Alto Networks, CrowdStrike, Datadog, or ServiceNow kept falling. Their Tuesday looks like a continuation of Monday’s unexplained weakness, not a reaction to anything new.
Why the distinction changes what you should believe
If this were a real rotation — capital leaving software specifically to fund hardware buying — you would expect the size of the outflow to roughly track the size of the inflow, and you would expect a plausible mechanism connecting the two (a specific fund rebalancing, a factor unwind, a sector-rotation call from a large allocator). Nothing in the four hardware catalysts references software valuations, security spending, or capital reallocation at all. Each is a self-contained, idiosyncratic story about the company it happened to. That is much closer to “four unrelated good days for hardware” coinciding with “four unrelated names failing to catch a bid” than it is to a single macro rotation. The shared Monday decline is a real pattern, computed directly from the data. The story that Tuesday reversed it for a coherent reason is not supported by anything dated I could find.
| Cohort | Monday avg. | Tuesday avg. | Two-day cumulative |
|---|---|---|---|
| AI hardware (SMCI, AMD, NBIS, MRVL) | −4.02% | +6.09% | +1.79% |
| Security software (PANW, CRWD, DDOG, NOW) | −1.78% | −1.99% | −3.73% |
Simple averages across the four named tickers in each cohort, not weighted by market cap or by my own position sizes. Source: Massive Market Data grouped daily bars.
Super Micro, the sharpest of the four bounces
Super Micro’s 9.35% Tuesday bounce is the largest of the four hardware moves, and the chart above shows why it reads as company-specific rather than sector-wide: SMCI has traded in a much wider band than SPY over the past two months, including a sharp run-up around late July on unrelated news, well before Tuesday’s Cisco announcement existed. A stock with that much idiosyncratic volatility already built in is exactly the kind of name where a single dated partnership announcement can move the price 9% in a session — it does not need a market-wide rotation as a co-explanation.
A projection: what this divergence compounds into
Rather than guess whether Tuesday’s split continues, here is what happens if each cohort’s average Tuesday move — hardware +6.08%, software −1.99% — simply repeated at that same daily rate for a few more sessions. This is deliberately not a forecast; it is the arithmetic of what “the gap keeps widening at Tuesday’s rate” would actually total, so the size of the current divergence is easier to judge.
| If Tuesday’s daily rate repeated for… | Hardware cohort, cumulative | Software cohort, cumulative | Gap |
|---|---|---|---|
| 1 more session | +6.09% | −1.99% | 8.08 pts |
| 3 more sessions | +19.39% | −5.85% | 25.24 pts |
| 5 more sessions | +34.36% | −9.56% | 43.92 pts |
Assumptions, stated plainly. This table holds Tuesday’s average daily move fixed for each cohort and compounds it forward — it is not a prediction that either cohort repeats Tuesday, and no real stock or sector sustains a single day’s move for five consecutive sessions. Daily catalysts like Tuesday’s (a partnership announcement, an analyst target) are one-time events, not a recurring daily rate, so the hardware cohort’s +34.36% five-session figure is best read as an illustration of how fast an unsustainable rate compounds, not as a target. What this table is for: making Tuesday’s roughly 8-point cohort gap concrete by showing how quickly it would become an obviously absurd number if extrapolated — which is itself the argument for treating Tuesday as four idiosyncratic events rather than the first day of a trend.
The one genuinely forward-looking numbers in today’s tape
Separate from Tuesday’s moves, Motley Fool contributor Harsh Chauhan published analyst estimates Tuesday projecting TSMC’s EPS could reach $47.75 by 2030, implying a $1,150 price target, roughly 182% above TSMC’s current level — contingent on Nvidia’s widely anticipated August 26 earnings report and continued AI-chip price increases. I am not adopting that target as my own view; it is a sourced, dated analyst projection, and its relevance here is narrow: TSMC manufactures for both AMD and Nvidia, so it sits downstream of exactly the kind of company-specific hardware demand that drove Tuesday’s SMCI, AMD, NBIS, and MRVL moves. A genuine, broad-based AI-hardware re-rating would likely show up in TSMC’s numbers too, well before it shows up as a plausible explanation for security software weakness.
Where this connects
This is the same discipline as Wednesday’s observation that a cohort hit by a shared shock didn’t move together on day two, and last Friday’s piece on how an index-level average can hide real single-name dispersion. The common thread across all three: a clean-looking pattern at one level of aggregation (an index return, a sector move, a two-day reversal) can dissolve into several unrelated stories the moment you check each name’s own dated catalyst. Tuesday’s AI-hardware bounce is real and well-sourced. Calling it a “rotation” out of security software is a story I don’t have the evidence for yet.