ChipMOS Is Riding a Memory-Pricing Upcycle, Not a Volume Ramp

A Taiwanese chip packager reporting back-to-back months of 30%-plus revenue growth reads, on its face, as a volume story: more chips shipped means more assembly-and-test work billed. ChipMOS Technologies (NASDAQ: IMOS) is growing, but not primarily because more units are moving through its factory. July revenue rose 43.6% year over year to a level the company calls a record since 2014, and August came in around 33% higher than a year earlier. What is actually driving the numbers is a memory-pricing upcycle and the operating leverage it hands to the packager that tests and packages the chips — a distinction that matters to anyone deciding whether the stock's doubling is a cycle to ride or a peak to avoid.

The growth lives in the memory line, not a broad ramp

ChipMOS is an OSAT, a contractor that packages and tests chips designed by others. Its revenue is a function of units handled, the price charged per unit, and how full the factory runs. The company's own attribution separates the contributions. For Q2 2026 it credited stronger pricing, a favorable product mix, and higher utilization — not volume expansion — for its record results. The mix data make the point concrete. Memory was 51% of Q2 revenue, split roughly between DRAM at about 21% and flash at about 30%, and DRAM test-and-assembly revenue was up over 70% year over year while flash rose more than 32%. Display-driver ICs, the other meaningful line at about 18% of revenue, were a smaller, steadier contributor weighted toward automotive panels.

That is the two-market structure doing the work. The DRAM upcycle — tight supply and AI-driven demand pushing memory pricing higher — is what lifts the whole company. An OSAT earns its best economics when the pricing constraint sits upstream in the memory makers, because it can pass along the rising cost of the substrates, lead frames, and gold it consumes while still collecting more per unit. Its own factories are not even full: overall utilization was 72% in Q2, test 74%, assembly 78%. The margin jump came from price and mix before occupancy, which is the signature of a pricing cycle rather than a capacity-squeeze from demand.

Margins confirm the mechanism, and the stock has already paid for it

The profit side shows the same story. Q2 gross margin hit 18.0%, up about 4 points from Q1 and roughly 11 points from a year earlier, and ChipMOS swung to net profit of about NT$892 million from a net loss in the year-ago quarter. That is a genuine, reported improvement. But the market is not paying a discount for it. The shares have risen about 97% this year even after pulling back to the low-$50s from a 52-week high near $78, and they trade at roughly 14x EV/EBITDA — an expensive multiple for a cyclical mid-cap whose cash flow barely covers its own expansion.

The premium is clearest against the larger OSAT peer. ChipMOS trades around 2.5x trailing sales and 14x EV/EBITDA, versus Amkor's roughly 1.7x sales and 9.6x EV/EBITDA. A smaller, more cyclical packager carrying a richer multiple than the sector leader means the market is paying up for the upcycle, not finding a bargain. When ChipMOS reported Q2, the stock fell despite beating profit expectations — a modest revenue miss and explicit plans for elevated capital spending. The multiple is already built on the two things that could disappoint: continued growth and discipline in deploying the cash it is generating.

The capex is the bet, and the condition to watch is memory pricing

Management is not sitting on the windfall. Capital spending is guided to exceed 25% of revenue in 2026 and again in 2027, up from a historical target near 20%, directed at memory assembly-and-test bottlenecks, AI-ASIC testing, and silicon photonics. In dollar terms that is ballpark $150 million of annual capex against roughly $140 million of annual operating cash flow — a plan to spend essentially all of the cash the cycle produces, plus a bit more, on capacity that presupposes the demand persists long enough to fill it.

That is the crux for the retail holder or watcher. The monthly revenue prints are real and the margin recovery is real; the question is whether the market is misattributing a cyclical, pricing-driven bounce to lasting structural demand. The August figure, while up roughly a third year over year, slipped from July's record — normal late-summer seasonality after panel procurement peaked, but a reminder that these are monthly snapshots of a cyclical business. If memory pricing holds through 2027 and the new capacity fills, the premium multiple compresses into earnings. If the pricing upcycle rolls over as it did after 2022, margin — for this company, effectively the memory price minus material costs — gives way on both revenue and profitability, and a stock that has already doubled has the furthest to fall. Memory pricing is the variable, not the next monthly revenue release, which is only a lagging confirmation of it.