Has the Memory Supercycle Topped? Our 71 Chip Scorecards Had the Answer Before the Crash

Month 6: a Crack Opened in Memory's "Perfection"

On 6/3, Micron closed at an all-time high of $1,079.57. Over the prior twelve months it had run from $103 to that level, a roughly tenfold gain (stockanalysis).

Two days later, on 6/4, the Nasdaq dropped 4% in a session after Broadcom's results came without an upgrade to its full-year AI guidance, and the selling carried into 6/5. Micron fell as much as 13.3% intraday, the worst performer in the S&P 500, and the memory-and-compute chain shed about a tenth over the two days (FXEmpire). By 6/14 Micron traded near $981.61, down about 9% from the prior week's peak but still sitting comfortably on a roughly tenfold gain for the year.

The spot a single pin can prick is usually the spot that's stretched tightest. This prick put the question the market had been arguing about for half a year squarely on the table: has the AI memory "supercycle" topped?

Bulls say this is just normal consolidation after a powerful run, with the HBM (high-bandwidth memory) shortage nowhere near resolved. Bears say the inventory build has already pulled demand forward and the classic memory-cycle peak is right in front of us. Both sides have data, and neither convinces the other.

We hold something most others don't, and it adds an angle to the fight: for every semiconductor company we cover, we've scored a full "Baillie Gifford growth-investing Ten Questions" card. Two of those questions are built precisely to answer "at this price, is it still worth owning?" Those two scores were set before the crash.

First, Split the Question in Two: the Demand Cycle, and the Stock's Valuation

The reason neither side wins is that the bulls and bears are answering two different questions while using the same word, "top."

The first question is demand. Here the bulls are basically right. Per TrendForce, global DRAM contract prices rose roughly 93%–98% quarter over quarter in the first quarter of 2026, with a forecast of another 58%–63% in the second quarter, and NAND flash contract prices rose 70%–75% over the same span (TrendForce). The mechanism is hard: HBM sits next to the AI chip at far higher profit per unit of capacity, crowding out wafer capacity for commodity memory, while cloud providers build out ordinary servers for AI inference at the same time. Both ends compete for the same supply. In 5/2026, Samsung (5/6), Micron (5/26), and SK Hynix (5/27) each crossed a trillion dollars in market cap within a single month (CNBC). On demand, the story is real.

The second question is valuation. After a tenfold run and trillion-dollar caps across the group, have these stocks already pulled that "real" demand forward? Here the bullish evidence is much thinner, and this is where the real fight is.

Split the two questions and the argument clears up at once: the demand supercycle can be real and these stocks can be expensive at the same time. The tightest bottleneck in the chain isn't necessarily the best link to own. Our earlier AI sector-rotation review named memory as one of the tightest bottlenecks of 2026. What this piece takes apart next is the gap between "tightest bottleneck" and "stock worth buying," the seam nobody wants to look at closely.

Right Before the Crash, Our Scorecards Already Scored Those Two Questions Low

In the Baillie growth Ten Questions, the ninth and tenth form a paired "pricing-perception" set:

Q9: What conditions would all have to hold for it to rise fivefold over ten years? Are those conditions realistic? What does today's share price imply about expectations?
Q10: Why hasn't the market caught on to all of this? Is it that it can't see it, won't bother, or can't see far enough?

These two ask about pricing, not company quality: at the current price, does the market still leave a perception gap? The scale runs to 10, and a higher score means the market understands less and there's more of a bargain to be had; a lower score means the market has already seen through it and the price leaves no room.

Our Micron card was finalized on 5/20, and our SK Hynix card on 5/22, both before the plunge in 6. For each, Q9 and Q10 came in at just 3 each.

The reasoning is blunt. Micron Q9 (source): "Today's share price already embeds the optimistic view that the memory supercycle never fades and that Micron permanently sustains near-peak earnings. There is no margin of safety, and it has worsened further since the report was finalized." Micron Q10: "The market has fully, even over-priced it (sitting at a trillion-dollar cap, P/E around 44), with no upside perception gap... this looks more like seeing through it too well, even overshooting, than failing to see or bother."

SK Hynix Q10 puts it harder still (source): "What the market underprices is precisely its cyclicality and fragility, not its growth... a company nobody 'bothers with' doesn't get bought up to a trillion dollars." It names memory's most classic valuation illusion, the peak-earnings trap: on the surface the forward P/E is only about 6, cheap enough to tempt, but that denominator is cycle-peak earnings. On a free-cash-flow basis, the true P/FCF is around 53 and the free-cash-flow yield is only about 1.9%.

Put differently, before the share price hit its all-time high on 6/3, our late-5 cards had already reached a verdict: the market saw through this story long ago; what's genuinely underpriced is the cyclical fragility of the business.

Not Just Two: Across 71 Scorecards, None Scores 4 on "Pricing"

If it were only Micron and Hynix, this would be a pair of anecdotes. So we widened the lens.

As of 6/15/2026, our Ten-Question scorecards cover 71 semiconductor and AI-hardware reports. Pull every Q9 and Q10 from them, and the result lines up with surprising tidiness:

Q9 averages 2.44 and Q10 averages 2.72;
Companies scoring 4 or above on Q9 or Q10: none. The top score on either question is just 3;
Even NVIDIA, the highest total in the set at 66/100, lands Q9=3, Q10=3.

In other words, across the entire chip and AI-hardware group we cover, not one company is flagged by our scorecard as still carrying a meaningful perception gap at its current price. That's a sector-level signal: full pricing is the state of the whole chain, not the quirk of any one stock.

Memory is especially dense here. Pull the memory and storage names out on their own:

Company	Baillie Total	Q9 Expectations in Price	Q10 Market Perception Gap
SK Hynix	49	3	3
Pure Storage (Everpure)	49	3	3
Micron	45	3	3
Samsung Electronics	44	3	3
Seagate	43	2	3
Western Digital	40	2	2
SanDisk	38	2	3
Kioxia	34	2	2

Totals run from 34 to 49, middling to below; Q9 sits at 2–3 across the board, Q10 at 2–3. That all but counts every memory name an investor could buy, and on these two pricing-perception questions not one reaches 4. The full ranking is on the Baillie growth board.

So What Was That Month-6 Crash, Really

Connect the two findings and the prick in early 6 has an explanation: it was valuation reverting to cyclical fragility, not demand collapsing.

The demand tap is still open. HBM is short, and second-quarter contract prices are still rising. What changed is the price the market is willing to pay for that demand. When a deeply cyclical stock gets priced like a "quasi-platform asset" and its P/E runs to about 46 (Micron's reading in 6, public.com, more than double its ten-year average of roughly 19), any doubt about "how long the boom holds" gets amplified into a double-digit single-day drop. The 6/4 trigger was nothing more than one un-raised Broadcom guidance line, and the memory chain fell a tenth over two days. That is the tape reading of "fully priced."

One distinction deserves honesty here: the cycle top (the peak in price and valuation) usually arrives ahead of the demand top (the peak in shipments and the boom). Granting that HBM demand can stay hot for a couple more quarters and judging that these stocks are already too expensive are not contradictory claims. The genuinely dangerous spot is held by anyone treating "demand is still here" as "buy the dip." By our scorecard, what they're buying is a peak-earnings trap.

What the Bulls Need to Win

Sharp is not the same as reckless. For the bull narrative of "this time is different, the price is fair" to hold, the following all have to happen at once, and we list them here as the falsification conditions for our own call:

Supply discipline holds. Samsung, Hynix, and Micron all stay restrained on capacity through this cycle, refusing to replay the "everyone expands, prices collapse" script that ends every memory cycle. This runs against industry history and is the hardest condition.
HBM "decyclicalizes" memory. HBM4's technical bar is high enough to turn advanced memory into a structurally short, scarce good, so gross margins no longer swing wildly with commodity DRAM. This is the bulls' strongest argument, but for now it's an assumption, not a delivered fact.
Earnings catch up to valuation. Real free cash flow over the next year or two has to work today's P/FCF of around 53 back into a reasonable range, rather than propping up the forward P/E on cycle-peak accounting earnings.

If all three hold at once, today's price has an explanation. Loosen any one of them, especially if new capacity comes online or the quarter-on-quarter rise in contract prices narrows to single digits, and the scale tips fast toward valuation reverting.

Conclusion: the Supercycle Is Still On, the Super-Stocks May Be Over

Laying the call out:

This is a valuation top, not a demand top. We put the subjective odds that HBM/AI memory demand still grows year over year over the next 12 months at roughly 75%. On demand, the bulls are right.
But the stocks are priced to perfection. 71 scorecards, with no Q9 or Q10 above 3, is the hardest cross-sectional evidence we can offer. We put the subjective odds of Micron drawing down at least 25% from its 6/3 high ($1,079.57) within 12 months at roughly 60%.
The most dangerous trade is buying the dip. "It fell, so it's an entry point" carries a large question mark for a stock at a P/E of 46 that our own scorecard flags as a peak-earnings trap.

To track this call, three public signals are enough: whether the quarter-on-quarter rise in DRAM contract prices narrows to single digits (the leading indicator of a cooling memory cycle), whether the three majors start announcing new capacity (whether supply discipline breaks), and whether, at the next scorecard re-rating, the first Q9 or Q10 at 4 or above finally appears among these 71 names (whether the perception gap reopens as prices pull back; we estimate the odds that none reaches 4 within one re-rating cycle at roughly 80%).

We'll keep updating these three signals.

The Evidence Boundary

The load-bearing numbers here fall into three classes with different bases, set out separately:

Market figures (Micron's price, all-time high, market cap, P/E, Hynix's level) are corroborated across several public market sources and are point-in-time snapshots with a shelf life measured in weeks; the 6/4–5 single-day drop and its trigger are taken from financial-media reporting.
Scorecard data (Baillie totals, Q9/Q10 scores and reasoning) come from our own scoring system, as of 6/15/2026. The "71 companies" is the count of semiconductor and AI-hardware reports under Ten-Question coverage at that time, from the same source as the Baillie growth board; the sample changes as reports are added, and scores may adjust at re-rating.
Probabilities and judgments (drawdown odds, demand-growth odds, perception-gap odds) are our own subjective view. This article does not constitute investment advice.

Memory is a textbook deeply cyclical business, and this piece's core claim, "demand is still here, the stocks are already expensive," would be falsified if the boom runs longer than expected or if supply discipline unexpectedly holds tight. The signals to watch for that are listed above.