Chip Armageddon: Falling Knife or Buy of the Decade?

What is happening

Chip shares have had a bruising ten days. Across the 193 companies we rate in the electronic components and manufacturing group, the typical company's shares have fallen 13.6% since 6 October, and 10.7% of that came in the past week alone. Some fell much further. In the week, Applied Optoelectronics lost 31%, Ichor 30%, LG Innotek 29%, Semtech 27% and Yangtze Optical Fibre and Cable 26%. These are not household names, but they make parts that go into the same data centres and devices as the famous chips, and they have been swept along with them. The Philadelphia Semiconductor Index, the main US index of chip shares, closed lower for an eighth session in a row on 16 October, its longest losing run since October 2008, leaving it 15.7% below its 6 October close.¹

The eight companies at the centre of it have not fallen together. Since 6 October, SK hynix is down 28.3%, Lumentum 27.3%, Lam Research 21.0%, AMD 18.6%, Micron 15.9%, Nvidia 10.5%, ASML 9.7% and TSMC 6.4%. The spread matters. The two memory-chip makers, SK hynix and Micron, are among the hardest hit, along with the companies that sell equipment and parts into the build-out of AI data centres. The two that have held up best, TSMC and ASML, are the two that reported results while chip shares were falling, and both reported records.

Three worries lie behind the selling: dearer money for the AI build-out, a coming glut in memory chips, and the risk that a fall in AI-linked shares spreads. The question is whether share prices already allow for that trouble, or whether the trouble is only beginning. We set the two sides against each other, each at its strongest, and against what today's prices already assume about the future, which settles part of the argument on its own.

The fear

The fear starts with money. A hotter-than-expected US inflation report revived worries that higher interest rates will make AI investment more expensive to fund, and the Nasdaq and the S&P 500 had their worst day since July.² The backdrop gives that worry weight. The US Federal Reserve raised its policy rate to 4.00% on 17 September, reversing the direction of its previous move, a cut in December. The yield on the US 10-year government bond, which sets the tone for long-term borrowing across the American economy, stood at 5.10% on 15 October, up 1.05 percentage points in a year. The data centres that buy AI chips are enormously expensive to build, and much of that building is paid for with borrowed dollars. Investors are questioning whether the spending on AI can last and how it is being paid for, and borrowing costs have jumped for some heavily indebted operators of AI data centres.³ If the customers find money harder to come by, the fear runs, the chip makers' order books will be the first place it shows.

The second strand is memory. Memory chips are the parts that hold data while a computer works on it, and the industry has a long history of piling into new factories when prices are high, then drowning in its own supply. China's ChangXin Memory Technologies, known as CXMT, said it had made significant progress on a denser memory-chip design aimed at server memory, the most profitable part of the market.⁴ Some investors fear the world could have too much memory-chip capacity as early as 2029.⁵ The memory makers have taken the heaviest blows, as set out above. And the picture from analysts on SK hynix is less one-sided than elsewhere: in the 30 days after its last results, on 29 July, their forecasts were raised at least four times and cut three times.

The third strand is contagion. The International Monetary Fund warned in its financial stability report that the consequences of an AI correction could be global.⁶ That warning matters to the fear because of who is making it. The IMF watches the financial system as a whole, and its concern is not confined to chip shares. If AI spending were to falter, the losses would not stay with the chip makers. They would fall on the investors who hold AI-linked shares and on the lenders who have financed the data centres, many of them far from the chip industry. On this view, the fall in chip shares is less a contained problem than an early sign of a strain that could travel. The fear's sharpest point is that some prices still assume years of very rapid growth, even after the falls. AMD is the clearest example: its price still asks for faster growth than any other in the group, as the arithmetic below shows, and a price like that has a long way to fall if the growth stalls. The fear is that the fall so far is the market only starting to reckon with that.

The case that it is overdone

The case that prices have already absorbed much of the trouble begins with the companies closest to the orders. TSMC, the largest contract chip maker in the world, builds the chips that Nvidia, AMD and many others design. In its results on 15 October, it reported a record third-quarter profit and said it expects fourth-quarter revenue of $52.3 billion to $54.3 billion.⁷ Its chief executive said demand is still far above what it can supply.⁷ Its sales are up 36% on a year earlier. Analysts responded within a day: their forecasts were raised at least 18 times and cut once.

The second plank is that orders in this industry are booked years ahead. ASML makes the machines that chip factories use to print the finest circuits, and a chip maker has to commit to a machine long before it arrives. ASML's results on 14 October reported record quarterly sales and said the machines it plans to build in 2027 and 2028 will not fully meet its customers' demand.⁸ In the two days after, analysts' forecasts were raised at least 28 times and cut twice. The pattern repeats elsewhere. After Nvidia's latest results, on 26 August, forecasts were raised at least 31 times and cut none in the month that followed. After Micron's, on 30 September, they were raised at least 32 times and cut twice. On this reading, the people paid to model these businesses have been revising their numbers up, not down, every time they have been given fresh results.

The third plank is that, for the hardest-hit companies, the prices already assume trouble. The memory makers' share prices are so low against their profits that they would still look ordinary even if those profits shrank every year for five years, as the arithmetic below shows. In other words, a memory glut is not a risk these prices ignore; it is closer to the starting assumption. On the China threat, Goldman Sachs told clients that the market has overestimated how quickly CXMT can close the technology gap in advanced memory.⁹ And neither the companies' results nor interest rates explain the fall on their own, since ASML and TSMC reported strong numbers in the middle of it.¹⁰ The suggestion is that the selling has been driven at least partly by mood rather than by anything the companies have reported.

Where the two meet

Each side has a reply to the other's strongest point. The fear's answer to the record results described above is that this is exactly what the top of a boom looks like. Micron's sales are up 256% on a year earlier and SK hynix's up 145%. Numbers like that are what persuade companies, including new Chinese rivals, to build new factories, and new factories take years to come on stream. So the fear does not dispute today's order book. Its worry is the capacity the world will have by the end of the decade, and whether the customers can keep paying at the borrowing costs set out above. A full order book today, on this view, says little about the state of the market three years out.

The case against the fear answers the dearer-money point in turn, and its reply is about timing. TSMC and ASML are the companies that actually see the orders, and they spoke while chip shares were falling, after the inflation report and the rise in bond yields had already happened. If higher borrowing costs were already choking off orders, these are the companies that would be expected to see it first, and neither did. As for the memory makers the fear has hit hardest, their prices already allow for a glut, so it would have to be deeper than the one already assumed to undermine them.

What neither side can do is prove its view of the years ahead. The fear is about the future of supply and financing; the case against it rests on the present evidence of demand. They are, in a sense, talking about different years.

What the prices assume

A useful way to see what a share price takes for granted is to ask how fast a company's profits would have to grow for its price to look ordinary. The yardstick here is the price-to-earnings ratio: the share price divided by a year's profit per share. The typical listed company costs about 19 times its earnings. A company costing 40 times is priced as though its profits will grow well beyond today's; one costing 6 times is priced as though they will not. So for each company we ask: if the share price stayed exactly where it is, how much would earnings have to grow each year for five years to bring the ratio down to 19? That is not a forecast. It is simply what the price already assumes. Beside it we set how fast its earnings have grown a year over the past three years: its own past record, and not a prediction either.

Read this way, the eight companies split into three groups. Nvidia, at 27 times earnings, needs its earnings to grow 8% a year, against a recent pace of 102% and sales up 83%. TSMC, at 28 times, needs 9% a year against a recent pace of 52%. That low bar is not fear in the price. Nvidia and TSMC have fallen the least of the eight, and their prices ask little only because their earnings grew so fast, from a low base, over three years of boom. That pace is their past, not a forecast. For these two the question is not how much fear the price allows for, but whether earnings that grew this fast can hold.

ASML and Lam Research ask for more. ASML, at 42 times, needs 18% a year against a recent pace of 23%. Lam Research, at 46 times, needs 20% against 29%. Lumentum, at 38 times, needs 15% a year measured on forecast earnings, with sales up 83%. Here the price asks for something closer to the recent pace, which leaves less room for a slowdown. AMD stands apart: at 136 times its earnings, it needs 49% a year for five years. Its recent pace has been 213%, so it has grown faster than that before, but the price leaves the least margin of any in the group.

The memory makers are the mirror image. Micron's price would still reach an ordinary level if its profits shrank 9% a year for five years; SK hynix's if they shrank 22% a year. What this arithmetic cannot say is whether last year's earnings are a fair place to start. If memory profits are at a peak, even a price that assumes shrinking profits may not have allowed enough. If the build-out runs on for years, it may have allowed far too much.

The companies

Nvidia designs the processors that power most AI data centres, and it has become the emblem of the whole build-out. That makes it the company most directly in the path of the dearer-money fear, since its customers are the data-centre builders whose borrowing costs are in question. Its latest results, on 26 August, came well before the fall began, so its own account of demand predates the inflation report and the rise in bond yields.

Taiwan Semiconductor Manufacturing, or TSMC, is the world's largest contract chip maker, building the chips that Nvidia, AMD and many others design. Its results are the main evidence for the case that the fear is overdone, and it has held up best of the eight. It is still exposed to the same customers as everyone else: if the data-centre builders pull back, its order book is where that would land.

Micron Technology, based in the United States, is one of the world's big memory-chip makers, and it sits squarely in the memory-glut argument. CXMT's claimed progress is aimed at server memory, the most profitable part of Micron's market. Its latest results, on 30 September, came just days before the fall began.

SK hynix, the South Korean memory maker, is a major supplier of the specialised memory used alongside AI processors. That puts it at the meeting point of the fear's first two strands: it depends on the AI build-out continuing, and it is exposed to any glut in memory and to any challenge from CXMT. Its latest results, on 29 July, are the oldest in the group, so the most recent evidence on its business comes from others.

Advanced Micro Devices, or AMD, designs processors for computers and data centres and is Nvidia's main rival in AI chips. It is where the fear's point about shares priced for fast growth bites hardest. It last reported on 4 August, long before the fall began.

ASML, the Dutch maker of the machines that print the finest circuits onto chips, sees orders years before they become sales, because a chip maker has to commit to a machine long before it arrives. That gives its results particular weight in the argument, since it is speaking about demand well beyond this year.

Lam Research makes equipment that chip factories use to etch and lay down the layers of a chip. Like ASML, it sells to the factories rather than to the data centres, so its orders show whether chip makers are still expanding. It last reported on 29 July, before the fall began, and is the first of the group to report again, as set out below.

Lumentum makes the optical parts that use light to link up the machines inside AI data centres. It belongs to the same corner of the market as Applied Optoelectronics, the week's largest faller, and it has been among the hardest hit of the eight. Its business depends directly on the pace of data-centre building, so it is exposed to the dearer-money fear more than to the memory one. It last reported on 11 August.

What would settle it

Several signals move well before any company reports, and each would tell the two sides apart. The first is the price of memory chips. Memory makers and their large customers agree contract prices months ahead, and those prices respond to the balance of supply and demand long before it shows up in a quarter's profits. If the glut the fear expects is on its way, contract prices for server memory would turn lower, even while today's order books look full. If AI demand is still outrunning supply, they would hold firm or rise, and the memory makers' low share prices would look like trouble assumed too early.

The second is what the customers plan to spend, and what it costs them to borrow the money. The companies building AI data centres set out their spending plans for the year ahead, and the most indebted operators have to keep raising fresh money.³ Plans held or raised, with lenders still willing at a bearable price, would mean dearer money has not yet bitten. Plans trimmed, or borrowing costs for those operators climbing further even while government bond yields hold steady, would be the fear arriving where it says it will, before any chip maker's orders reflect it. Closely tied to this are TSMC's sales, which it publishes every month between its quarterly results, and the waiting times for machines from ASML and Lam Research. A chip maker that defers or cancels a machine is acting on its own view of demand years out, so shrinking waits or cancellations would side with the fear, and lengthening waits with the case against it.

The third is CXMT's actual shipments, and to whom. Its claim of progress concerns a design.⁴ Goldman Sachs's view is that the market has overestimated how quickly it can close the gap.⁹ What separates the two is whether server customers begin testing and then buying its memory in volume, and whether governments tighten or loosen their rules on selling chips and chip-making equipment to China.

The companies' next results are the next check on whether these trends hold: Lam Research on Wednesday 21 October, AMD on Tuesday 3 November and Lumentum on Thursday 5 November, each after the close of trading.

Where they stand with us

We award medals to the companies we rate, and before the fall, at the end of September, most of this group stood well with us. TSMC, Micron, AMD, ASML and Lumentum held gold medals, Nvidia and Lam Research silver, and SK hynix bronze. Across the wider group of 193 electronic components and manufacturing companies there are 34 golds. A medal describes where a company stands with us; it is no verdict on whether the fear or the case against it is right, and it says nothing about where any share goes next.

The judgement that is left

The numbers do settle one thing: where the fear already sits in the price. For SK hynix and Micron, a memory glut is close to the starting assumption. Nvidia and TSMC are a different case: their prices ask little because their earnings have grown so fast, not because fear has marked them down, and they have fallen the least; for them the question is whether those earnings hold. For AMD most of all, and for Lam Research and ASML to a lesser degree, the prices ask for growth close to their recent pace, leaving less room for a slowdown.

What the numbers cannot settle is whether recent earnings are a fair starting point, and that turns on a single judgement: whether the spending on AI data centres keeps going for years, or slows as the money to pay for it gets dearer. If it keeps going, today's earnings are a reasonable base, and the falls of the past ten days look large against what the prices assume. If it slows, the memory makers' low prices may turn out to be justified, and the companies priced for fast growth are the ones most exposed. Whether the build-out outlasts the rise in the cost of paying for it is the question on which every price in this group now rests.

Notes

1. TECHi, 16 October 2026. techi.com ↩
2. CNN, 15 October 2026. cnn.com ↩
3. Seeking Alpha and CNBC reports, 15 October 2026. seekingalpha.com ↩
4. CNBC and Investing.com, 13 October 2026. cnbc.com ↩
5. CNBC, 15 October 2026. cnbc.com ↩
6. The IMF, as reported by CNN, 13 October 2026. cnn.com ↩
7. TSMC's results, as reported by Yahoo Finance and The Motley Fool, 15 October 2026. finance.yahoo.com ↩
8. ASML's results release, 14 October 2026. sec.gov ↩
9. Goldman Sachs, as reported by 24/7 Wall St., 16 October 2026. 247wallst.com ↩
10. TECHi, 16 October 2026. techi.com ↩