Lou Miller Says Neoclouds Are Down 57% and Optical Stocks 46%. Past Manias That Fell That Far Took 13 to 18 Years to Get Back, if They Did

Lou Miller runs the Goldman Sachs desk that builds custom baskets of stocks for clients, so he watches the AI trade in pieces. Asked on the bank’s podcast how the data center trade was holding up after the midterm elections, he said, “Worse than I would have thought.” He put data center stocks down 36% from their highs earlier in the year, then listed the higher-risk corners.

Miller said the early reaction was more extreme than he expected, and he found a silver lining in it. “The multiple for the data center trade is about half of what it was in June,” he said. “We’re at liberation day lows in terms of multiples.” In his view, valuation is no longer a headwind and the bar for positive surprises is much lower. He recorded the episode on Thursday, Nov. 5, the day the S&P 500 fell 4.4%.

Checking the Drops

Goldman’s baskets are its own, so we built simple versions: equal amounts of each stock bought at the end of 2025 and held. At the Nov. 4 close, the last before Miller spoke, a neocloud basket of CoreWeave CRWV, Nebius NBIS and IREN IREN was 56% below its peak, close to his 57%. Our optical basket of Lumentum LITE, Coherent COHR, Ciena CIEN, Applied Optoelectronics AAOI and Fabrinet FN was down 59%, and a memory basket of Micron MU, Sandisk SNDK, Western Digital WDC and Seagate STX was down 50%. By our measure, his optical and memory figures were the milder ones.

The selling got worse on the day he spoke. The neocloud basket hit its low on Nov. 5, 62% below its June 18 peak, and then bounced. As of the Nov. 9 close it was 56% below the peak, and the optical basket was 59% below its own. For the neoclouds, the slide has also reached the bond market: CoreWeave’s bonds yield about 15%.

The drop and the record, as of the Nov. 9, 2026 close

Every Past Mania We Could Chart

To see what usually follows a fall like this, we looked at 14 past stock manias we could chart from their peak with total returns, using an exchange-traded fund where one existed at the time and an equal-weight basket of the leaders where it didn’t. The list runs from the 2000 tech peak through housing, clean energy, gold miners, 3D printing, biotech, cannabis and the 2021 crop: clean energy again, Chinese internet stocks, cloud software, IPOs and lithium. Thirteen of the 14 fell 57% or more at some point, and the median group bottomed 79% below its peak.

Of those 13, 5 have gotten back to their old high, and none did it in under 13 years. Clean energy took 13.1 years from its December 2007 peak, gold miners 14.0, homebuilders 14.2, the Nasdaq-100 14.6 and chip stocks 17.6. The one that fell less, biotech in 2015, dropped 44% and was back in 2.2 years.

Every time before

Of 13 past stock manias that fell as far as the neoclouds, 5 got back to their peak. It took 13 to 18 years.

Each line is a past mania’s total return from its peak month, drawn until it got back to that peak or, if it never did, to Nov. 9, 2026. The three from 2000 are dark. The 2026 neocloud basket (CoreWeave, Nebius, IREN) is coral.

The closest match to Miller’s optical group is the optical boom of 2000. An equal-weight basket of JDS Uniphase (now Viavi), Ciena and Corning GLW lost 98% by 2002. It came within 2% of its August 2000 peak at the end of June 2026, during this year’s AI optics rally, and then fell back. At the Nov. 9 close it was 54% below that 26-year-old high. The telecom companies that bought the gear in 2000 spent $2.77 for every $1 of new revenue, and the suppliers’ stocks have yet to fully recover. Single names tell the same story: Cisco’s 2000 crash took 25 years to recover.

What the Record Can and Can’t Tell You

This is a base rate, a description of what happened to groups that fell this far before, and it carries three limits. First, the groups are ones the market came to call manias, which tilts the list toward the ones that ended badly. Second, the five from 2021 have had only about five years; all five are still below their peaks, from 30% to 68%. Third, the neoclouds are a few months into their slide. In past manias a 57% drop was often a stop on the way down: the median group kept falling to 79%.

Miller’s argument is about price, and the record above is about time. He says the reset has made the data center trade cheap enough that good news can move it. The history says that when a hot group falls this far, getting all the way back to the old high has been rare and slow, even for the groups that went on to recover.

What to Watch

Miller said he is watching employment, inflation and the rest of earnings season, and over the longer run the equity and debt capital markets. He also named the Fed’s Dec. 9 decision. Asked about a year-end rally, he predicted “a relaxed Santa rally.” For the baskets here, the test is simple: whether the Nov. 5 lows hold, and whether the capital markets stay open to the companies doing the building.