Start with the number the sell side refuses to lead with: Coherent trades at 160 times trailing earnings. Lumentum sits at 146 times. These are not growth multiples. They are momentum-cult multiples—the valuation a name receives once the underlying business has stopped being the primary driver and the only remaining variable is whether tomorrow’s inflows exceed today’s. COHR is up 105 percent year-to-date heading into the session. LITE is up 142 percent. Both just surrendered double-digit percentages in a single afternoon—Coherent off 12 percent, Lumentum off 7 percent—and every wire covering the move reached for the identical language: “profit-taking ahead of earnings,” “no stock-specific catalyst,” “classic de-risking.” I have read that exact formulation about this trio at least four times since May. July 2. July 15. July 28. Now today. At some point “no catalyst” ceases to function as an explanation and becomes an admission that no one watching the sector actually understands what is moving these stocks, higher or lower, on any given Tuesday.
Here is the simpler read. This is a crowded trade carrying a floating cost-of-capital problem that has been dressed up as a fundamentals story. Every leg higher is attributed to hyperscaler capex commitments, optical-transceiver export limits on Chinese suppliers, CPO backlog announcements—real demand inputs, and I am not disputing them. Datacenter interconnect revenue is genuinely growing 40 percent-plus at several of these names. But a legitimate demand story trading at 150 times does not require “no-catalyst” 10 percent air pockets every three weeks to reset. It requires those air pockets because position sizing across momentum funds and retail options flow has become so top-heavy that any pause in incremental buying is functionally indistinguishable from selling.
Contrast that with what Bank of America is doing on the other side of the AI-infrastructure ledger this week: reiterating its buy on Nvidia, calling the shares cheap “despite how fast it’s growing,” and explicitly labeling the circular-financing concerns around AI capex—the vendor-financing-the-customer arrangements that have been circulating since spring—“overblown.” Possibly. I would trust that dismissal more if it arrived with arithmetic rather than adjectives. Nvidia reports fiscal Q4 on August 26, one day before Jackson Hole opens, and the consensus still expects a beat-and-raise on the Vera Rubin ramp. Fine. Notice, however, the asymmetry in how skepticism is allocated: the optics suppliers, one layer down the AI capex chain and posting real datacenter growth, get punished on vibes and multiples that are no more extreme than Nvidia’s own historical precedent. Meanwhile the largest node in the entire chain receives a “concerns are overblown” verdict from a research desk that maintains fee relationships with nearly every participant in the ecosystem.
Place those two observations side by side and the actual risk map emerges, not the one in the headlines. AI infrastructure spending is not fictional. The market simply has no reliable mechanism at present for locating where in the stack the real fragility resides—supplier, hyperscaler, financier, or the equity holder three steps removed buying LITE at 146 times because the chart says so. SOXX barely registered the optics selloff, which confirms the move is idiosyncratic to the photonics names rather than a broad semiconductor unwind. That is the more troubling signal, not the less troubling one. A sector-wide derating can be explained with a single macro variable—rates, dollar, whatever. A repeatedly isolated, repeatedly unexplained collapse inside one narrow sub-segment of the AI trade, four times in ten weeks, each time carrying the identical “no catalyst” write-up, indicates that the market’s price-discovery mechanism for this corner of the complex has effectively ceased functioning. No one is modeling COHR or LITE. They are trading an options-flow feedback loop that has been rebranded as an earnings preview.
Elsewhere in the capex ecosystem, capital is already rotating into the picks-and-shovels tier that does not carry triple-digit multiples. Teledyne is paying $18.90 a share in cash for Varex Imaging. Blackstone’s Safe Harbor Marinas is taking MarineMax private at a 46 percent premium. Neither transaction is AI-adjacent, yet both represent cash buyers stepping into hard assets at the precise moment the marginal dollar in growth equities is growing more skeptical, not less. That is the signal I would actually trust over another “profit-taking, no catalyst” wire note: when the M&A desk begins moving on tangible cash-flow businesses in the same week the most crowded momentum names in tech suffer their fourth unexplained double-digit drawdown since May, someone with real capital has already rendered a judgment the retail options flow has not yet absorbed.
Recommendation: treat Wednesday’s Coherent print and Tuesday’s Lumentum numbers as the actual data points, and stop pretending the intervening chart action carries informational content.
