Coherent's Record Quarter and the Five Questions Behind the Sell-Off

Coherent's Record Quarter and the Five Questions Behind the Sell-Off

Coherent just reported the strongest quarter in its history — $2.05 billion in revenue, up 33.7% year over year, and $1.74 of non-GAAP earnings per share, both comfortably ahead of the roughly $1.99 billion and $1.62 the sell-side was modeling. It guided the next quarter to about $2.3 billion at the midpoint, a raise well above the $2.14 billion analysts had penciled in, and told investors that data center and communications — the AI optics engine of the business — was 79% of revenue with orders booked into 2028 and beyond.

The stock fell anyway. It has kept sliding since the report, and in today's session it dropped another 12.75% to $306, roughly 30% beneath its $440 high, on a market cap near $60 billion.

That is not a contradiction. It is the market answering the same questions the analysts on these calls keep asking, and the questions have not changed in six months. Back in February, when the quarter was $1.69 billion and the stock was $230, the same five analysts — JPMorgan's Samik Chatterjee on demand visibility, Raymond James's Simon Leopold on 1.6T and co-packaged optics, Stifel's Ruben Roy on margin mix, Barclays' Thomas O'Malley on backlog, Jefferies' Ezra Weener on indium phosphide sourcing — set a template. Last week's call covered the same ground. Walk through the five, and you can see why a record quarter produced a falling stock.

Question one: How far out is the demand actually booked?

The visibility question is the oldest on the list, and it has the most rehearsed answer. In February, CEO Jim Anderson told JPMorgan the demand picture was the best it had ever been, with most of calendar 2026 already booked and 2027 filling quickly. This quarter the company extended the claim: customer purchase orders and long-term agreements now run into calendar 2028 and beyond, with minimum take-or-pay volumes — contracts that require the buyer to pay whether they take delivery or not — and agreed prices.

In one sense the question has been answered so decisively that it is no longer the point. Coherent is not a demand story anymore; it is a supply story. Demand visibility this long out is precisely the kind of guarantee that ends up being decomposed per unit when the cycle turns, and take-or-pay protects near-term volume, not the durability of hyperscaler buildout. But the reason analysts moved off this question is that it stopped being the binding constraint. The binding constraint is now inside Coherent's own factories.

Question two: Can the indium phosphide ramp actually deliver margin?

This is the engineering question, and it is the heart of the whole earnings call. Management was explicit that indium phosphide laser output — not assembly or test capacity — is the ceiling on transceiver growth. The numbers attached to the answer are impressive: June-quarter InP laser production was up roughly 80% year over year, internal InP output is set to double by the end of the current September quarter, a quarter ahead of plan, with a third six-inch facility in Zurich coming online in the first half of calendar 2027. The cost logic is the payoff: six-inch wafers yield roughly four times the output of three-inch wafers at about half the cost, and yields across CW lasers, electro-absorption modulators, and photodiodes now beat the old lines.

This is the kind of claim that should survive scrutiny, because the cost structure is real and it compounds. An 80% year-over-year jump in the coefficient of the business's main input, one quarter ahead of schedule, is execution, not propaganda.

And then there is the number that should give every shareholder pause. With data center optics at 79% of revenue and the next quarter's data center growth guided above 80% year over year, management guided non-GAAP gross margin to only a 39.5% to 41.5% range, targeting "above 42%" for the full fiscal year. Think about what that says. The hottest product line in technology, at the center of the AI buildout, in a capacity-constrained market, still caps gross margin in the low forties. That is the tell: the optics supplier does not hold the pricing power. The hyperscalers that buy this stuff do. Optics content is a small slice of total cluster cost — which is exactly why the buyer can squeeze the seller every generation and still get the capacity it needs. All that InP engineering buys low-40s gross margin, not pricing power.

Question three: What does the 1.6T and co-packaged optics transition do to the business you already have?

The transition question is where this story gets genuinely complicated, and the analysts keep pressing it because nobody has an answer yet. On one hand, the 1.6T ramp accelerated relative to the guidance given just three months ago, with 800G still growing. On the other, co-packaged optics — putting the optical engines directly on the switch or accelerator package rather than in a pluggable transceiver — is coming faster than management previously let on: initial CPO revenue is now guided for the December quarter, with scale-up revenue in the second half of calendar 2027, and management says essentially every large strategic customer has a CPO or near-packaged optics program underway, an incremental addressable market it sizes at more than $15 billion.

Here is the part the market is quietly discounting. Coherent's current record revenue is built on pluggable transceivers. CPO does not add a business on top of that business; it re-architects the supply chain under it, moving the optics content into the package and consolidating module-level margin into whoever owns the switch and the digital signal processor — think Broadcom and Marvell, not the laser vendor. Management's answer, that co-packaged and near-packaged content per socket is comparable because Coherent supplies the whole stack — lasers, isolators, fibers, photonic integrated circuits — is the bull case. But nothing on the call proved that the margin survives the migration. The question keeps getting asked because the outcome is genuinely unresolved, and the stock is priced as if the bull case is guaranteed.

Question four: Are you turning this boom into cash?

This is the number no analyst had to ask about, because the cash flow statement answers it. Over the trailing year Coherent generated about $80 million of operating cash flow while spending more than $1.1 billion on capital expenditures — netting negative free cash flow of roughly $1 billion, a negative free-cash-flow margin near 8%, with inventory sitting at $2.6 billion. Revenue grew more than half a billion dollars year over year in the June quarter alone, and the cash did not follow.

Read that plainly. The fastest-growing franchise in AI infrastructure is buying its growth — the fab capacity, the wafers, the Zurich site — faster than the business earns it. This is the affordability question in its rawest form: a $60 billion company generating $80 million of trailing operating cash. It is not a red flag in isolation; a company this far into a capacity super-cycle is supposed to spend ahead of revenue, and net debt is a manageable $1.2 billion after last year's divestitures. But it means the entire bull case rests on the margin and cash conversion arriving on schedule — which brings you straight back to the 42% gross margin ceiling in question two. The boom is real. Whether it is profitable is still a forward promise, not a recorded fact.

Question five: How fragile is the pricing?

The last question is the one management itself flagged, on a record quarter, which should tell you how seriously to take it. Laser pricing, in the company's own framing, is a "live debate" in the market. US import restrictions are already influencing how customers decide where things get manufactured. Industrial revenue, the legacy business underneath the AI optics rally, fell 15.8% year over year. Competitive pressure from international players is called out as a risk to pricing and margin expansion.

Every one of these is a version of the same threat: the forces that can erase the InP cost advantage fastest are price erosion and a forced re-wiring of the supply base. Hyperscaler optics pricing falls per bit every generation, as it always has. The take-or-pay contracts protect volume and near-term price; they do not protect the five-year price curve, and they do not protect against an import regime that pushes manufacturing decisions — and eventually orders — toward whoever can build in the right geographies.

What the questions add up to

Coherent is not a company that missed its quarter. It beat, it raised, it put up record numbers in the hottest segment of the market, and the complex around it de-rated anyway — Fabrinet fell 19% today, Ciena and Lumentum roughly 9% to 10%, with Lumentum still trading near 26 times sales. The September-quarter call six months ago was rewarded; this one sold off.

That is what the five questions add up to: they are one question repeated from different angles. At what cost is this growth, and is that cost sustainable? The income statement says the AI optics boom is real. The margin guide says the pricing power sits with the buyers. The cash flow statement says the cost is still being paid. And the valuation — 8.4 times trailing sales, 78 times trailing earnings, 67 times EBITDA — says the market has already front-loaded the margin and cash conversion Coherent has not yet recorded.

The judgment here is directional, not hedged: the argument about this stock has moved off demand entirely and onto whether a low-40s gross margin and negative free cash flow can compound into the numbers the multiple implies. The break condition is visible. If gross margin holds above 40% while free cash flow turns positive on a revenue base that keeps growing above 30%, the record quarter was a starting point rather than a peak. If the InP ramp slips, or CPO consolidates the margin into the switch vendors, or the pricing debate resolves downward, then a stock that has already de-rated 30% from its high has further to correct. That is the trade in one sentence: the boom is priced, the cost is not.