πŸ”†

COHR

πŸ’‘
Thesis. Coherent owns the scarce laser layer and proved it again β€” Q4 revenue up 34% with non-GAAP gross margin expanding 215 basis points to 40.2%, an order book stretching into calendar 2028, and NVIDIA's $2B behind it. The complication is that Broadcom is now building its own EML and CW laser capacity, and the stock fell on a beat.
Reviewed 11 September 2026, after Q4 FY2026 results. Next review after Q1 FY2027 earnings on 4 November 2026.

What it does

Coherent builds optical transceivers β€” the light engines that move data between racks in an AI cluster. 800G is the volume product today, 1.6T is ramping. Datacom and communications is now around 75% of revenue. There's also a legacy industrial and materials laser business, now a shrinking share.
The part that matters competitively is that it's vertically integrated. Coherent grows its own indium phosphide laser chips, on the industry's first 6-inch production line in Sherman, Texas.
A 6-inch wafer yields more than four times the chips of a 4-inch wafer at less than half the cost per chip. In a business where margins are thin and everyone buys the same way, owning the cheapest fab is the moat.

Bull case

  • Q4 FY2026, reported 12 August: revenue $2.05B, up 34% β€” and up 42% pro forma once the aerospace and defence divestiture is adjusted out
  • The margin story is the proof. GAAP gross margin 38.5% (+277bps), non-GAAP 40.2% (+215bps). On a physical manufactured product, that is hard-won
  • Earnings are compounding faster than sales. Non-GAAP EPS $1.74, up $0.74 on the year. GAAP EPS $1.19, a $2.02 swing. For FY26 as a whole, non-GAAP EPS grew at more than twice the rate of revenue
  • Unusual visibility for a component supplier. The order book extends into calendar 2028, with a multi-year NVIDIA supply agreement running through 2030
  • NVIDIA put roughly $2B in in March 2026 β€” part of about $4B split with Lumentum β€” explicitly to accelerate indium phosphide capacity
  • The 1.6T cycle is starting while 800G is still growing through CY2026
  • Optical circuit switching is real revenue, with seven customers shipping and Google's Ironwood architecture validating the category
  • Western supply chain is an advantage against Chinese suppliers as sourcing gets scrutinised

Bear case

  • Optics is brutally cyclical, and this is the good half of the cycle. The industry has a long history of boom followed by bust
  • Broadcom is moving in. It disclosed in September that it is investing in EMLs and CW lasers for its own demand β€” expanding directly into the layer Coherent and Lumentum own
  • Innolight dominates merchant 400G and 800G volume, which means permanent price pressure from below
  • Co-packaged optics is the structural risk. If optics moves onto the switch ASIC, the pluggable transceiver market shrinks late-decade
  • Merger debt is still on the balance sheet, which amplifies any downturn
  • Hyperscaler orders are lumpy by nature, so a quarter can miss for reasons that say nothing about demand
  • The market is already sceptical. The stock fell 4% after hours on a beat with above-consensus guidance, having dropped 12% in pre-earnings de-risking

Major customers

  • NVIDIA β€” invested roughly $2B in March 2026 to accelerate indium phosphide capacity, and is a named collaborator on silicon photonics for the Spectrum-X platform. A customer that buys equity to guarantee supply is the strongest demand signal there is
  • Hyperscalers β€” buying transceivers directly and through switch vendors. Datacentre is now 75% of revenue, up from 41% pro forma a year earlier
  • Optical circuit switch customers β€” seven named as shipping, with Google's Ironwood architecture validating the category
  • Cisco, Ciena and other systems vendors β€” components and transceivers
  • Industrial and aerospace β€” the legacy laser business, now a shrinking share
πŸ”—
Read-through.
πŸ›οΈ
NVDA
platform roadmaps are the single most important external variable here β€” Coherent's demand curve is drawn by NVIDIA's product cadence.
Downstream, Coherent supplies
πŸ”†
CIEN
,
πŸ•ΈοΈ
ANET
and the module makers, so its capacity commentary leads their product availability. It sells laser chips to
🏭
TSEM
for silicon photonics work.
πŸ”†
FN
manufactures for Coherent, so Fabrinet's volumes partly reflect Coherent's shipments.
It competes with
πŸ•ΈοΈ
AVGO
and Mitsubishi in EML supply, and sits alongside
πŸ”†
LITE
as one of the two names that own the laser bottleneck.
πŸ›οΈ
TSM
naming lasers as a co-packaged optics bottleneck confirms the scarcity from the foundry side.
The tell: Broadcom investing to build its own lasers confirms the shortage before it threatens the position. Watch which of those two it turns out to be.

What would change the view

  • Gross margin trend β€” the 6-inch indium phosphide line either proves out in the margin or it doesn't
  • 1.6T program wins against Innolight and Lumentum
  • Debt paydown from the merger
  • Co-packaged optics adoption pace, and whether it expands or shrinks the addressable market
  • Whether Broadcom's in-house laser capacity starts to show up as competition rather than confirmation
  • Order book duration β€” it runs into 2028 today; any shortening is the early warning

Update log

  • 11 Sep 2026 β€” Reviewed. Broadcom's EML and CW laser investment added to the bear case as a new competitor, and TSMC naming lasers a co-packaged optics bottleneck noted as confirmation of scarcity. Verdict held.
  • 12 Aug 2026 β€” Q4 FY2026 results: revenue $2.05B (+34%, +42% pro forma), non-GAAP gross margin 40.2% (+215bps), non-GAAP EPS $1.74. Beat with above-consensus guidance; stock fell 4% after hours on a name that had already run.
  • 25 Jul 2026 β€” Full review.

Research and education only β€” not investment advice.