Marvell’s $115bn bid for Lumentum reads like a chip industry merger – for data centre operators it points at a component that most procurement teams have never had to track.
By Steven Foley | Datacenter Focus | 13 November 2026
Every fast link inside an AI cluster runs on light, and light needs a laser. On 2 November, Marvell agreed to buy optical specialist Lumentum in an all stock deal worth $115bn and investors punished both stocks on the news – sending Marvell down 12.7% and dragging Lumentum lower alongside it – over concerns about share dilution and valuation. Operators must look past the share prices though and ask what the deal says about supply.
The part that isn’t going away
The networking industry is rearranging where light is produced, not removing the need for it. Analysts at Counterpoint Research note, newer co-packaged optics (CPO) designs move the laser out of the switch and into an external light-source module. The laser is still there, and it runs at higher power.
This creates an awkward combination; The laser is the part operators can swap, and it is also the part in shortest supply. In October, Lumentum chief executive Michael Hurlston told Bloomberg that the company’s AI optical component capacity is completely sold out through early 2029, leaving it unable to meet roughly 70% of near-term demand for certain products.
The queue forms before you arrive
Is your place in the queue already spoken for? In March, Nvidia signed a multiyear, non exclusive agreement with Lumentum featuring a $2bn equity investment, multibillion-dollar purchase commitments, and future capacity access rights. Earlier this month, Lumentum also disclosed a second major co-packaged optics win backed by more than $5bn in purchase commitments. Most data centre operators don’t buy indium phosphide lasers directly – they buy switches and transceivers and the lasers arrive inside them – that means your delivery timeline is dictated by your switch vendor’s supply contracts – and not yours. Is this the GPU allocation story all over again with the largest balance sheets being served first?
What the merger changes
Until now, Marvell owned no laser fabs, whereas rival Broadcom manufactures its own lasers in house. Buying Lumentum closes that vertical integration gap – it also narrows the merchant market. Counterpoint analysts point out that any switch or optics vendor competing with Marvell will now want an independent second source for lasers, pushing more demand toward rivals like Coherent.
Marvell’s leadership has stated it intends to maintain Lumentum’s open merchant supply model – this is a stated intention and not a binding contract. The deal remains subject to shareholder and regulatory approval with closing not expected until the first half of 2028, leaving more than a year in which wafer and laser allocation decisions will be made behind closed doors first, and scrutinised by regulators later.
The operator takeaway
Four things operators should consider doing:
- Audit your optics bill of materials. Ask each switch and transceiver vendor who manufactures their high-speed lasers, and verify whether they have a qualified second source.
- Decouple module lead times. Request delivery schedules for external light source modules and optical transceivers separately from switch chassis. One is useless without the other.
- Elevate optics in critical-path planning. Put laser and transceiver supply into your facility delivery plan alongside grid power, transformers, and cooling slots.
- Watch antitrust reviews through 2027. How US, European, and Asian regulators view indium phosphide laser concentration will determine how much supplier choice remains by 2028.
The infrastructure question has mostly been about who can get GPUs, is it now shifting towards who can get light? Operators who interrogate their switch vendors’ optical supply chains now, will have answers before a missing laser holds up a live hall.
