Marvell Agreed to Pay $115 Billion to Get Past 1.25 Metres of Copper

There are two ways to read Marvell's agreement to buy Lumentum. The first is as a valuation event: $115 billion in stock, one of the largest technology acquisitions on record, about 1.7 times Microsoft's $68.7 billion Activision Blizzard purchase on its face. The second is as a measurement, and it is the more useful one.

The measurement is roughly 1.25 metres. That is how far copper will carry a 400 Gbps signal before it stops being a signal, according to The Register's Tobias Mann. Every time the per-lane speed doubles, that reach halves. Optics do not obey the same rule. A photonic link can carry the same data for kilometres.

Everything else in the announcement follows from that. A rack built on copper is a rack whose compute domain ends at the top of the cabinet. A rack built on optics can extend across an aisle, across a row, across a hall. The difference is not a cable choice. It sets how many racks you need, how much cable tray you fill, how much air you move, and how often you tear the floor up.

What $115 billion buys

The terms are all-stock and fixed. Lumentum holders get 4.500 Marvell shares for each share they hold, an implied enterprise value of $115 billion, about $1,132 per Lumentum share and a 23.9% premium to the unaffected closing prices on 28 October, the last trading day before Bloomberg reported the talks. Marvell holders end up with about 67% of the combined company and Lumentum holders about 33%. TechCrunch, citing the Wall Street Journal, put the equity value at about $80 billion, with the rest debt. Both figures are published; they are not the same measurement and should not be used interchangeably. Both boards approved it unanimously. Matt Murphy stays as chairman and chief executive, and three Lumentum directors, including chief executive Michael Hurlston, join a Marvell board that currently has seven.

What Marvell is buying is the part of the optical chain it did not have: high-power lasers, transceivers, optical circuit switching, and indium phosphide manufacturing, the compound semiconductor layer that makes the lasers possible. Marvell already designs the electrical side, the retimers and transimpedance amplifiers at the front of an optical link, and it already builds custom accelerators for the hyperscalers. It does not own the light.

Lumentum's first quarter, reported the same morning, explains why the price was not negotiated down. Revenue was $1.26 billion, up 136.5% year on year and 25.5% sequentially. Non-GAAP gross margin was 53.2%, non-GAAP operating margin 42.3%, and GAAP net income $360.3 million against $4.2 million a year earlier. Guidance for the current quarter is $1.575 billion to $1.625 billion, which the company said would be about 140% growth on the year-earlier quarter.

Hurlston used the release to disclose a second major co-packaged optics win, "backed by purchase commitments exceeding $5 billion," and said demand visibility for optical circuit switching had expanded enough to make it "a multi-billion-dollar annual revenue stream." Those are commitments, not revenue, and the release gives no delivery schedule. It is still the clearest signal in the transaction that the customers have already voted. For scale, Marvell's own revenue was $8.19 billion in fiscal 2026 and analysts expect about $12.07 billion this year, which puts the headline price at roughly nine and a half times the buyer's expected annual sales.

The 20 kilowatts Nvidia would not spend

Optics stayed outside the rack for as long as they did because of power. When Nvidia showed its first Blackwell rack systems in 2024, Jensen Huang said the company had deliberately avoided optics because they would have added 20 kW to a system already drawing 120 kW under load. An optical transceiver replacing a copper cable costs several watts, and a rack of accelerators has a great many links.

That arithmetic has been changing. Marvell paid about $3.5 billion for Celestial AI in February for near-packaged and co-packaged optics, where the optical engine sits beside the switch or compute die rather than at the end of a pluggable cable. Lumentum's lasers are the input to that design. Put the two together and the phrase in the release, "end-to-end extreme co-design," starts to mean something specific: the laser, the driver, the modulator and the package get designed by one company instead of four.

This is the part of the deal a spec sheet will eventually reflect. Scale-out networking, the links between racks, is already mostly optical. The contest is over scale-up, the links inside a rack between accelerators, where copper still wins on power and latency, and over scale-in, the newer category both companies keep invoking. If optics crosses into the package, the rack stops being a copper chassis with a few fibre uplinks and becomes an optical assembly with compute in it. Rack pitch, cable volume, cooling design and service procedure all change with it.

The market repriced the deal before the ink dried

Marvell's shareholders were less enthusiastic than its board. The stock closed at $171.41 on Monday, down 12.7% on the day and about 30% since 28 October, against a Nasdaq 100 that Bloomberg put down roughly 5% over the same stretch. The company was worth about $221 billion before the talks leaked.

Because the ratio is fixed and the consideration is all stock, that slide is not a separate event from the deal. It is the deal. At Monday's close, 4.500 Marvell shares were worth about $771, roughly a third less than the $1,132 printed in the announcement, and below the price Lumentum traded at before the talks began. A 23.9% premium to $1,132 implies an unaffected close near $914. Bloomberg's Chris Hughes put the effective offer below $800 a share, around 10% under the pre-announcement level, and noted that the promised savings of more than $1 billion a year could justify the $20 billion premium only if the market valued them on the multiples these two stocks recently commanded.

The analysts split along predictable lines, and the split is informative. Bank of America cut Marvell to Neutral from Buy and cut its target to $225 from $400, citing the overhang from an all-stock deal, an extended regulatory process, limited earnings accretion, and potential conflict with customers and channel partners. TD Cowen went to Hold from Buy with a $175 target, arguing the deal lands at or near the peak of an optical mini-cycle and modelling mid-single-digit earnings dilution in calendar 2028. William Blair kept Outperform and called dilution and integration the concerns to watch. Raymond James raised Lumentum to Strong Buy on the view that the market reaction was overdone. Stifel cut Lumentum to Hold.

TD Cowen's Sean O'Loughlin made the sharpest numerical case against the price. He noted that bottom-up consensus for optical transceiver shipments in 2027 implies a market near $100 billion, while his own forecast, built from accelerator shipments, is $47 billion. If the transceiver build runs ahead of the accelerators that consume it, Lumentum's operating leverage works in reverse, and the dilution Marvell is promising to avoid in year one arrives in year two anyway.

The bull case is not stupid, which is what makes the disagreement real. Bernstein analysts expect the optical transceiver market to grow roughly five-fold from this year to nearly $300 billion by 2030, according to Reuters Breakingviews. Bank of America's Vivek Arya, before he cut the stock, argued Marvell is the only custom silicon partner working across every major hyperscaler programme and would win in scale-up optics whichever architecture customers chose. Owning the lasers puts that supply inside the company.

Twenty-one months is a long time to hold a fixed ratio

The merger agreement, signed 1 November, sets an outside date of 1 November 2027, extendable in three stages to 1 August 2028 if regulatory approvals are still outstanding. That is not boilerplate. Marvell has agreed to pay Lumentum $3.2 billion if the deal fails for regulatory reasons with every other condition satisfied, and the same agreement obliges both companies to offer remedies that include divesting businesses, holding them separate, changing contracts, and giving third parties "minimum supply, pricing or other similar commitments."

Read that last clause twice if you buy lasers. A competition authority joining a company that designs custom silicon to a company that makes the scarce component inside everyone else's optical modules has an obvious remedy available: keep the supply flowing on terms. The agreement anticipates it.

China is the open question, and neither company has said anything about it. Reuters Breakingviews' Karen Kwok wrote that Beijing could scrutinise the combined company's control of critical optical components, that an outright block may be less attractive because Chinese manufacturers rely on Lumentum's lasers, and that supply guarantees are the likelier demand. That is a columnist's read, not a disclosed fact. What is disclosed is enough to make it material. Lumentum's most recent annual report lists significant manufacturing in the United States, Thailand, China, the United Kingdom, Slovenia and Japan, and the agreement's definition of regulatory law covers foreign investment on national security grounds, not just antitrust.

There is a commercial version of the same problem. Marvell sells custom chips to companies that compete with Lumentum, and Lumentum sells lasers to companies that compete with Marvell. Bank of America flagged the channel conflict directly. Broadcom, which sells both switching silicon and optics and is the closest thing to a formed competitor, has every reason to make the review interesting.

What to ask before you specify the next rack

None of this changes the physics, and the physics is the part that will still be true when the regulatory files are forgotten. Copper's reach halves with every lane-speed doubling, and the industry is moving from 400G to 800G and beyond on a schedule that does not care who owns whom. The rack you specify in 2027 will be installed in 2029, after this deal is supposed to close, which means the interconnect inside it is being designed now by two companies that will not be one company for another twenty-one months.

That gap is where procurement risk lives. Ask what lane speed the design targets and at what reach, because a vendor quoting 400G at 1.25 metres and a vendor quoting 800G at 0.6 metres are answering different questions. Ask where the copper stops and the optics start, and whether that boundary sits inside the rack or at its edge. Ask who makes the laser, where the fab is, and what happens to supply if the ownership of that fab changes. Ask whether the same interconnect is available from a second source, because a five-billion-dollar purchase commitment from one customer is a signal about demand and also about allocation.

The industry has just said where it thinks the ceiling is. Two boards agreed to combine at a price that only makes sense if optics keeps moving closer to the die, and they agreed to wait up to twenty-one months and pay $3.2 billion if regulators disagree. That is a lot of conviction to put behind a length of copper.

For the hardware teams working out where their next generation sits relative to that ceiling, the useful work starts before the purchase order, not after. At DMC we spend our time on exactly that boundary, the place where a component decision becomes a rack design, a cooling load and a refresh cycle. If you are stress-testing an interconnect roadmap against supply that may change hands before your product ships, let's talk.