Marvell Buys Lumentum for $115 Billion in the Biggest Chip Deal of the A.I. Era

Set Trending Topics as a preferred source on Google.

Marvell is paying $115 billion for a company most people have never heard of, and the market is sending both stocks down.

U.S. chip designer Marvell Technology is acquiring photonics specialist Lumentum in an all-stock transaction. Lumentum shareholders receive 4.5 Marvell shares for each of their shares, which works out to about $1,132 (about €974) per share, a premium of 23.9 percent over the price before the first takeover rumors. Including debt, the deal values Lumentum at $115 billion (about €99 billion), with an equity value of $102.5 billion. After closing, Marvell shareholders will hold about 67 percent of the combined company and Lumentum’s about 33 percent. Both boards approved the deal unanimously; Marvell chief Matt Murphy stays on as chairman and CEO, and Lumentum CEO Michael Hurlston joins the Marvell board along with two other directors. The transaction is not expected to close until the first half of 2028, pending approval by shareholders and antitrust authorities.

For scale: Broadcom paid $69 billion for VMware in 2023, Dell paid $67 billion for EMC in 2016, and the largest chip takeover ever attempted, Broadcom’s $130 billion bid for Qualcomm, was blocked by the U.S. government in 2018. Marvell and Lumentum call it one of the biggest tech deals ever; according to Bloomberg, it is the year’s largest merger by equity value. For Marvell, it is by far the biggest acquisition in its history, the previous record being the $10 billion purchase of optical chipmaker Inphi in 2021.

What Marvell Is Known For

Marvell is one of those chip companies that profit from the A.I. boom without building a graphics processor of its own. Founded in 1995 and based in Santa Clara, the company designs semiconductors for data infrastructure: networking chips, switches, storage controllers, chips for optical connections in data centers and, for several years now, above all custom A.I. accelerators, so-called custom XPUs. Alongside Broadcom, Marvell is the most important partner for the hyperscalers developing their own A.I. chips as an alternative to Nvidia: Marvell works on the Trainium chips for Amazon and the Maia series for Microsoft, and in August it signed a contract with Google for chips that sit alongside Google’s own TPUs, worth up to $120 billion (about €103 billion) through fiscal 2033. Google secured the right to a $12.2 billion stake in the process, Nvidia invested $2 billion in Marvell earlier this year, and Nvidia CEO Jensen Huang called the company the next “trillion-dollar company” over the summer.

The numbers explain the rise. Revenue grew from $3 billion in fiscal 2021 to $8.2 billion (about €7 billion) in fiscal 2026, and the market value climbed to more than $170 billion. At its investor day in October, Marvell raised its revenue target for 2028 to $20 billion and projected $70 billion to $90 billion for 2031, about $37.5 billion of that from interconnect alone. In February, Marvell already bought the photonics start-up Celestial AI for $3.5 billion. Lumentum, for its part, was spun out of network equipment maker JDSU in 2015 and builds high-performance lasers, optical transceivers and optical circuit switches; the San Jose company also runs its own fabs for compound semiconductors such as indium phosphide. In the past quarter, Lumentum posted revenue of $1.26 billion and expects up to $1.63 billion for the current one, and here too Nvidia has held a $2 billion stake since March.

Why Light Instead of Copper

The reason for the deal lies in the physics of A.I. data centers. The more chips are connected into a cluster, the more data has to flow between them, and copper cables are hitting their limits in bandwidth, distance and energy consumption. The industry is therefore shifting to optical connections, right down into the chip packages themselves, known as co-packaged optics. So far, Marvell has supplied the electrical side and Lumentum the lasers and optical components. Together, the two want to offer fully integrated, pre-validated optical solutions and expand their addressable market to $550 billion (about €473 billion), according to the press release. Murphy speaks of “an opportunity that comes along once in a generation,” and Hurlston points to a second major co-packaged optics win backed by purchase commitments of more than $5 billion. Marvell promises cost synergies of more than $1 billion (about €860 million) a year and a positive contribution to adjusted earnings per share in the first year after closing.

The Market Has Doubts

Investors reacted unusually. Lumentum shares lost about 6 percent on Monday even though the company is the one being acquired, and Marvell fell 3 percent. Since the talks leaked in late October, Marvell has lost more than 20 percent of its value, 13 percent on the day of the leak alone. That Lumentum trades well below the implied purchase price points to doubts about Marvell stock as a currency, about the long road to closing in 2028 and about antitrust review. The break-up fees are correspondingly high: Marvell pays $5.8 billion if it walks away for a competing offer and $3.2 billion if regulators block the deal. “Investors are right to ask whether Marvell is paying too much,” Axel Rudolph of IG Group told Reuters. Add to that the general nervousness about the hyperscalers’ A.I. spending, with margins tightening and cash flows thinning. Competitors in the optics business, Coherent in the United States and Zhongji Innolight in China, gained instead, Coherent by 2 percent after a 25 percent slide over the past month.

For the deal’s supporters, the sell-off is a misunderstanding. Analysts at Stifel and Piper Sandler reaffirmed their buy ratings on Marvell, arguing that whoever is first to cover the transition to optical interconnect end to end, from the laser fab to the custom chip, sets the standard for decades, and that a proposed U.S. ban on new Chinese transceivers could shift additional demand to Lumentum. Whether the math works out comes down to a question that preoccupies the entire A.I. industry: whether the hyperscalers keep building out their data centers in the coming years the way the $550 billion forecast assumes.