Marvell agrees $115 billion Lumentum acquisition to expand AI connectivity
Marvell Technology, a US designer of semiconductors for data infrastructure, has agreed to acquire optical technology manufacturer Lumentum in an all-share transaction with an implied enterprise value of $115 billion.
Announced on 2 November 2026, the agreement would unite Marvell’s custom processors, networking chips and electrical and optical interconnect technologies with Lumentum’s lasers, optical transceivers and specialised manufacturing. The companies say the combination would help them design these technologies together and supply optical connections at volume, addressing the growing demands of AI infrastructure.
Lumentum shareholders would receive 4.5 Marvell shares for each share they hold, giving them approximately 33% of the combined company on a fully diluted basis. Marvell chairman and chief executive Matt Murphy would retain those roles. The transaction is expected to close in the first half of 2028, subject to shareholder and regulatory approvals and other customary conditions.
What Marvell and Lumentum bring to AI connectivity
Based in Santa Clara, California, Marvell supplies custom computing chips, switches that direct network traffic, and circuitry supporting high-speed electrical and optical connections. Its customers use these technologies to move, process, store and secure data.
The acquisition would build on that existing optical business by adding more of the components and manufacturing processes needed for complete connections.
Lumentum, headquartered in San Jose, supplies lasers, optical modules and subsystems for data centres, telecommunications and other applications. Its optical transceivers send and receive information using light, while its optical circuit switches direct light signals between fibre connections.
The company also brings expertise in making optical devices from compound semiconductors, including indium phosphide. These materials enable high-performance lasers, adding a different manufacturing capability to Marvell’s silicon-based chip technologies. The purchase would give Marvell direct access to those production capabilities alongside its existing semiconductor design and optical technologies.
Why AI data centres need optical connections
The value of combining these capabilities becomes clearer when looking at how large AI systems operate. When an AI task is distributed across processors, they must exchange intermediate results as their calculations progress. A processor waiting for information from another cannot immediately continue the dependent work, so the connections between them help determine how effectively the system uses its computing capacity.
Keeping processors supplied with data requires connections that can carry sufficient information without consuming excessive power. Copper remains important, particularly over short distances, but maintaining electrical signal quality becomes more demanding as transmission speeds rise, reducing its practical reach.
Optical links carry information using light and already connect data centres and racks. Bringing optical conversion closer to processors can extend the reach of high-speed connections while shortening the electrical paths that precede them.
The acquisition is intended to address this range of requirements, from connections between separate data centres to links within advanced semiconductor packages. The engineering challenge changes with the distance involved, but in each case the electronic and optical components must operate as a coordinated system.
Why chips and co-packaged optics are designed together
To understand that coordination, consider what happens when data enters an optical link. Electronic circuitry prepares it for transmission. A laser supplies light, whose properties are changed to encode the information being sent. At the receiving end, a photodetector converts the incoming light into an electrical signal for further processing.
The electronics that drive the optical devices must match their operating requirements, while the receiver must reliably recover the information. The layout and packaging also affect signal quality and power consumption. Choosing a faster optical device alone therefore does not determine the performance of the connection; the surrounding circuitry and physical arrangement matter too.
Co-packaged optics (CPO) addresses part of this problem by placing optical engines alongside electronic chips within the same package. Shortening the electrical path between them can reduce signal losses and the power needed to move information.
Marvell has announced such an architecture for custom AI accelerators, processors designed to handle AI workloads. It places optical engines in the same package as the computing chip, allowing data to enter optical fibre without first taking a longer electrical route across a circuit board or through copper cabling.
Lumentum would contribute additional optical devices and production expertise to this integration strategy. The companies intend to offer combinations of components validated together, reducing the work customers need to do when developing new equipment and shortening the path to deployment.
Delivering those combinations at volume also requires repeatable assembly and testing. The deal’s stated rationale consequently includes coordinating semiconductor design, advanced packaging and specialised manufacturing, as well as expanding the product portfolio.
Consolidation across AI chips and optical connectivity
Marvell has already been assembling these capabilities through acquisitions spanning electronic switching and optical transmission.
On 2 February 2026, Marvell completed its purchase of Celestial AI, whose Photonic Fabric technology is designed to connect large AI systems using high-bandwidth, low-latency optical links. On 10 February, it completed the acquisition of XConn Technologies, adding electronic switching products and expertise for directing data between devices.
Marvell then announced its acquisition of Swiss photonics company Polariton Technologies in April. Its specialised devices encode information onto light, complementing the electronic switching and optical interconnect expertise acquired earlier in the year.
The pattern reaches beyond Marvell. Processor designer AMD announced its acquisition of photonic chip developer Enosemi in May 2025, linking the purchase to the development of optical connections and co-packaged optics for future AI systems. Together, these deals point to semiconductor companies bringing more of the technology for moving data into their own businesses.
Investment and supply agreements provide another route to securing optical technology. In March 2026, AI computing company NVIDIA announced a $2 billion investment in Lumentum as part of nonexclusive strategic agreements covering research, manufacturing expansion, purchase commitments and future access to laser-component capacity. The arrangement tied funding for optical production to the requirements of a major AI infrastructure supplier.
Marvell’s proposed acquisition would bring a broader set of these activities under common ownership. For customers building AI systems, the intended benefit is practical: more of the work of matching chips, light sources, optical receivers and packaging could happen before those components reach their own development teams. That would help turn demanding connection specifications into equipment ready for production and deployment.
Liked this article? You can support our independent journalism via our page on Buy Me a Coffee. It helps keep MoveTheNeedle.news focused on depth, not clicks.