Marvell Bets $115 Billion on AI's Optical Future

Marvell Technology is buying optical components maker Lumentum in a $115 billion all stock deal, one of the biggest technology deals of 2026 and one of the boldest bets yet that the AI spending boom is not a passing phase. Announced on Monday, November 2, the deal joins a custom chipmaker with a company that makes the light based parts AI data centers use to move data at high speed. If the bet pays off, Marvell will own a much larger slice of the plumbing inside every large AI system. If it does not, it will have spent a fortune at the top of a market that investors are already starting to question.

What the deal is

Under the agreement, Marvell will pay $1,132 per Lumentum share in stock, a 23.9 percent premium to Lumentum’s last unaffected closing price on October 28, before deal talks were first reported. The equity value is $102.54 billion, and the enterprise value, including debt, is $115 billion. Marvell shareholders will own about 67 percent of the combined company, and Lumentum shareholders about 33 percent.

Marvell expects the deal to close in the first half of 2028, though it said closing could extend to August 2028 if regulatory approvals are still pending. That long timeline matters. This is not a deal that changes the market tomorrow. It is a deal that assumes the market will still be growing strongly two years from now, and that regulators in the United States and elsewhere will allow two suppliers in the AI supply chain to combine.

Marvell said the combined company will address a total market of $550 billion. That number is a claim about the future, not a measure of current sales. Still, it shows how Marvell wants investors to see the deal: not as a chipmaker buying a parts supplier, but as the creation of a much broader AI infrastructure company.

Why optical parts matter now

Modern AI systems are not limited mainly by how fast a single chip can calculate. They are limited by how fast thousands of chips can talk to each other. Training a frontier model means moving enormous amounts of data between processors, racks, and buildings, again and again. Copper wires run out of room at those speeds and distances. Light, carried through fiber and switched by optical components, does not.

That is why Lumentum matters. Its lasers, transceivers, and optical components are the parts that turn electrical signals into light and back again, letting data centers move information quickly without melting down or slowing to a crawl. As AI models grow and clusters get larger, demand for those parts has grown with them. Marvell is betting that demand will keep outrunning supply.

Marvell has already been moving in this direction. It closed a $3.5 billion purchase of Celestial AI in February 2026, a photonics company, as rising use of AI agents and chatbots drove demand for faster optical links between chips. Lumentum is the same bet at a far larger scale.

The race with Broadcom

The deal is also a direct challenge to Broadcom, the larger rival that has long been Google’s main custom chip partner and offers customers a broader menu of AI data center hardware under one roof. Marvell wants to close that gap by selling more of the system, not just one piece of it.

Marvell has the momentum to try. Just three years ago it was a relatively little known chipmaker. Its market value has since risen above $170 billion, helped by its work helping technology giants such as Amazon develop custom AI chips and networking technology for data centers. In August it signed a deal to supply chips that sit alongside Google’s AI processors, worth up to $120 billion in sales through fiscal 2033, with Google gaining the right to buy a $12.2 billion stake. Its annual revenue has more than doubled, from $3 billion in fiscal 2021 to $8.2 billion in fiscal 2026. Nvidia chief executive Jensen Huang has called Marvell the next trillion dollar company, and Nvidia invested $2 billion in each of Marvell and Lumentum earlier this year.

That history explains the confidence, but it also raises the stakes. Marvell’s acquisition of Lumentum is its largest purchase since its $10 billion buyout of Inphi in 2021, more than ten times larger. A deal of this size cannot be quietly undone if the strategy is wrong.

What could go wrong

The first risk is timing. The deal will not close until 2028, and the AI market has already shown it can turn quickly. Chip stocks sold off sharply after deal talks were first reported late in October, following a weak forecast from Intel, with Marvell down about 22 percent and Lumentum down about 8 percent in that period. Investors are no longer willing to pay any price for AI exposure. They are asking harder questions about who will actually profit, and when.

The second risk is regulation. A $115 billion combination in the AI supply chain will draw close scrutiny, and the long closing window is itself a sign that Marvell expects a demanding review. Geopolitics adds another layer: Lumentum competes with US listed Coherent and China’s Zhongji Innolight, and a proposed US ban on new Chinese optical transceivers could shift demand toward domestic suppliers, which would help Lumentum, but it also shows how political this market has become. A deal that depends on cross border approvals is exposed to decisions Marvell cannot control.

The third risk is integration. Marvell is buying a company roughly half its size, in a related but distinct technology area. Combining chip design with optical manufacturing, aligning product roadmaps, and keeping Lumentum’s customers, some of whom compete with Marvell’s customers, will take years of careful work. The $550 billion market figure will mean little if the combined company cannot execute.

What it means for builders

For founders and engineers building on AI, this deal is a useful signal about where the cost and the bottlenecks are moving. The most valuable parts of the AI stack are no longer only the models. They are the connections between models, the networking, and the infrastructure that lets large systems run at scale. When a chipmaker pays $115 billion for optical parts, it is saying that moving data is now as strategic as processing it.

That has practical consequences. First, expect infrastructure costs to stay high, and expect the suppliers of that infrastructure to consolidate. Fewer, larger suppliers usually mean less negotiating power for buyers, at least in the short term. Second, watch for more vertical integration: companies that once sold one layer of the stack are now trying to sell several, which means product choices made today may lock you into a supplier’s broader system tomorrow. Third, keep an eye on the physical side of AI. Power, cooling, fiber, and optics are becoming first order constraints, not afterthoughts, and they will shape which AI products are economical to run.

What to watch

Three things will tell us whether this bet was wise. The first is the regulatory path: whether approvals arrive on schedule, and whether any conditions limit what the combined company can sell or to whom. The second is demand: whether Big Tech’s AI spending keeps growing through 2027, or whether the recent investor caution turns into real budget cuts. The third is execution: whether Marvell can turn two companies into one product story that customers actually want to buy, rather than just a bigger company on paper.

Marvell is not buying Lumentum because the AI boom is certain. It is buying Lumentum because, if the boom continues, the companies that move data the fastest will be among its biggest winners. That is a reasonable bet. At $115 billion, it is also an expensive one, and the next two years will test whether Marvell timed it right.