Chipmaker Marvell Technology Inc.’s stock moved higher today after the company announced plans to acquire the network interconnect startup XConn Technologies Holdings Inc. in a stock-and-cash deal valued at $540 million.

XConn is a maker of interconnect switching technologies for artificial intelligence and data center infrastructure. The addition will help to strengthen Marvell’s own switching portfolio, the chipmaker said. Switching is necessary to connect large numbers of AI chips into giant clusters, so they can work together to run powerful large language models.

Shares of Marvell rose more than 4% on news of the acquisition, which will also see the company add XConn’s engineering teams to its own networking unit.

The deal will help to fuel Marvell’s efforts to improve its AI infrastructure story and make its custom AI processors a more compelling story for data center operators such as Amazon Web Services Inc. Marvell was once seen as one of the most promising bets outside of Nvidia Corp., and many experts believed it was primed to become a major beneficiary of the AI boom.

But that hasn’t happened. Though Marvell’s business has grown over the last three years, it has struggled to match the momentum of Nvidia or another key rival, Broadcom Inc., which also makes custom AI processors on behalf of large companies.

Marvell has been trying to change that story through acquisitions. Last month, it swooped to buy the optical interconnect technology firm Celestial AI Inc. for $5.5 billion. Optical interconnects are becoming essential in AI, because they enable clusters of AI accelerator chips to communicate with each other with subsecond latencies – much faster than traditional data center interconnects. The acquisition of XConn adds another piece of the puzzle, for its switches can likely be paired with Celestial AI’s optical interconnects to help Marvell build larger clusters of “XPUs,” which are the customized AI processors it develops for the likes of AWS.

Holger Mueller of Constellation Research told SiliconANGLE that the ability to offer tightly integrated accelerator hardware is fast becoming tablestakes in AI factories. This means central processing units and graphics processing units that are closely coupled with storage and networking services. “When you have expertly engineered integration, it reduces the total cost of ownership for AI workloads, and that is what matters for enterprises and data center operators,” Mueller explained. “Networking especially is key to running AI platforms efficiently, and this is why Marvell is picking up XConn, so it can build more integrated clusters of XPUs. The key will be how quickly Marvell is able to integrate XConn’s switching technology with its own chips, and enterprises will expect to see the results later this year.”

“This combination creates a compelling switching platform for accelerated infrastructure, advancing Marvell’s connectivity strategy for next-generation AI and cloud data centers,” said Chief Executive Matt Murphy (pictured). “Combined with the pending acquisition of Celestial AI, we will be well positioned to deliver customers the performance, flexibility and architectural choice they need as AI systems grow in size and complexity.”

Ultimately, the acquisition will help Marvell to better compete with Broadcom. Though Marvell is known to develop XPUs for AWS, its rival is thought to be developing chips for at least four different hyperscalers, and it has sold many more of them over the past couple of years. Marvell is really playing catch up, and it needs to offer a more compelling product. XConn will hopefully enable it to do that.

The acquisition will also help to bolster Marvell’s finances. The chipmaker said it expects XConn to contribute about $100 million in revenue in fiscal 2028. The deal is expected to close early this year, subject to regulatory approval.

Photo: Marvell Technology