0

Chipmaker outlines strategy for electrical and optical AI infrastructure to capture growing market share

In sum – what we know:

• A strategic pivot – Marvell is moving beyond its roots in chip manufacturing to position itself as a comprehensive “end-to-end” connectivity provider for critical AI infrastructure.
• The Golden Cable initiative – Launched on December 9, this program is designed to accelerate the Active Electrical Cable (AEC) ecosystem, helping hyperscalers deploy 1.6T networking faster.
• Celestial AI acquisition – A massive $3.25 billion acquisition of the pioneer behind Photonic Fabric technology, signalling a long-term commitment to optical interconnects as the successor to copper.

Marvell Technology is going through some changes. Shifting focus from pure-play chip manufacturing to a broader, “end-to-end” connectivity strategy, the company used its recent analyst day to outline a vision that directly addresses one of the more pressing infrastructure challenges of generative AI: moving data.

As AI workloads become increasingly massive, raw compute power is no longer the sole, or even the main, limiting factor. Because these workloads are partitioned across processors, the speed at which data travels between them has become a new bottleneck. While manufacturers have spent years optimizing the processor, Marvell is betting its future on optimizing the pathways between them.

This commitment was on display at OCP 2025, where Marvell demonstrated significant performance gains using CXL compute accelerators. This emphasis on the Compute Express Link (CXL) standard serves as a reinforcement of their intent to manage the full spectrum of AI infrastructure needs. This, of course, comes alongside growing concerns around increasing competition in the chip manufacturing space — and it comes shortly after Marvell’s deal to Acquire Celestial AI.

Golden Cable

To address the immediate physical layer challenges of AI scaling, Marvell has introduced the “Golden Cable” initiative. The program focuses on Active Electrical Cables (AEC) specifically designed to accelerate 1.6T networking deployments in hyperscaler environments.

The initiative is a direct response to the “time-to-market” pressure felt by major tech companies. By providing partners with validated reference designs, firmware, and calibration data, Marvell allows manufacturers to bypass lengthy R&D cycles. This preserves design flexibility while ensuring the cables meet the rigorous performance standards of modern data centers.

The financial stakes here are significant. The AEC market is projected to more than double, growing from $644 million in 2025 to an estimated $1.4 billion by 2029. The Golden Cable program is Marvell’s bid to capture the lion’s share of that expansion by offering a solution that lowers both cost and power consumption compared to competing approaches.

Industry adoption is already underway. Foxconn Interconnect Technology utilized the program’s validated architectures to complete its first design in just two months. Similarly, Luxshare Technology has combined Golden Cable with its Optamax high-performance bulk cable to accelerate AEC deployment, proving the model works in the real world.

Celestial AI

While AECs solve near-term issues, Marvell recognizes that copper has physical limitations. On December 2, the company signed a definitive agreement to acquire Celestial AI for $3.25 billion, a clear signal that they believe the future of AI scaling is optical.

The centerpiece of this acquisition is Celestial’s Photonic Fabric technology. As bandwidth requirements skyrocket in massive AI clusters, traditional copper connections struggle with reach and signal integrity. Celestial AI solves this by integrating electrical and optical components into compact form factors, enabling light-speed connectivity at the package, system, and rack levels.

According to Marvell, Celestial has already engaged with multiple hyperscalers, with plans to co-package their tech with custom XPUs and scale-up switches. This positions Marvell to enable the industry’s first large-scale commercial deployment of optical interconnects.

Matt Murphy, Chairman and CEO of Marvell, described the acquisition as “a transformative step in Marvell’s evolution and expands our leadership in AI connectivity, as scale-up becomes the next frontier in AI infrastructure.”

Competition ahead

Despite the strategic clarity, the road ahead is crowded. Marvell faces stiff competition from other semiconductor giants and infrastructure players who are also chasing the lucrative AI connectivity market.

Specific risks remain regarding key customers. Analysts have flagged concerns that Amazon could shift technology development to other partners. While Amazon Trainium2 shipments are expected to bolster Marvell’s near-term results, there is speculation that the latest generations — Trainium3 and Trainium4 — may have been awarded to chip design firm Alchip instead. However, Marvell still has a robust portfolio of higher-profile clients, including Microsoft and Google. While competition is heating up, these existing customers continue to ramp up spending with no signs of slowing down.

Sandeep Bharathi, President of Marvell’s Data Center Group, emphasized that the speed of transformation in AI infrastructure is unprecedented, requiring scale-up fabrics that can deliver bandwidth, power efficiency, and reach simultaneously.

Ultimately, 2026 will be the proving ground. Marvell has laid out the strategy and acquired the necessary technology; now, their execution with hyperscaler customers will determine whether this pivot successfully establishes them as the dominant architect of the AI data center.