At Computex in Taipei, Marvell Technology chief executive Matt Murphy and Nvidia chief Jensen Huang said the next phase of artificial intelligence growth will hinge less on raw processing power and more on the high-speed networks that link chips together. The remarks, including Huang calling Marvell a potential trillion-dollar company, helped send Marvell’s share price up more than 16 percent in extended trading and over 18 percent in early trading on June 2.
Nvidia flags Marvell as key AI infrastructure player
Huang’s endorsement came alongside Nvidia’s recent 2 billion-dollar strategic investment in Marvell, underscoring Nvidia’s view that connectivity is now as critical as compute in large-scale AI systems.
Marvell’s profile has already shifted sharply toward this theme. Its data center business has grown from less than 10 percent of revenue a decade ago to more than 75 percent last quarter, with annual growth near 40 percent. Based on company guidance, market projections point to Marvell’s revenue reaching about 16.4 billion dollars in the next fiscal year.
Traders responded aggressively to the new framing of Marvell’s role in the AI build-out, adding tens of billions of dollars to the company’s market value in a single day. The reaction reflects growing recognition that the bottleneck in AI performance is moving from the processor itself to the network fabric connecting tens of thousands of processors.
From compute to connectivity as AI bottleneck
Both Murphy and Huang argued that advances in processing and memory are now outpacing the ability of existing connections to move data. Murphy said that as computational and memory constraints have eased, connection capacity has become the next limiting factor for large-scale data infrastructure.
Industry analysts following the event said major cloud providers are reassessing their network architectures to cope with the scale and complexity of distributed AI workloads. The focus is shifting toward the hardware and technologies that can move data between chips at ever higher speeds and over greater distances, while keeping power use and physical footprint in check.
Copper near limits as data rates climb
Huang described a dual-medium approach for data center interconnects, with copper cables used “where possible” and optical fiber “where necessary.” He said that as data rates rise toward 400 gigabits per second, copper interconnects will struggle to span even a single server rack, pushing the sector toward broad adoption of optical links within the next five to ten years.
Murphy added that every doubling of bandwidth roughly halves transmission distance, meaning today’s 200 Gbps copper systems already operate near physical limits of about 2.5 meters. To move beyond those limits, Marvell is directing substantial investment into co-packaged optical, or CPO, technology, which integrates optical fiber directly into the switch package. The approach is designed to cut energy consumption and reduce space demands in dense AI data centers.
New high-capacity switches and CPO debut
At Computex, Marvell unveiled a new 100-terabit Ethernet switch targeted at AI data centers. The company also demonstrated a 51.2-terabit switch based on CPO that eliminates traditional copper traces on the circuit board.
Murphy said the longer-term goal is to remove rigid hardware boundaries inside data centers so that compute and memory resources can be dynamically arranged around the needs of specific AI models. In this architecture, the network fabric becomes the key determinant of how efficiently those resources are used.
NV Link Fusion aims at flexible cloud architectures
The two companies also introduced NV Link Fusion, a platform that combines Nvidia’s system architecture with Marvell’s networking products to support heterogeneous data center environments.
Huang said the initiative allows cloud operators to use parts of Nvidia’s platform while still designing their own processors, enabling more modular and customized facilities. Murphy framed Marvell’s position as that of a neutral partner, linking a wide range of compute and storage suppliers across the industry.
The collaboration signals a broader pivot in AI infrastructure spending, away from an exclusive focus on chip performance and toward the networks that bind those chips into coherent systems.
Data center capex surges on AI build-out
This strategic shift is being supported by rapid growth in real-world spending. Global data center capital expenditures jumped 51 percent to 455 billion dollars in 2024 and are projected to climb by more than 30 percent in 2025, driven largely by AI infrastructure.
The four largest cloud service providers increased their capital spending by 62 percent to more than 251 billion dollars in 2024, highlighting the scale of the physical build-out needed to support new AI services and large language models.
For market participants, the message coming out of Taipei is that the AI opportunity now extends beyond the makers of graphics processing units to the companies designing the high-speed data pathways that allow those GPUs to work together.
Interconnect and optics markets set for rapid growth
The broader data center interconnect market, which underpins communication between servers and accelerators, is projected to grow from just over 10 billion dollars in 2024 to more than 20 billion dollars by 2030. Some forecasts see the market approaching 26 billion dollars in that timeframe, driven by unrelenting demand for high-bandwidth links to handle AI and machine learning workloads.
Within that, co-packaged optics stands out as a high-growth segment. The CPO market is expected to expand from about 165 million dollars in 2026 to more than 764 million dollars by 2031, implying a compound annual growth rate of nearly 36 percent. Other studies point to a similar trajectory, with CPO revenue projected to grow at a CAGR of more than 35 percent through 2035.
For traders tracking these developments, several signals are becoming central:
- Capital expenditure reports from major cloud operators, which provide early evidence of whether the shift toward networking infrastructure is translating into large-scale purchase orders and deployments.
Murphy and Huang’s joint message at Computex suggests that those numbers will increasingly reflect spending on the foundational technologies that move data, not just on the chips that process it.
Explore how AI transforms finance next—discover our in‑depth guide on AI complementing blockchain in modern digital infrastructure.
Disclaimer: The content on this page is provided for general informational purposes only and does not represent the views or financial advice of Toobit. We make no guarantees regarding the accuracy or completeness of this information and shall not be held liable for any errors, omissions, or outcomes resulting from its use. Investing in digital assets involves risk; users should independently evaluate their financial situation and the risks involved. For further details, please consult our Terms of Service and Risk Disclosure.
