The next "trillion-dollar technology king" is it? Huang Renxun personally promoted Marvell (MRVL.US) "standing in the light" to strengthen the AI interconnection theme again.

Focus on the customization of AI chips (AI ASIC) and optical interconnect technology in large AI data centers, as well as Marvell Technology, Inc. (MRVL.US), one of the largest partners in the AWS Trainium series AI ASIC by Amazon.com, Inc. The stock price of Marvell Technology, Inc. surged nearly 30% in pre-market trading on Tuesday as NVIDIA Corporation CEO Huang Renxun and Marvell Technology, Inc. stood side by side at the 2026 Computex in Taipei, China. The core logic undoubtedly lies in Huang Renxun's statement that Marvell Technology, Inc. has the potential to become the next trillion-dollar market value listed company, and he emphasized its crucial role in large AI data center interconnect infrastructure systems. This clear endorsement from the leading player in the AI computing power field has greatly enhanced market confidence in Marvell's long-term growth path. Marvell Technology, Inc. CEO Matt Murphy's keynote speech at the 2026 Computex conference, titled "The Future of AI Scaling Depends on Connectivity," emphasized that as the demand for AI computing resources and large-scale AI infrastructure clusters grows exponentially, performance improvements have gradually outpaced data movement and high-speed DCI interconnectivity within and between data centers. Therefore, the next stage of AI computing power expansion will be dominated by "connectivity rather than single-node computing power." Marvell Technology, Inc.'s optical interconnect, ultra-high-speed DSP, and silicon photonics technology chain are at the core of this new infrastructure paradigm. In a subsequent conversation between NVIDIA Corporation's founder and CEO Huang Renxun and Marvell Technology, Inc. CEO, they deepened the understanding of the trend towards upgrading the AI computing power industry to focus on "connectivity." This conversation not only exchanged technical viewpoints but also declared the highest level of dynamics and trends in the global AI computing infrastructure industry chain. In terms of AI revenue prospects, Anthropic, which launched a series of AI intelligent entities in February that beat global software stocks, recently released a compelling profit trajectory. This strong revenue growth and better-than-expected profit path will further accelerate the construction of "AI computing power super empires" by technology giants. Anthropic expects its Q2 revenue to double and achieve operating profit for the first time, highlighting a shift from a narrative of burning money to a narrative of cash flow circulation in the AI industry. The most striking point is Anthropic's single-month ARR (Annual Recurring Revenue) surged to $11 billion, equivalent to the combined value of Palantir, Snowflake, and DataBricks over a decade - a unprecedented miracle in capitalist history. Anthropic also expects second-quarter revenue to jump from $4.8 billion in the first quarter to $10.9 billion and achieve approximately $559 million in operating profit, indicating that advanced AI applications are not just consuming computing power but are transforming enterprise programming, intelligent entity workflows, network security, and data analysis into high-value Token income. The explosion in demand for Claude and AI tools drove Anthropic close to its first profitable quarter, with its cost of computing power per $1 of revenue dropping from about 71 cents in the first quarter to about 56 cents in the second quarter, showing that scale effects and reasoning efficiency are improving the AI application economic model. This is why Wall Street financial giant Morgan Stanley has declared that the AI computing power arms race is entering a phase of systemic expansion, with the institution significantly increasing its capital expenditure expectations for major US tech giants in 2026 from $433 billion a year ago to $805 billion. Capital expenditures in 2027 are expected to reach $1.1 trillion, exceeding the previous forecast of $950 billion, and by 2028, close to $3 trillion in AI-related infrastructure investments will flow through the global economy, with over 80% of spending still ahead. Morgan Stanley's latest forecasts highlight the global popularity of AI entities, with the main investment theme in AI computing power shifting from the "competition around AI GPUs" to the "all-stack computing power system driven by AI entities." The supply chain bottleneck at the AI computing power infrastructure level has expanded from "large-scale purchases of GPUs/ASICs" to simultaneously addressing a complete set of AI data center delivery processes including data center power equipment, liquid cooling, data center CPUs, DRAM/NAND/HBM, optical communication/optical interconnect, high-performance Ethernet network infrastructure/data center DCI high-speed interconnectivity, transformers, gas turbines, etc., especially storage bases and optical interconnection systems, which are currently the biggest bottlenecks in data center computing systems. Huang Renxun's prediction of the trend in the development of AI computing power is increasingly bullish on the underlying logic of Marvell. He points out that the commercialization of AI has entered the era of "Useful AI," meaning that Token production can not only drive research and development experiments but also bring real profits. With AI tasks being split and run in a distributed manner, the demand for "massive connectivity" has exceeded the scope of single-node upgrades, making system-level expansion the center of design. Therefore, after the standardization of distributed AI architecture, the connectivity layer will become one of the fastest-growing and highest-value subfields in the market, which is the logical starting point for the long-term value space of Marvell. Huang Renxun personally confirmed that "high-speed interconnect is the future"! He made a bold bet that Marvell Technology, Inc. is the next "trillion-dollar tech giant." Huang Renxun emphasized that the $20 billion strategic investment in Marvell Technology, Inc. and their cooperation are not unidirectional dependence but ecological complementarity. Under the NVLink Fusion framework, Marvell can provide customized optical interconnect and ASIC solutions for hyperscalers, while NVIDIA Corporation ensures that data center infrastructure can balance between universality and high performance through its widespread deployment of GPUs and system-level platforms. This synergy of "hardware + data center optical interconnect + ecosystem" gives Marvell longer-term growth potential in future AI factories, AI-RAN, and edge AI layouts, supporting its position as a potential trillion-dollar AI computing power infrastructure core enterprise. Huang Renxun explicitly stated that in the era of "agentic AI" sweeping global enterprises and large-scale distributed training/inferencing configurations, performance in computation and memory has gradually exceeded. The next core limitation is no longer single computing power or memory capacity, but the efficient movement of data between nodes. As AI training/inferencing workloads are split across hundreds of thousands, tens of thousands, or even more collaborative processing units, achieving ultra-high-speed, low-latency, and high-efficiency internal data center interconnect or data center-to-data-center DCI interconnectivity is essential to ensure that the computing power and memory advantages can be translated into overall system performance improvements. Huang Renxun has emphasized multiple times in his speeches that Marvell Technology, Inc.'s solutions at the architectural level of data center optical interconnects are in a critical position. In this context, Huang Renxun's positive outlook on Marvell is based on the first layer of logic. Traditional data center networks are dominated by copper cable infrastructure, and as the workload of AI training/inferencing expands to cross-rack and cross-data center levels, the demand for optical interconnects within data centers and across data centers, and even the cutting-edge optical technology - silicon photonics technology, is expected to experience strong growth. Huang Renxun emphasized in the conversation, "Use copper where you can, use light when you must," indicating that at the point of overcoming traditional physical bandwidth limitations, Marvell is one of the few suppliers that can provide high-speed optical interconnects, optical DSP, and potential custom silicon photonics solutions simultaneously. Huang Renxun's prediction of Marvell reaching trillion-dollar market value is not just a slogan but is based on a dual-layer ecological synergy logic in the AI computing power industry chain. NVIDIA Corporation has not only established technical compatibility with Marvell in the NVLink Fusion platform but has also expanded its AI infrastructure strategy from simple GPU to deep collaboration in rack-level, system-level, and cross-vendor custom ASIC. NVLink Fusion allows customers to build scalable, compatible, and heterogeneous AI computing systems combining NVIDIA Corporation's exclusive GPU/DPU and Marvell's custom XPUs and optical network interconnect technologies under a unified architecture. Huang Renxun believes that this cooperation allows Marvell to occupy an irreplaceable computing power pivot in future AI data center construction, enabling it to transcend the traditional chip valuation framework. This also highlights the strategic investment of $2 billion from NVIDIA Corporation into Marvell under Huang Renxun's leadership, which not only serves as a financial vote of confidence but also signifies ecosystem lock-in. By deeply integrating capital and technology, NVIDIA Corporation can ensure the optimization of its NVLink interconnect standard, AI factory architecture, and Marvell interconnect solutions, reducing the risk of ecosystem fragmentation caused by customers choosing independent custom paths. This "soft and hard compatibility" cooperation framework makes customers more inclined to deploy unified deployments in the NVLink/Marvell network combination when building large-scale AI computing infrastructure, thereby securing future market share. As AI data centers increasingly trust the power of light, the "optical interconnect + silicon photonics" engine ignites a new bull market for Marvell This chip giant's stock price has risen by as much as 160% this year, significantly outperforming the S&P 500 index and the Nasdaq 100 index, which is known as the "technology stock barometer." The core engine of Marvell Technology, Inc.'s revaluation and fundamental expansion is the resonance of the two hardcore logics of "AI ASIC/XPU + optical technology," rather than being driven solely by the theme of chip demand. According to recent statements from Wall Street analysts, the market no longer views Marvell Technology, Inc. solely as a generic data center AI chip company but rather as a company that is positioning itself to both cater to the wave of self-developed XPU/ASIC by large cloud computing giants and take the lead in large-scale upgrades based on optical interconnectivity and silicon photonics development in the explosive expansion of AI computing power clusters. Broadcom Inc. and its largest competitor, Marvell Technology, Inc., are primarily focused on using their absolute advantages in high-speed interconnect and chip IP areas to work together with cloud computing giants such as Amazon.com, Inc., Alphabet Inc. Class C, and Microsoft Corporation to create AI ASIC computing clusters customized to their specific needs in AI data centers. This ASIC business has become a very important business for both companies; for example, the TPU AI computing cluster created by Broadcom Inc. in collaboration with Alphabet Inc. Class C is a typical route of AI ASIC technology. Optical interconnect is the second major core pillar that analysts on Wall Street believe will drive the prospects and fundamental expansion logic of Marvell Technology, Inc., and is also a key factor in NVIDIA Corporation's $20 billion investment in the company. When the connection distance exceeds approximately 10 meters, copper cable interconnects are no longer able to meet the bandwidth and distance requirements of large AI data centers at high speeds; at this point, it is necessary to switch to optical interconnect systems, and the core hardware in optical devices is a range of optical DSP and optical interconnect products that Marvell excels in. Marvell has introduced high-bandwidth network chips such as the Teralynx T100 102.4Tbps switch chip designed for AI data centers to meet the interconnect requirements of AI computing clusters across nodes, racks, and even data centers that traditional copper cables cannot match in terms of bandwidth and latency requirements. In addition, the company's technical accumulation and product portfolio in the areas of optical interconnect and silicon photonics, combined with acquisitions (such as Celestial AI and XConn), further enhance its competitiveness in the high-end optical interconnect and custom ASIC (XPU, etc.) markets, which are essential infrastructure components in the era of scalable AI inferencing and training. Optical interconnect focuses on high-speed links between servers, switches, racks, and data centers. Looking ahead, Marvell's leading silicon photonics technology is also a key logic for Wall Street's positive outlook on the company's stock price. Silicon photonics is not a substitute concept parallel to optical interconnect but is more like a critical technology that pushes light from outside the cabinet, between racks, further into racks, systems, and even packaging. More accurately, silicon photonics technology leans towards advancing optical transmission to the chip/package/I/O side, which is why Marvell chose to acquire Celestial at a high cost. Marvell's technical accumulation and product portfolio in high-speed optical DSP, optical interconnect modules, and ASIC/XPU custom fields are at the most scarce and highly competitive parts of the AI computing power infrastructure system. In the era of increasing AI inference and training computing power, these components are more difficult to replace than single computing power chips, making Marvell not just a "AI chip stock" but the architectural core of optical interconnect/data center DCI interconnectivity and a wider data movement infrastructure, providing longer-term and more stable revenue and profit expansion space.