On June 2, 2026, during the Computex 2026 keynote speech at the Taipei Computer Show, Nvidia CEO Huang Renxun publicly stated on site that Marvel is expected to grow into a trillion dollar semiconductor company in the future, setting a new record for strategic cooperation among semiconductor giants in recent years. After the statement was released, the stock price of Meiweier in the US stock market surged by 37% in a single day, with a daily market value increase of over 18 billion US dollars, making it the most explosive semiconductor stock at this year's Computex exhibition; In sharp contrast, Broadcom, a traditional leader in AI interconnect chips, has achieved a historic high in its financial report, but its stock price has fallen instead of rising. The capital market predicts that its exclusive monopoly on high-end AI interconnect chips will soon be broken, and the restructuring of the global AI computing power supporting chip industry chain has officially begun.

The complete hardware system of AI supercomputing servers is divided into three core components: GPU training and inference computing chips, CPU general-purpose processors, and high-speed interconnect switching chips. Nvidia firmly controls the absolute global dominance of GPUs, while Broadcom monopolizes the high-speed interconnection track within global AI servers through Ethernet switching chips and PCIe high-speed interconnect chips. In the procurement budget of each 10000 card intelligent computing center, the purchase amount of interconnect supporting chips accounts for 18% -22% of the total hardware cost, making it an indispensable essential core component for computing power clusters. In the past, Nvidia's GPU ecosystem was highly dependent on Broadcom interconnect chips, and the two sides were deeply bound for many years. Broadcom continued to raise the supply prices of high-end interconnect chips, while Nvidia's overall hardware costs continued to rise. The space for transferring costs to downstream cloud vendors gradually narrowed, and the cooperation conflicts between the two sides continued to accumulate.

This time, Nvidia officially partnered with Maiwell to release a complete joint optimization software and hardware adaptation solution. Nvidia's new generation Rubin GPU is natively compatible with Maiwell's latest generation 800G Ethernet switch chip and PCIe 6.0 interconnect controller chip. The operator layer and high-speed interface layer have completed bidirectional deep collaborative optimization. The end-to-end interconnection latency of the entire computing cluster has been reduced by 19%, and the concurrent data throughput limit has been increased by 23%. The comprehensive hardware performance surpasses Broadcom's supporting solution. At the same time, Nvidia simultaneously recommended alternative solutions for the Maiwell interconnect chip to global cloud service providers and intelligent computing center operators, breaking the exclusive supply lock of Broadcom. Downstream computing power purchasers now have dual supplier backup space for the first time, significantly increasing their bargaining power.

Maiweier has decades of technological accumulation in the traditional enterprise network chip field, with a stable global market share in Ethernet controllers and vehicle communication chips. Previously, it had not been able to enter the high-end AI computing interconnect chip track, and its core weakness was the lack of joint optimization capabilities with top GPU manufacturers at the bottom layer. After the full opening and adaptation of Nvidia's ecosystem, Maiweier has filled its biggest gap and has a strong certainty of explosive growth in orders. According to institutional calculations, orders for Maiweier AI interconnect chips are expected to increase by 220% month on month in the second half of 2026, gradually eroding Broadcom's high-end market share, and shifting the high-end interconnect chip market from exclusive monopoly to a duopoly competition pattern.

The chain reaction of the industrial chain continues to spread outward: the bidding rules for hardware procurement of self built intelligent computing centers by global public cloud giants have been synchronously modified, and interconnect chip suppliers must set up two or more qualified shortlisted manufacturers. Broadcom cannot continue to monopolize large orders exclusively; Overseas second tier cloud providers are accelerating the switch to the Maiweier solution to control the total cost of computing power cluster construction. Through this strategic cooperation, Nvidia is no longer constrained by a single upstream chip supplier. The cost of Rubin's new generation GPU solution has been effectively controlled, and the global price competitiveness of its products has been further strengthened, consolidating its position as the global leader in GPUs.

For domestic high-speed interconnect IC design manufacturers, after the formation of the duopoly competition pattern of overseas giants, Broadcom and Maiwell will moderately delegate mid-range interconnect chip technology authorization, and domestic manufacturers will have a window period for technology introduction and joint research and development; At the same time, domestic Xinchuang Intelligent Computing Centers and local state-owned newly built computing power clusters have mandatory procurement requirements for localization, and domestic high-speed Ethernet switching chips and PCIe controller chips have ushered in opportunities for mass import. Several domestic interconnect chip design companies have completed 800G switch chip engineering samples and entered the joint debugging and testing stage of top server manufacturers, accelerating commercialization through the window period of overseas giant reshuffle.

At the risk level, Broadcom holds a massive amount of stock bound orders, and the customer replacement cycle is generally 12-24 months, so the short-term market share will not be quickly lost; Maiweier's advanced interconnect chip production capacity is highly dependent on TSMC's advanced processes, leading to tight wafer casting capacity and uncertain order delivery cycles. However, in the medium to long term, the collapse of the exclusive monopoly pattern is irreversible, and AI supporting interconnect chips have entered a stage of full competition. The overall procurement cost of downstream computing hardware industry chains has declined, and the speed of large-scale construction of AI computing clusters is expected to further accelerate.

The essence of this giant's strategic cooperation is the reconstruction of the upstream and downstream game of the computing power industry chain. NVIDIA consolidates its leading position in GPU, Maiweier achieves breakthroughs in the high-end track, and downstream cloud vendors gain bargaining power, forming a win-win situation for all parties; Domestic chip manufacturers, taking advantage of the global reshuffle window period, have entered the mid-range interconnect chip market in a staggered manner, and the domestic computing power chain has gained external favorable opportunities for independent and controllable processes.