On March 31, Nvidia announced a US$2 billion investment in Marvell and plans to further integrate its NVLink Fusion technology with Marvell's XPU services for customer use. Although Nvidia revealed partnerships with several ASIC service providers around NVLink Fusion technology earlier in 2025, this direct investment signals a closer collaboration between Nvidia and Marvell. The move raises questions about how the two companies will expand their presence in the cloud AI market and whether ASIC customers will embrace this integrated solution.
During a lecture hosted by the Chinese National Association of Industry and Commerce (CNAIC), DIGITIMES Chairman Colley Hwang analyzed the East Asian industrial landscape. While headlines often focus on the chip wars between the US and China, Hwang shed light on a quieter, more structural divergence: the widening "resilience gap" between Taiwan and South Korea, as manifested through the lens of currency.
China's AI accelerator server market is gradually shifting away from using Nvidia chips to domestic GPU and AI chip makers, with Chinese companies capturing almost 41% of the market last year, according to an IDC report.
Amid AI-driven shifts, Arm launched its AGI CPU in March 2026 to address system-level optimization lacking in highly customized data center CPUs. Partnering with Meta and supported by OpenAI, Arm seeks to offer a standardized solution that enhances ecosystem efficiency without directly competing with clients.
Dixon Technologies is accelerating its push into display module manufacturing, backed by an INR11 billion (US$118.74 million) investment in a new facility in the Noida–Greater Noida region. The plant, approved under India's Electronics Component Manufacturing Scheme (ECMS), will serve as the company's first dedicated display module fabrication unit and marks a significant milestone in its backward integration strategy.
In late March 2026, a series of developments converged to reshape sentiment in the large model sector. Anthropic faced a major source code leak of Claude Code due to an engineer error. At nearly the same time, Chinese large model firm Z.ai released its first annual report since listing, with CEO Zhang Peng explicitly naming Anthropic as the company's benchmark; meanwhile, rising contender DeepSeek experienced three consecutive days of service disruptions from March 29 to 31, affecting both web and API access.
The architecture of host CPUs in custom AI servers is undergoing a significant shift, with proprietary Arm-based designs steadily displacing traditional x86 processors, according to Counterpoint Research's latest data on AI server compute ASIC shipments.
Microsoft has unveiled back-to-back investments across Southeast Asia, committing US$5.5 billion to Singapore and more than US$1 billion to Thailand in a coordinated push to expand artificial intelligence (AI) and cloud infrastructure in the region.
As the global AI race accelerates, Huawei's 2025 annual report leaves little ambiguity: AI now sits at the core of its strategy. The 147-page filing references "AI" 421 times, an unusually explicit signal of strategic depth. The company is pursuing a "foundation first, expansion later" model, pairing heavy R&D with infrastructure buildout to scale its AI position.
Spot prices for memory have plunged by 35%, yet contract prices continue to rise. Notebook supply chain sources say there are no signs of supply easing, with shortages expected to persist through the end of 2026. Rising component costs are also being passed on to end markets, prompting brand vendors to adopt a conservative stance, and a full-year decline in shipments now appears inevitable.
Rising demand for computing power from generative AI and the semiconductor industry's move toward high-performance computing are reshaping the smartphone supply chain, with global implications for device availability, pricing, and component sourcing as memory prices rise, capacity shifts toward AI workloads, and smartphone production forecasts point to substantial declines ahead.
More coverage
