- About Us
-
Research Report
Semiconductors
LED
Consumer Electronics
Emerging Technologies
- Selected Topics
- Membership
- Price Trends
- Press Center
- News
- Events
- AI Agent
This issue highlights server ODMs, CPUs, GPUs, and key changes in the thermal supply chain. AI servers are the focus, CSPs keep ordering, ODMs prep for new platforms, Meta and Google push in-house CPUs, with rising cooling needs. Geopolitical risks still impact the outlook.
In late June 2025, Google DeepMind introduced a new AI model named Gemini Robotics On-Device. This model further integrates the multimodal capabilities of Gemini 2.5 directly into terminal devices. Its design focuses on edge or terminal inference, enabling humanoid robots to perform complex tasks by integrating language, vision, and sensory inputs, even in environments without communication or cloud connectivity. This approach marks a significant departure from previous AI systems that relied heavily on cloud computing for processing. Consequently, it enhances the practicality of robots in time-sensitive or network-constrained scenarios such as healthcare and disaster rescue.
In 2025, global AI chips focus on high-end HBM memory; NVIDIA’s new Blackwell platform drives growth, amid geopolitical limits and steady AI server demand, with rapid HBM technology evolution toward HBM4 in 2026.
North American CSPs & OEMs drive AI market growth; new Blackwell platform shipments will expand. China's market faces variables due to geopolitics affecting AI solution supply. Overall AI server shipments are expected to maintain double-digit growth.
On May 25, Figure AI CEO Brett Adcock announced that the company’s latest humanoid robot, Figure 03, has achieved bipedal walking, claiming it integrates the most advanced hardware system seen to date. The previous model, Figure 02, is currently undergoing testing at a BMW factory, recently completing 20 consecutive hours of operation. As early as March 2025, Figure revealed that the design of the third-generation model had been finalized, with a focus on cost control and mass production scalability. The company plans to use injection molding and diecasting technologies. As of now, a full video of Figure 03 walking has not been released.
Amazon is rapidly expanding its self-developed AI ASIC servers to bolster AWS's cloud AI training competitiveness. The new generation Trainium chips feature diversified designs catering to various training needs. On the NVIDIA platform, Delta leads as the main power supplier, advancing high-power AI server development.
COMPUTEX 2025 was convened for four days in mid-May, and invited 1,400 suppliers to showcase how AI is currently actuating industrial upgrades. The grand event focused on “AI NEXT” as this year’s topic as a continuance from “Connecting AI” of 2024, and carried on with the concentration on products related to AI, where various applications that came to fruition on-site also symbolize how the industry is marching towards a new era of AI empowerment.
As 2025 progresses, CSPs continue to build out AI infrastructure, with Meta driving growth in the demand for the B300 and GB300...
Global market intelligence firm TrendForce reports that NVIDIA achieved a total revenue of US$44.1 billion for FY1Q26...
According to TrendForce, NVIDIA's data center revenue grew by 73% year-over-year in FY1Q26, driven by the gradual ramp-up of the new Blackwell GPU platform. After 2Q25, NVIDIA is expected to significantly increase production of the GB200 rack and launch the MGX RTX PRO to meet edge AI demands. The GB300 NVL72 design was reverted to the Bianca architecture for a smoother product launch, causing short-term pressures on the liquid cooling supply chain. In the future, NVIDIA's data center business will continue to expand, focusing on Tier 2 data centers and sovereign cloud projects, which will become key growth areas.
Research Areas:
Industry News
