Techlnsights: The semiconductor industry as a whole showed resilience in May and maintained the expected growth trend

The Zhitong Finance App learned that on June 12, Techlnsights released the May summary of the artificial intelligence market. May saw many turbulent economic situations around the world, with repeated tariffs and export controls. NVDA.US (NVDA.US), a leading company in the AI semiconductor industry, has reported an asset write-down of 4.5 billion US dollars due to export control restrictions. However, the semiconductor industry as a whole has shown resilience and maintained the expected growth trend. Furthermore, the global market for AI-powered processor chips and accelerators is expected to reach US$457 billion by 2030, with a compound annual growth rate (CAGR) of 23%.

2025 AI Data Center Chip Forecast Update

GPU accelerators are expected to lead the market, while ASIC accelerators are expected to get more attention, driven by cloud service providers such as Google and Amazon. In the short term, key challenges facing the industry include increasing memory capacity, improving connectivity protocols, and addressing growing power consumption issues.

Overall, the market is expected to continue to grow rapidly as the demand for performance of AI models continues to rise.

B200 secrets revealed

TechlInsights analyzed the floor plan of the GPU-chip GB102-A01 in the Nvidia GB100-886N-A1 package. The GB100-886N-A1 package has been removed from the NVIDIA HGX B200 accelerator cluster in the supermicro SYS-A22GA-NBRT GPU superserver. It includes eight NVIDIA Blackwell GB100 GPU packages and two Intel Xeon 66900 “Granite Rapids” central processing units (CPUs).

Artificial intelligence models compete for L4 autonomy

The development of autonomous driving systems involves the use of end-to-end (E2E) or composite artificial intelligence (CAIS) models. While end-to-end artificial intelligence is the most popular approach, CAIS promoted by MobilEye provides a more efficient and secure alternative. For example, the Toyota Guardian system uses CAIS to provide a safe housing and seamlessly integrates human-machine inputs. The CAIS architecture divides artificial intelligence tasks into three components: Primary (P), Guardian (G), and Fallback (F), which work together to ensure safe navigation. CAIS provides efficiency, scalability, and affordability, making it an attractive alternative to end-to-end artificial intelligence. However, adoption of CAIS has been limited as OEMs tend to develop their own end-to-end AI models. The future direction of the automotive industry is still uncertain, but CAIS has shown appeal and has been adopted by several car manufacturers, including Volkswagen and Polaris.

AI pushes the limits of packaging

The high performance computing (HPC) and artificial intelligence markets are driving advances in advanced packaging technology, driving pressure to adopt existing solutions such as 2.5D and 3D packaging on a large scale. At the core of these packages are high-density interconnect solutions, where solder-based micro-bumps and hybrid bonding techniques are currently used to connect different small chips and components.

New technologies such as ultra-low pitch microbumps and interconnects with spacing as low as 1 micron are emerging. Currently, vertical interconnects such as silicon through holes (tsv) are commonly used, but alternatives such as permeable insulator holes (tiv) are also being developed to reduce cost and density.

Semi-capital expenditure remains stable, driven by artificial intelligence

The global semiconductor supply industry has shown resilience in the face of macroeconomic turbulence, and demand driven by artificial intelligence remains a key driver of growth. Recent quarterly reports from major vendors such as TSMC, MediaTek, and Sun Moon Light show strong revenue growth due to strong demand for 3nm and 5nm process technologies for high-performance computing and artificial intelligence applications. Predicted sales of semi-equipment also showed a positive trend, indicating that the industry is expected to meet its goals despite the challenges of tariffs and macroeconomic pressure.

Artificial intelligence continues to drive power specifications

The growing power demand for AI workloads in data centers is pushing existing 54V power distribution systems to the limit, prompting companies such as Nvidia to explore high voltage direct current (HVDC) architectures as a solution. Two main strategies are emerging: ±400V HVDC and 800V HVDC, which operate directly at 800V to increase efficiency and reduce wiring requirements. Power semiconductor vendors are prepared to benefit from this transformation, but must prioritize core broadband gap (WBG) technology, develop scalable solutions, and utilize cross-market synergies to meet the needs of HVDC architectures. As voltage levels expand and new technologies emerge, enterprises will need to take a proactive and flexible approach to take advantage of the huge growth opportunities brought by AI-driven data center transformation.