https://www.leiphone.com/category/chips
DeepSignal tracks AI updates from 雷峰网芯片, filtering research and product signals into plain-English summaries, signal scores and source-linked article pages.
Current topics: China Hardware, AI Startup, Inference, GPU, Agent · Companies: Copilot
High-signal updates

The price of the B300 AI chip has surged over threefold in six months, now exceeding 12 million CNY, while major companies are avoiding intermediary computing power stations due to security and performance concerns. Additionally, significant management changes at an AI Infra company highlight instability in the sector, as the industry shifts from a focus on hardware to operational efficiency.
The tripling price of the B300 AI chip signals a growing demand for high-performance hardware, impacting builders and PMs who must consider cost and availability in their projects. Additionally, the shift away from intermediary computing power stations indicates a need for more direct and efficient infrastructure solutions, which could influence investment strategies in the AI sector.

Qualcomm's Snapdragon 8 Gen 2 chip is now the official processor for five major mobile esports leagues, showcasing its ability to deliver stable performance under pressure. With benchmarks like 144fps in KPL and low latency, Snapdragon is transforming gaming experiences, while AI integration is poised to redefine game development and interaction.
Qualcomm's Snapdragon 8 Gen 2 chip being adopted by five major mobile esports leagues signals a significant leap in mobile gaming performance and stability. For builders and PMs, this integration of AI in gaming can lead to innovative game mechanics and user interactions, while investors should note the potential for growth in the mobile esports market driven by advanced technology.
Zeng Yi, founder of Sunzhan, emphasizes that AI is reshaping chip architecture, enabling RISC-V's rise as companies like Qualcomm and Google adopt customizable AI processors. He believes RISC-V's flexibility and open-source nature will drive its adoption in the semiconductor industry, particularly in AI applications.
The rise of RISC-V as a customizable AI processor, as highlighted by Zeng Yi, indicates a significant shift in chip design that lowers barriers for innovation in the semiconductor industry. Builders and PMs should consider leveraging RISC-V for tailored solutions, while investors should recognize the potential for growth in companies adopting this flexible architecture.

Qualcomm emphasizes that personal AI requires not just computational power but also context awareness, memory, and low-power sensing for effective user interaction. The shift from AI for devices to AI for users necessitates a distributed system that understands individual user needs across multiple devices.
Qualcomm's emphasis on the need for context awareness, memory, and low-power sensing in personal AI indicates a shift towards more user-centric AI solutions. Builders and PMs should focus on developing distributed systems that can adapt to individual user needs, while investors may find opportunities in startups that prioritize these capabilities in their AI products.

Allwinner Technology's V881 chip achieves 2GB-level image quality using only 128MB of memory, redefining video processing capabilities. This innovation is crucial for AI applications, enabling real-time understanding and decision-making in various sectors, from security to sports analytics.
Allwinner Technology's V881 chip demonstrates a breakthrough in video processing by achieving 2GB-level image quality with only 128MB of memory. This advancement is significant for builders and PMs as it allows for more efficient AI applications in real-time analysis, potentially reducing costs and enhancing performance in industries like security and sports analytics.

At WAIC 2026, AI Infra executives emphasized a shift from scaling GPU resources to optimizing token production efficiency, with predictions that 99% of tokens will be consumed by agents. The rise of AI factories and token factories marks a new era in AI infrastructure, focusing on stable, low-cost token generation for diverse applications.
The shift from scaling GPU resources to optimizing token production efficiency, as highlighted by AI Infra executives at WAIC 2026, signals a critical transition in AI infrastructure. Builders and PMs should focus on developing applications that leverage low-cost token generation, while investors may find opportunities in AI factories and token factories that cater to this emerging demand.

At WAIC 2026, AI chip executives emphasized the shift to 'supernodes' for AI computing, moving from single-chip performance to system-level collaboration. The focus is on inference workloads, with companies like Huawei showcasing advanced supernodes like the Ascend 950, while the industry grapples with challenges such as low-latency requirements and the need for a cohesive ecosystem.
The emphasis on 'supernodes' for AI computing, as highlighted by Huawei's Ascend 950, indicates a shift towards system-level collaboration in handling inference workloads. Builders and PMs should consider the implications for architecture design and integration, while investors should note the potential for scalable solutions in the evolving AI ecosystem.
Yuan Chuan Wei, founded by Huawei veteran Yang Bin, has secured hundreds of millions in Pre-A funding to develop LPU+ chips aimed at optimizing AI inference. The company emphasizes creating high-value solutions over cost-saving, targeting the emerging Agentic AI market with a focus on low latency and high stability.
Yuan Chuan Wei's successful Pre-A funding round to develop LPU+ chips highlights a significant investment in AI inference optimization, which is crucial for builders and PMs focusing on high-performance applications in the Agentic AI market. This development signals a shift towards prioritizing stability and low latency in AI solutions, attracting investor interest in emerging technologies.
Despite a surge in Token usage, with daily calls exceeding 180 trillion, profits remain concentrated with Nvidia, which reported $81.6 billion in revenue and a 74.9% gross margin. In contrast, OpenAI faced a staggering $9.3 billion operating loss, highlighting the structural challenges in the AI industry where costs are dominated by upstream hardware.
The surge in Token usage, reaching over 180 trillion daily calls, highlights the growing demand for AI services, but the significant operating loss reported by OpenAI underscores the financial challenges in the industry. Builders and PMs should consider the reliance on hardware providers like Nvidia, while investors need to evaluate the sustainability of AI business models amid high operational costs.

At WAIC 2026, Arm China emphasizes that is not a scaled-down version of cloud AI but a new computing market defined by power consumption, real-time capabilities, and reliability. Their Star 300 AIoT platform aims to enable AI capabilities in resource-constrained environments, while the Zhouyi X3-Pro addresses complex inference needs across diverse edge scenarios.
Arm China's introduction of the Star 300 AIoT platform and Zhouyi X3-Pro highlights a shift towards edge AI, emphasizing its unique requirements for power efficiency and real-time processing. This development signals to builders and PMs the need to adapt their AI solutions for edge environments, while investors should recognize the potential growth in this emerging market.

The Shuguang 8000, China's first 100,000-card AI supercluster, debuted at WAIC, achieving over 150,000 daily tasks in its first week. This system, utilizing 'super-intelligent fusion' technology, is poised to meet 5-10% of the nation's token demand, but faces challenges in performance efficiency and reliability as it scales.
The debut of China's Shuguang 8000 supercluster, capable of handling over 150,000 daily tasks, signals a significant advancement in AI infrastructure that builders and PMs can leverage for scalable applications. However, its challenges in performance efficiency and reliability highlight the need for ongoing innovation and investment in robust AI systems.

Moore Threads' CEO Zhang Jianzhong introduced the concept of 'Three AI Factories' at WAIC 2026, focusing on model training, token production, and agent manufacturing, emphasizing the need for a unified infrastructure to support continuous AI development. The approach aims to enhance AI capabilities and reduce costs, with significant implications for the industry as it transitions from model training to practical applications.
Moore Threads' introduction of the 'Three AI Factories' concept highlights a unified infrastructure for model training, token production, and agent manufacturing, which can streamline AI development and reduce costs. This is significant for builders and PMs as it suggests a shift towards more efficient AI deployment, while investors may see potential for scalable solutions in the evolving AI landscape.

GMI Cloud showcased its AI-native cloud solutions, including the Inference Engine and Agentbox, at WAIC 2026, emphasizing high-performance GPU services and innovative AI infrastructure. Their collaboration with DDN aims to enhance AI deployment efficiency, addressing global market needs with a focus on scalable, secure solutions for enterprises.
GMI Cloud's launch of AI-native cloud solutions like the Inference Engine and Agentbox at WAIC 2026 signals a significant advancement in scalable AI infrastructure. This development offers builders and PMs enhanced tools for deploying AI applications efficiently, while investors should note the growing demand for high-performance GPU services in the enterprise sector.

At WAIC 2026, Enflame Technology showcased its advancements in AI computing infrastructure, including the ESL64-O and ESL64-C supernodes, which support large-scale AI applications. The company emphasizes the importance of high-performance computing for AI deployment, aiming to enhance the domestic AI ecosystem and address industry challenges with a focus on Token economies.
Enflame Technology's introduction of the ESL64-O and ESL64-C supernodes at WAIC 2026 highlights significant advancements in AI computing infrastructure, which are crucial for scaling large AI applications. This development signals an opportunity for builders and PMs to leverage enhanced computational power, while investors may see potential in supporting a growing domestic AI ecosystem focused on Token economies.

At the WAIC forum, Qualcomm's Dr. Xu Haoliang emphasized the transformative potential of , highlighting a tenfold increase in token processing needs for devices. He introduced new architectures designed for continuous online workloads and collaboration with partners like Mianbi to enhance edge model capabilities.
Qualcomm's introduction of new architectures for edge AI, as highlighted by Dr. Xu Haoliang, signals a significant shift towards meeting the growing demand for real-time processing in devices. Builders and PMs should consider how these innovations can enhance product capabilities, while investors may see opportunities in companies leveraging this technology for competitive advantage.

此芯科技 launched its AGX Agentic Compute strategy and Agentic OS, aiming to redefine computing for intelligent agents by enhancing CPU roles in task execution and resource management. The new OS facilitates automation and security, addressing challenges like data safety and model fragmentation in AI applications.
此芯科技's launch of the AGX Agentic Compute strategy and Agentic OS represents a significant shift in how computing resources are managed for AI applications, enhancing automation and security. This development is crucial for builders and PMs as it addresses data safety and model fragmentation, potentially streamlining the deployment of intelligent agents and attracting investor interest in more robust AI infrastructures.

Yuntian Lifei unveiled its AI inference chip roadmap at WAIC 2026, introducing three specialized chips—DeepVerse100P, DeepVerse100D, and DeepVerse100L—aimed at optimizing different stages of inference to reduce token generation costs to one cent per hundred billion tokens. The chips will enhance system efficiency in large-scale heterogeneous clusters by addressing specific computational needs.
Yuntian Lifei's introduction of three specialized AI inference chips aims to drastically reduce token generation costs to one cent per hundred billion tokens. This development signals a significant opportunity for builders and PMs to optimize AI applications and for investors to capitalize on cost-efficient AI infrastructure that can enhance performance in large-scale deployments.

At WAIC 2026, AI chip supplier Aixin Yuanzhi unveiled its new 'Yuanxi' AI inference series, featuring over 1000 TOPS performance and advanced edge computing capabilities. The products aim to enhance AI deployment across various sectors, including industrial automation and smart education, while addressing cloud dependency issues and reducing operational costs.
Aixin Yuanzhi's launch of the 'Yuanxi' AI inference series with over 1000 TOPS performance signifies a shift towards more powerful edge computing solutions, which can reduce cloud dependency and operational costs. This development is crucial for builders and PMs looking to implement AI across various sectors, while investors should note the potential market expansion in industrial automation and smart education.
Oriental Computing has launched the DF1000, China's first software-defined 3D AI chip, focusing on inference efficiency with 520T BF16 performance and 6.4TB/s memory bandwidth. This marks a shift from traditional GPU reliance, emphasizing a complete domestic AI ecosystem.
The launch of the DF1000, China's first software-defined 3D AI chip, signifies a strategic shift away from GPU dependency, highlighting advancements in inference efficiency with impressive performance metrics. This development opens new avenues for builders and PMs in creating optimized AI solutions while presenting investors with opportunities in a burgeoning domestic AI ecosystem.
Novasilicon, a startup in Shanghai, has secured millions in funding to revolutionize chip design using AI, aiming for a significant market share in custom chip development. Led by CEO Li Linyang, the company plans to leverage large models for design optimization, targeting industries with increasing chip demands.
Novasilicon's funding for AI-driven chip design signifies a shift towards automation in semiconductor development, which can drastically reduce design time and costs. Builders and PMs should consider the implications of AI in optimizing hardware solutions, while investors may see a lucrative opportunity in the growing demand for custom chips across various industries.

Intel's 'Intelligent PC' concept aims to run a 35B model on 32GB memory, enabling local processing to reduce costs and improve efficiency. This hybrid approach addresses the high costs of cloud-based AI while providing a user-friendly interface, as demonstrated by partners like remio and QClaw.
Intel's 'Intelligent PC' aims to run a 35B model on just 32GB of memory, which could significantly lower costs for deploying AI locally instead of relying on cloud services. This development is crucial for builders and PMs as it enables more efficient and scalable AI applications, while investors should note the potential for reduced operational costs in AI solutions.
The rise of Token billing in AI has transformed costs into operational expenses, with prices varying significantly due to factors like model efficiency, energy costs, and contract terms. As companies shift from GPU hours to Token-based billing, understanding the hidden complexities behind Token pricing becomes crucial for effective budgeting.
The shift to Token-based billing in AI represents a significant change in how operational costs are managed, necessitating a deeper understanding of pricing complexities. Builders and PMs must adapt their budgeting strategies to account for fluctuating Token prices influenced by model efficiency and energy costs, while investors should evaluate the financial implications of these new cost structures on AI startups.

The DPU market is rapidly expanding as AI infrastructure shifts focus from GPU performance to network efficiency, with companies like Cloud Leopard achieving significant milestones in DPU development, including a 400Gbps product. This transition highlights the critical role of DPU in optimizing AI systems for high-frequency inference and resource scheduling.
The rapid expansion of the DPU market, exemplified by Cloud Leopard's 400Gbps product, signifies a shift in AI infrastructure priorities towards network efficiency. This development is crucial for builders and PMs as it enhances system performance for high-frequency inference, while investors should note the potential for significant returns in this emerging sector.
The resurgence of Groq's LPU in NVIDIA's Vera Rubin platform marks a shift towards specialized chips for AI inference, with Groq's SRAM bandwidth reaching 150 TB/s, significantly outperforming traditional HBM solutions. As the industry embraces heterogeneous computing, the viability of LPU as a standalone business remains uncertain amid rising competition and evolving market demands.
The resurgence of Groq's LPU in NVIDIA's Vera Rubin platform highlights a significant shift towards specialized chips for AI inference, offering an impressive SRAM bandwidth of 150 TB/s. Builders and PMs should consider how this could impact their hardware choices, while investors need to assess the competitive landscape as the viability of LPU as a standalone business remains uncertain.

Leading tech firms are directly purchasing thousands of B300 units, signaling a shift in procurement strategies. A storage giant's valuation may rise to a trillion yuan, impacting existing semiconductor companies. Meanwhile, a Shanghai AI chip firm has secured Pro-IPO financing, and the market faces rising storage costs affecting AI chip strategies.
The direct purchase of thousands of B300 units by leading tech firms indicates a strategic shift in hardware procurement, which could influence AI infrastructure costs and availability. This, coupled with the rising valuation of a storage giant and the securing of Pro-IPO financing by an AI chip firm, suggests a tightening market that builders and investors need to navigate carefully.

JiuZhang Cloud's AI Factory aims to revolutionize AI deployment by standardizing computational power measurement and enhancing model production efficiency. With the introduction of DCU (standardized computational unit), the company addresses the industry's infrastructure gap, enabling scalable AI solutions that can adapt to various business needs.
JiuZhang Cloud's introduction of the DCU (standardized computational unit) addresses a critical infrastructure gap in AI deployment, allowing builders and PMs to develop scalable solutions more efficiently. For investors, this innovation signals a potential increase in market competitiveness and a pathway for broader AI adoption across diverse industries.
Shenzhen-based Taixin Semiconductor has developed AI plush toys with integrated cloud capabilities, costing under $40, driven by a $1 chip and $2 module. Despite low production costs, user engagement wanes after two weeks, raising concerns about long-term viability.
Shenzhen's Taixin Semiconductor has launched AI plush toys with cloud capabilities for under $40, utilizing a $1 chip and $2 module. This development highlights the potential for affordable AI integrations in consumer products, but the rapid decline in user engagement suggests that builders and PMs need to focus on sustainable user experiences to ensure long-term product viability.
Suin Technology is set for a crucial IPO meeting on June 15, 2026, after achieving an impressive 81.32% revenue CAGR from 2023 to 2025, with projected revenues reaching 9.90 billion CNY. The company aims for profitability by 2026 or 2027, leveraging its proprietary DSA architecture and strong partnerships with leading internet firms.
Suin Technology's upcoming IPO meeting on June 15, 2026, is significant as it highlights the company's impressive 81.32% revenue CAGR and potential profitability by 2026 or 2027. This signals strong market demand for innovative AI solutions and could attract builders and investors looking for high-growth opportunities in the tech sector.

Intel's Xeon 6+ processor, with 288 E-cores, can run over 1000 AI agents simultaneously, addressing a 417% surge in China's AI computing demand. Key technologies QAT and IAA enhance performance and reduce memory costs, making Agentic AI production-ready.
Intel's Xeon 6+ processor, capable of running over 1000 AI agents simultaneously, signals a significant leap in AI computing power, enabling builders and PMs to deploy scalable Agentic AI solutions efficiently. For investors, this development highlights a growing market demand for advanced computing infrastructure that supports the rapid evolution of AI applications.

Intel's Xeon 6+ processor, featuring 288 efficient cores, can deploy 400-500 agents simultaneously, marking a shift in AI infrastructure from GPU dominance to CPU importance. This new architecture enhances task management and resource allocation, achieving up to 2.26x performance improvement over its predecessor.
Intel's new Xeon 6+ processor, with 288 cores, enables the simultaneous operation of 400-500 agents, signaling a shift in AI infrastructure from GPU to CPU. This development suggests that builders and PMs can optimize resource allocation and task management, while investors should consider the implications for future AI hardware investments.