The world's first 1.8nm chip is here!

The world's first 1.8nm chip has finally officially debuted. On January 5th, at the International Consumer Electronics Show (CES) 2026, Intel announced its annual flagship AI PC chip - the third-generation Core Ultra series processor (codename Panther Lake). This is the first computing platform based on Intel 18A process (1.8nm level), bringing AI PCs into the era of AI with edge-side AI computing power reaching up to 180TOPS. The mobile product line of the third-generation Intel Core Ultra introduces the new Intel Core Ultra X9 and X7 processors, both integrated with Intel Iris Xe graphics, designed to meet the demands of complex multitasking, capable of handling intricate workloads such as gaming, content creation, and productivity. The flagship model is equipped with up to 16 CPU cores, 12 Xe cores, and 50TOPS of NPU computing power, delivering up to 60% improvement in multithreading performance, up to 77% improvement in gaming performance, and achieving up to 27 hours of lasting battery life, covering over 200 PC product designs. For example, the Lenovo IdeaPad reference design featuring the Core Ultra X9 388H, with a battery capacity of 99Whr and a 2.8k OLED display, can achieve a streaming playback time of up to 27.1 hours on Netflix, which Intel has dubbed the "x86 battery king." The first consumer laptops equipped with the third-generation Intel Core Ultra processors will begin pre-sales on January 6th and will be available globally starting January 27th. More product designs will be launched throughout the first half of this year. It is worth mentioning that this AI PC chip has also "crossed over" into humanoid robots. The third-generation Intel Core Ultra has achieved the first-ever synchronous release of edge processors and their corresponding PC versions, receiving testing and certification for embedded and industrial edge scenarios, and will support fields such as embodied intelligence, smart cities, automation, and healthcare. Intel claims it "has proven to be the ideal choice for mission-critical edge and physical AI." In terms of performance improvements for large language models (LLM), end-to-end video analysis, and visual language action (VLA) models, the third-generation Intel Core Ultra demonstrates outstanding performance, and is also more competitive in AI inference costs compared to traditional multi-chip CPU and GPU architectures. Intel is actively promoting the integration of edge AI and embodied intelligence technology, which means placing processors that were previously installed in PCs into various types of robots, such as humanoid robots, fixed robotic arms, and autonomous mobile robots At the exhibition, Intel showcased a humanoid robot powered by the third-generation Core Ultra. To support these emerging scenarios, Intel collaborated with edge ODMs to launch reference boards and development kits, integrating the Intel Robotics Software Suite, tools, frameworks, and applications, which will be provided simultaneously with hardware releases. As a result, the third-generation Core Ultra will become Intel's most widely covered and globally available AI PC platform in history. Intel also announced that it will launch a complete handheld gaming device platform based on Panther Lake, with more details to be revealed later this year by hardware and software partners. This press conference also marked the first public appearance of Intel CEO Eric Wu in 2026. Eric Wu shared that Intel's mission is to make intelligence accessible, efficient, and deployable everywhere. According to him, over the past year, Intel has continuously pushed the limits in architecture, manufacturing processes, and cross-software and hardware collaborative optimization, fulfilling its commitments in product shipments and completing the first batch of products based on the Intel 18A process. Jim Johnson, Intel's Senior Vice President and General Manager of the Client Computing Group, discussed that the entire industry, along with Intel, is approaching a strategic turning point in 2026. Intel is delivering leadership products based on leading process nodes, integrating computing, graphics, and AI, and expanding together with customers to create its most extensive AI PC platform ever. Aravind Srinivas, CEO and co-founder of Perplexity, also attended the event, stating that local computing will become increasingly important, primarily due to performance, security, cost-effectiveness, and control. For these reasons, Perplexity will launch the enterprise-focused AI browser Comet next month. After the event, Feng Dawei, Vice President of Intel's Client Computing Group and General Manager of the Client Segments Division, and Gao Song, Vice President of Intel and General Manager of Software Engineering and Client Products in China, engaged in in-depth discussions with media outlets such as Zhidx. Feng Dawei shared that from a hardware architecture perspective, one significant method for Panther Lake to achieve low power consumption is through low-power islands, which have four E cores (energy-efficient cores) placed in a lower power range, with dedicated MagCache. The vast majority of non-performance applications can be retained in the low-power island without spilling over. Gao Song mentioned that 2026 is a critical year for PC development, and this year is the turning point. Whoever can launch a SoC that adapts to the future development of PC products at the beginning of the year will be able to seize this turning point. First use of 1.8nm process, local AI computing power reaches 180TOPS The third-generation Core Ultra series processors (codename Panther Lake) adopt the Intel 18A process, with AI computing power up to 180TOPS, integrating the high energy efficiency of the Core Ultra 200V series (codename Lunar Lake) and the high performance and scalability of the Ultra 200H series (codename Arrow Lake-H) It is a key hardware that carries Intel's Agentic AI ambitions. The Intel 18A process is the first 2nm node developed and manufactured in the United States. Through two major technological breakthroughs, RibbonFET fully wrapped gate transistors and PowerVia backside power delivery, it pushes the energy efficiency and density of chips to new heights. The former allows architects to control current more precisely, while the latter enhances power transmission and signal integrity. Compared to Intel 3, Intel 18A improves performance per watt by more than 15%, and transistor density increases by 30%. Thus, Intel becomes the first company in the industry to combine fully wrapped gates and backside power delivery at a large-scale production node. Samsung and TSMC's 2nm processes also adopt fully wrapped gate (GAA) transistor technology and are expected to be launched this year. TSMC plans to introduce backside power delivery technology at the N16 node this year, while Samsung may not use backside power delivery technology until next year. Intel 18A will become the cornerstone of Intel's next three generations of PC and data center products. Its engineers have also begun developing more advanced technologies such as Intel 14A. Panther Lake uses Intel's Foveros-S packaging technology, consisting of various modules produced by different process technologies: Compute tile (Intel 18A) Graphics tile (Intel 3/TSMC N3E) Platform controller tile (TSMC N6) Base tile (Intel 1227.1) Fill tile (used to maintain the integrity of the entire chip) This processor is designed for Agentic AI, with edge AI computing power of up to 180TOPS, LPDDR5x bandwidth of up to 9600MT/s, and a maximum capacity of 96GB, running large language models 4.3 times faster than AMD HX 370 and 2.0 times faster than Core Ultra 200H. CPU: 10TOPS, fast, suitable for running lightweight AI GPU: 120TOPS, high bandwidth, suitable for gaming and creative AI tasks NPU: 50TOPS, high energy efficiency, suitable for running AI assistants The AI acceleration dedicated unit NPU 5 of Panther Lake focuses on high efficiency, with a smaller area than the previous generation, simplifying backend functions. By doubling the scale of the MAC array, it improves performance per unit area by more than 40% and natively supports two FP8 data formats: E4M3 and E5M2. At the same time, the low-power NPU with 50TOPS can achieve always-on, always-inference, covering typical scenarios such as video conferencing and security Based on heterogeneous hardware architecture, Intel has achieved synergistic effects among CPU, NPU, and NPU through deep software optimization. By combining OpenVINO with a maximum of 96GB of system memory, the third-generation Core Ultra can run models with 70 billion parameters at a 32K context length. Previously at the Intel Industry Solutions Conference 2025 in China, Intel also revealed that through a series of key technological innovations in sparse attention, inference decoding, KV Cache compression, and more, Panther Lake is now capable of supporting 80 billion parameters and 32K context windows for MoE (Mixture of Experts) models, with the first word response time within 30 seconds, and the token throughput for running agent applications increased to 2.7 times the original. Multiple configurations, unified packaging, joint hardware and software enhancement of energy efficiency Significant improvements have been made to almost every SoC subsystem in Panther Lake. Intel has redesigned the core for Intel 18A to operate at lower voltages, achieving single-core performance improvements while further enhancing energy efficiency. By adding up to 8 energy-efficient cores (E cores), Intel has improved the multi-core performance of the third-generation Core Ultra. Compared to Lunar Lake, Panther Lake's performance has improved by 60%, and even with fewer performance cores (P cores), it is faster and more efficient than its previous highest-performing SKU. The low-power subsystem has also been restructured, introducing low-power energy-efficient cores (LP-E cores) with independent caches to run hundreds of workloads optimized for batch processing and endurance, such as web browsing and video conferencing, achieving higher performance, lower power consumption, and longer battery life. At the platform level, Intel has deployed multiple technologies focused on energy efficiency, including Intel Smart Display, context-aware charging, low-power video conferencing image processing, and long-duration connectivity based on Intel Wi-Fi. Through advanced packaging technology, Intel offers various configurations of the third-generation Core Ultra in the same package form factor, providing customers with greater design flexibility, including more memory options and the ability to customize power supply solutions. The 16-core CPU + 12 Xe3 configuration additionally expands 8 PCIe 5.0 lanes, enhancing connectivity capabilities for high-performance devices. For more configurations and technical details about Panther Lake, as well as how this chip achieves a balance between high performance and high energy efficiency, please refer to the article "Decisive Battle at 1.8nm! Intel's AI Roadmap Released, Two Major AI GPUs to be Launched." Announcing the largest integrated GPU, supporting AI multi-frame generation At the same time, Intel announced the new generation integrated GPU - Arc B390. It features more than 50% of the graphics units, a cache capacity that has doubled, and integrates 96 XMX AI acceleration units, providing 120 TOPS of graphical AI computing power. Compared to Lunar Lake, Panther Lake has improved gaming performance by 70% and AI inference performance by 50%, allowing PC gamers to experience near discrete graphics card-level gaming performance on thin and light laptops. Test results show that the new Arc GPU can not only continuously provide a smooth gaming experience but also, under similar power consumption and memory configurations, achieves an average frame rate that is 70% higher than AMD Radeon’s latest products, with some games even reaching 2 times. For example, in first-person shooter games like "Painkiller" and "Delta Force," under a 10W TDP, high-quality settings, and with XeSS super-resolution enabled, the third-generation Core Ultra can deliver outstanding performance. The truly unique aspect of the third-generation Core Ultra graphics architecture is that it is designed for the future, which Intel refers to as " Modern Rendering." This means that when enabling high-load visual effects like ray tracing, better lighting quality can be achieved; sharper images can be delivered through higher quality textures and more complex geometric assets; and an extremely smooth experience can be realized through AI-generated frames. Intel's new generation Arc is the world's first integrated graphics card to support AI multi-frame generation at launch: for every 1 frame rendered, 3 frames are generated by AI. As a result, it can achieve 4 times the frame rate, allowing AAA games to run smoothly at high quality. A prime example is "Battlefield 6," released at the end of last year. Intel collaborated deeply with EA, and under the highest load "Overkill" graphics settings in the game, the Arc B390 can already provide a smooth experience through super-resolution. When multi-frame generation is enabled in the driver, its frame rate can expand to over 120 FPS, nearly 3 times that of AMD processors. Currently, over 300 games worldwide support XeSS technology. An AI application called GameSkill CS2 AI Coach, developed by Beijing AI gaming startup New Wisdom Games, also appeared in Intel's exhibition area. This is an AI product designed for tactical reviews of CS esports professional teams, optimizing training efficiency through AI technology, accelerating the speed at which coaches and analysts review training or professional competitions, and has been deeply optimized based on Intel's Core Ultra series chip XPU architecture and ARC graphics cards With the support of Intel's technical team, GameSkill's professional AI coach for CS adopts a full-end solution. The technical team fine-tunes the Qwen3-14B model by analyzing and processing data from tens of thousands of game rounds. The entire inference process runs on Intel B60 graphics cards, generating tactical reviews. Hybrid AI is the Trend By integrating the entire technology stack—from hardware, software, operating systems to firmware—Intel is at the forefront of the industry, redefining the platform foundation for client-side AI computing, allowing local intelligence to truly land on various edge devices, including PCs. It is reported that Intel's computing power shipped to the edge exceeds 4 ZettaOPS, equivalent to the computing capacity of 40 data centers. The key is, how to activate this computing power more efficiently? Intel believes that hybrid AI is the trend, and to truly implement this model, it is necessary to enable better collaboration between edge models and cloud models. Local AI is responsible for securely executing tasks on devices, ensuring that data remains on the device; cloud AI handles global inference, planning, and multi-agent orchestration. The communication between the two brings higher security, better privacy protection, superior performance, and lower costs. It is worth mentioning that ByteDance is the only Chinese partner to have a dedicated PPT in this presentation. ByteDance launched the AI rough cut cloud function in CapCut in 2025, which automatically edits and generates short video summaries that conform to story logic, and it is expected to cover millions of users this year, but this puts immense pressure on cloud computing power. Realizing this, ByteDance utilized the massive computing power of Intel AI PCs in collaboration with the cloud, resulting in achieving the same quality and faster video editing speed, while the product operator can reduce operational costs by lowering the demand for cloud computing power. Against the backdrop of increasingly limited data center infrastructure, more and more AI workloads will migrate to the edge, but it will not completely detach from the cloud; instead, it needs to work seamlessly with the cloud. The third-generation Core Ultra series processors mark the starting point of this hybrid AI era. In addition, Intel emphasizes that it will optimize and support users to run mainstream AI models locally on PCs as soon as possible, citing the Chinese market as an example. Last year, it achieved Day0 support for Alibaba's latest Qwen3 large model on Core Ultra, allowing developers to plug and play. Intel is actively collaborating with ecosystem partners to expand solutions that originally relied solely on the cloud into a hybrid architecture. At the same time, Intel has made significant investments in software engineering and infrastructure, supporting hundreds of leading ISVs by providing the tools needed to optimize AI applications on Intel CPUs, GPUs, and NPUs. For example, Adobe Premiere Pro is using Arc GPUs for media search. Just describe what you want to find, and the system can help you locate the corresponding clips, even if the material is not labeled or buried deep within a large amount of video For example, Zoom has launched a virtual fill light feature, which runs on a low-power NPU, automatically brightening the subject while dimming and weakening the background. Intel ensures that the software can run on the most comprehensive set of frameworks and tools, supporting mainstream industry deployment paths such as LLaMA.cpp and PyTorch; the OpenVINO tool provides production-ready deep optimizations across various computing engines; Intel has also deeply integrated key open-source optimizations into Microsoft Windows ML. In addition, Intel is also an important partner in the launch rhythm of Copilot+. Copilot+ will be launched alongside the entire third-generation Core Ultra product line. Conclusion: Designed for Future AI Native PCs, Evolving Around Agents and Multimodal Perception Overall, the third-generation Intel Core Ultra processors offer longer battery life, faster graphics performance, higher overall performance, and more AI capabilities, all without waiting. Edge AI computing is an inevitable trend. Since Intel proposed the concept of AI PCs, the development of AI PCs has gradually evolved from "AI-enhanced PCs (traditional applications + AI plugins)" to "AI native PCs." The five major capabilities of AI PCs—perception, cognition, execution, memory, and learning—are co-evolving, ushering in a new era of intelligent applications. The future AI native PC will evolve into a partner with multimodal perception capabilities. Its interaction will no longer be limited to keyboards but will be based on natural interactions involving language, screens, and cameras, possessing deeper perception and understanding abilities. Today, Intel stands at a strategic turning point in client-side AI computing. The third-generation Core Ultra has achieved multidimensional performance growth and overall energy efficiency optimization through comprehensive upgrades in process, packaging, and architecture, and is supporting the proliferation of more intelligent experiences in edge AI devices through innovative designs from OEM and ISV partners. Risk Warning and Disclaimer The market has risks, and investment should be cautious. This article does not constitute personal investment advice and does not take into account the specific investment goals, financial conditions, or needs of individual users. Users should consider whether any opinions, views, or conclusions in this article align with their specific circumstances. Investment based on this is at their own risk