"At the time when 'Lobster Fever' sweeps the globe, Huang Renxun leads the open-source model to dominate the scene! NVIDIA Corporation (NVDA.US) has ambitious plans to support the 'AI bull market narrative'."

With the introduction of Claude Cowork by Anthropic and the popularity of AI agents such as OpenClaw (also known as "lobster") that can autonomously perform tasks globally, the "AI chip superpower" NVIDIA Corporation (NVDA.US) seeks to grasp this wave of the AI agent super wave. The company has introduced its open-source large-scale AI model "Nemotron 3 Super," specifically designed for extremely complex agent-based AI intelligent systems. On the Pinchbench benchmark level, Nemotron 3 Super is the top performer, sitting comfortably at the top of the open-source models. In terms of success rate on OpenClaw tasks, it scored a high 85.6%, coming close in performance to closed-source models like Claude Opus 4.6 and GPT-5.4. NVIDIA Corporation's latest move significantly strengthens this globally highest-valued company (approximately 4.5 trillion USD) in its transition from a simple AI chip supplier to a full-stack platform of "models, toolchains, cloud inference services, and AI ecosystems." For NVIDIA Corporation's stock price, it may soon reach a new all-time high and drive the global AI computing industry towards a new upward trajectory. The "NVIDIA Corporation full-stack AI ambition" is playing a crucial role in supporting the "AI bull market narrative" in the capital markets amidst the intense volatility caused by the political situation in the Middle East GEO Group Inc. In an official statement, NVIDIA Corporation stated that this model combines state-of-the-art inference capabilities, enabling it to efficiently and accurately complete large-scale AI tasks suitable for enterprise-level autonomous AI agent systems. NVIDIA Corporation stated that this new 120 billion parameter open model adopts a hybrid expert model (MoE) architecture, incorporating three innovations. Compared to the previous generation Nemotron Super model, the inference performance has more than tripled, throughput can be increased up to five times, and accuracy can be doubled. According to NVIDIA Corporation, AI search leader Perplexity has started to provide Nemotron 3 Super to its users for systemized AI agent-driven searches, as one of the 20 models in Computer. Technology companies providing advanced software development agents, such as CodeRabbit, Factory, and Greptile, are integrating this model with their proprietary AI large models to achieve higher accuracy at a lower cost and significantly improve business efficiency. According to NVIDIA Corporation, life science and cutting-edge AI research institutions like Edison Scientific and Lila Sciences will use this flagship open-source model to support their agent modes, for complex functions such as deep literature retrieval, data science, and molecular understanding. NVIDIA Corporation also mentioned that companies such as Amdocs, American AI and data analysis leader Palantir (PLTR.US), EDA chip design software leader Cadence (CDNS.US), as well as two major European giants - Dassault Systemes and Siemens, are actively deploying and customizing NVIDIA Corporation's model. This will help realize the vision of agent-based automation in sectors such as telecommunications, network security, semiconductor design, and manufacturing, as well as comprehensive updates and subscription-based products. Looking at the model structure and deployment parameters, Nemotron 3 Super is not simply a model with 120 billion parameters stacked, but rather a system more suitable for enterprise-level agent-type workflows a "high total participation, low activation" system. With 120 billion total parameters, it only activates 12 billion during inference, supports a native context window of 1 million tokens, and has a minimum deployment threshold of 8xH100 80GB. Its backbone is a mixed architecture of LatentMoE+Mamba-2+ a small amount of Attention, with two shared-weight Multi-Token Prediction (MTP) layers. The base model has 88 layers, a model dimension of 4096, 32 Q heads, 2 KV heads, 512 experts per layer, top-k activation of 22 experts, and a MoE Latent Size of 1024, according to the official technical report. NVIDIA Corporation's innovative open-source AI agent model design serves a very clear engineering purpose: to control activation costs with MoE, extend context and throughput with Mamba, and maintain precise retrieval and inference stability with Attention. Therefore, it is more like an "agent orchestration brain" focusing on multi-agent orchestration, long-chain tool calling, and long-context memory, rather than simply pursuing scores for single-turn dialogue with large models. The CEO of Qualcomm, the world's largest smartphone chip company, Cristiano Amon, recently stated at the Mobile World Congress in Barcelona that the upcoming "AI agent" super wave will change a broader digital ecosystem, shifting from a mobile-centric, application-centric ecosystem to an era-centric ecosystem centered around agents. Amon stated that 2026 will be the "year of AI agents," emphasizing the central role agents will play in observing, interpreting, and acting. NVIDIA Corporation's ambition: Not only to be a chip supplier but also an "AI infrastructure contractor" In terms of efficiency metrics, the absolute selling point of Nemotron 3 Super is not its overwhelming accuracy but its ability to reduce the inference throughput and cost of agent systems while maintaining similar accuracy levels. According to NVIDIA's official blog, Nemotron 3 Super achieved 85.6% in the PinchBench full test suite, making it the "best open-source model in its class." In terms of success rate on OpenClaw tasks, it scored 85.6%, approaching the performance of Claude Opus 4.6 and GPT-5.4. Therefore, a more accurate positioning of NVIDIA Corporation's new Nemotron 3 Super would be: if an ordinary enterprise needs to perform complex multi-step agents, long process orchestration, code/terminal/tool calling mixed workloads, Nemotron 3 Super may not be the strongest in a single point, but it is likely one of the closest to a scalable deployed "agent intelligence brain" within the current open-source vs. paid closed-source competitive landscape. Regarding moats, with Huang Renxun leading the way, NVIDIA Corporation has built a solid "super AI moat" with its AI GPU computing system + CUDA. The emergence of Nemotron 3 Super will undoubtedly strengthen this moat. As mentioned earlier, NVIDIA Corporation's role is increasingly resembling an AI infrastructure contractor, rather than just a chip provider. According to NVIDIA Corporation's blog, Nemotron 3 Super is optimized not just for NVIDIA Corporation GPU platforms but specifically designed for Blackwell inference efficiency and agent scenario optimization. It is said to offer up to a 5x increase in throughput and a 2x increase in accuracy compared to the previous Nemotron Super model. In an 8k input / 64k output environment, its throughput can reach 2.2x that of GPT-OSS-120B and 7.5x that of Qwen3.5-122B. When running heavy-duty inference tasks on Blackwell using NVFP4, it is up to 4 times faster than Hopper FP8. This synergy between "model architecture - quantization format - inference framework - flagship GPU platform" will make it difficult for CUDA, TensorRT-LLM, NIM, DGX/Blackwell's linkage to be replaced by any single variable, indicating that NVIDIA Corporation is elevating its AI system capabilities from "single GPU performance and CUDA barriers" to a full set AI system capability of "model architecture - inference stack - GPU platform - enterprise deployment." Recent attitudes of Wall Street analysts towards NVIDIA Corporation have taken a more bullish turn on the margins with the advent of Nemotron 3 Super. The trend of Nemotron 3 Super and NVIDIA Corporation's shift towards a "full-stack AI platform" are likely to collectively propel NVIDIA Corporation's stock price to break its previous all-time high set in October at 212.167 USD. As of the end of Wednesday's closing on the US stock market, NVIDIA Corporation's stock price closed at 186.03 USD. The latest channel surveys from Morgan Stanley show that the global "AI computing supply-demand gap is expanding significantly on a daily basis," and cloud computing giants (Hyperscalers) continue to aggressively drive growth in AI workloads. Even though some of these Hyperscalers' customers (such as Amazon.com, Inc. and Meta) may develop their own AI ASICs or purchase AMD AI GPU clusters, it is expected that the procurement of NVIDIA Corporation products by these super clients will increase by over 80% in 2026. Morgan Stanley points out that the upcoming GTC 2026 conference will showcase NVIDIA Corporation's leading technological roadmap to effectively counter market concerns about share loss. The Vera Rubin architecture and NVIDIA Corporation's latest efforts in Physical AI will open up a completely new market total addressable market (TAM). As model scales, inference pathways, and multi-modal/agent-type Agentic AI workloads drive exponential expansion in computing power consumption, the main theme of technology giants' capital expenditure is more inclined towards centralized AI computing infrastructure. Global investors continue to anchor the "AI bull market narrative" around NVIDIA Corporation, AMD, and the delivery of AI computing clusters as one of the most certain investment narratives in the global stock markets. This also indicates that investments in electricity, liquid cooling systems, optical interconnect supply chains, and other areas closely related to AI training/inference will follow the lead of AI computing leaders such as NVIDIA Corporation, AMD, Broadcom Inc., Taiwan Semiconductor Manufacturing Co., Ltd. Sponsored ADR, and Micron in the midst of uncertainty in the political situation of the Middle East GEO Group Inc. According to the latest analyst expectations compiled by institutions, Amazon.com, Inc. together with the parent companies Alphabet Inc. Class C, Meta Platforms Inc., Oracle Corporation, and Microsoft Corporation, are expected to collectively spend approximately 650 billion USD on AI-related capital expenditures by 2026. Some analysts believe that the overall expenditure may exceed 700 billion USD - indicating a year-over-year increase in AI capital spending of more than 70%. It is worth noting that these five major US super tech giants are expected to invest around 1.5 trillion USD between 2023 and 2026 to build a massive AI computing infrastructure. In comparison, these tech giants have invested approximately 600 billion USD throughout their entire historical period before 2022. In the official blog, NVIDIA Corporation stated that Nemotron 3 Super reached 85.6% in the full PinchBench test suite, positioning it as the "best open-source model in its class," and achieved 85.6% success rate on OpenClaw tasks. It also highlighted that it approached the performance level of Claude Opus 4.6 and GPT-5.4. Therefore, for NVIDIA Corporation's latest Nemotron 3 Super, a more accurate positioning is: if an ordinary enterprise needs to execute complex multi-step agents, long process orchestration, code/terminal/tool calling mixed workloads, Nemotron 3 Super may not be the strongest in a single point, but it is likely one of the closest to a scalable deployed "agent intelligence brain" within the current open-source vs. paid closed-source competitive landscape. In terms of moats, under the leadership of Huang Renxun, NVIDIA Corporation has built a solid "super AI moat" with its AI GPU computing system + CUDA. The emergence of Nemotron 3 Super will undoubtedly strengthen this moat. As mentioned earlier, NVIDIA Corporation's role is increasingly resembling an AI infrastructure contractor, rather than just a chip provider. According to NVIDIA Corporation's blog, Nemotron 3 Super is optimized not just for NVIDIA Corporation GPU platforms but specifically designed for Blackwell inference efficiency and agent scenario optimization. It is said to offer up to a 5x increase in throughput and a 2x increase in accuracy compared to the previous Nemotron Super model. In an 8k input / 64k output environment, its throughput can reach 2.2x that of GPT-OSS-120B and 7.5x that of Qwen3.5-122B. When running heavy-duty inference tasks on Blackwell using NVFP4, it is up to 4 times faster than Hopper FP8. This synergy between "model architecture - quantization format - inference framework - flagship GPU platform" will make it difficult for CUDA, TensorRT-LLM, NIM, DGX/Blackwell's linkage to be replaced by any single variable, indicating that NVIDIA Corporation is elevating its AI system capabilities from "single GPU performance and CUDA barriers" to a full set AI system capability of "model architecture - inference stack - GPU platform - enterprise deployment." Recent attitudes of Wall Street analysts towards NVIDIA Corporation have taken a more bullish turn on the margins with the advent of Nemotron 3 Super. The trend of Nemotron 3 Super and NVIDIA Corporation's shift towards a "full-stack AI platform" are likely to collectively propel NVIDIA Corporation's stock price to break its previous all-time high set in October at 212.167 USD. As of the end of Wednesday's closing on the US stock market, NVIDIA Corporation's stock price closed at 186.03 USD. The latest channel surveys from Morgan Stanley show that the global "AI computing supply-demand gap is expanding significantly on a daily basis," and cloud computing giants (Hyperscalers) continue to aggressively drive growth in AI workloads. Even though some of these Hyperscalers' customers (such as Amazon.com, Inc. and Meta) may develop their own AI ASICs or purchase AMD AI GPU clusters, it is expected that the procurement of NVIDIA Corporation products by these super clients will increase by over 80% in 2026. Morgan Stanley points out that the upcoming GTC 2026 conference will showcase NVIDIA Corporation's leading technological roadmap to effectively counter market concerns about share loss. The Vera Rubin architecture and NVIDIA Corporation's latest efforts in Physical AI will open up a completely new market total addressable market (TAM). As model scales, inference pathways, and multi-modal/agent-type Agentic AI workloads drive exponential expansion in computing power consumption, the main theme of technology giants' capital expenditure is more inclined towards centralized AI computing infrastructure. Global investors continue to anchor the "AI bull market narrative" around NVIDIA Corporation, AMD, and the delivery of AI computing clusters as one of the most certain investment narratives in the global stock markets. This also indicates that investments in electricity, liquid cooling systems, optical interconnect supply chains, and other areas closely related to AI training/inference will follow the lead of AI computing leaders such as NVIDIA Corporation, AMD, Broadcom Inc., Taiwan Semiconductor Manufacturing Co., Ltd. Sponsored ADR, and Micron in the midst of uncertainty in the political situation of the Middle East GEO Group Inc. According to the latest analyst expectations compiled by institutions, Amazon.com, Inc. together with the parent companies Alphabet Inc. Class C, Meta Platforms Inc., Oracle Corporation, and Microsoft Corporation, are expected to collectively spend approximately 650 billion USD on AI-related capital expenditures by 2026. Some analysts believe that the overall expenditure may exceed 700 billion USD - indicating a year-over-year increase in AI capital spending of more than 70%. It is worth noting that these five major US super tech giants are expected to invest around 1.5 trillion USD between 2023 and 2026 to build a massive AI computing infrastructure. In comparison, these tech giants have invested approximately 600 billion USD throughout their entire historical period before 2022.