Representative End User Clients
Representative Automation Clients
Representative Software Clients
The Consumer Electronics Show (CES) 2025 has come and gone, but the impact of Nvidia’s announcements continues to reverberate through the tech world, particularly in the context of the ongoing “AI Wars”. As highlighted in my December 2024 blog, the AI landscape is characterized by several key battlefronts: Datacenter Hardware, Edge Hardware, General Purpose AI Software Platforms, Edge AI Software, and Industrial-Grade AI. This year Nvidia’s CES keynote showcased a significant shift in focus, moving beyond just generative AI, to one that incorporates physical AI, AI agents, and a more accessible infrastructure for AI development. This shift is enabling the creation of more realistic simulations and digital twins, the deployment of advanced AI models, and the expansion of AI capabilities to new applications and industries. Nvidia’s announcements at CES 2025 demonstrates their comprehensive approach that addresses these critical battlefronts, pushing the boundaries of what's possible with AI and providing developers with the tools they need to innovate across all sectors.
Recap of the AI Wars Battlefronts
Before diving into the details, let’s revisit some of the key battlefronts in the AI Wars:
Datacenter Hardware: The competition for high-performance, efficient, and cost-effective hardware to train and deploy large AI models in data centers.
Edge Hardware: The push to bring AI capabilities closer to the data source, requiring optimized chips for edge devices.
General Purpose AI Software Platforms: The race to provide comprehensive tools for training, validating, and deploying diverse AI models and techniques.
Edge AI Software: The focus on reducing the complexity and cost of deploying AI models to edge devices, enabling real-time processing without relying solely on cloud connectivity.
Industrial-Grade AI: The specific needs and alliances around domain expertise, data fabrics, and digital twins in industrial contexts.
Nvidia's Multifaceted Announcements at CES 2025
Nvidia’s CEO, Jensen Huang, presented a vision that emphasizes how the company is advancing AI and graphics capabilities across all these fronts. While Project DIGITS is a major step forward for edge AI, the versatility of the new GeForce RTX 5000 series GPUs demonstrates Nvidia’s strategy for broad market penetration.
GeForce RTX 5000 Series GPUs: Beyond Gaming
Nvidia's unveiling of the GeForce RTX 5000 series, powered by the Blackwell architecture, goes beyond catering solely to the gaming community. The series, led by the flagship RTX 5090, showcases significant advancements in real-time ray tracing, AI-driven Deep Learning Super Sampling (DLSS), and overall compute capabilities. The RTX 5090 boasts 92 billion transistors, 4000 TOPS, four petaflops of AI, 380 ray tracing teraflops, and 125 shader teraflops. It also features 1.8 terabytes per second of memory from Micron. The flagship card is priced at $1,999.
The RTX 5070, offering performance comparable to the previous generation RTX 4090, is priced at $549, making high-end graphics more accessible. Moreover, Nvidia is integrating these powerful GPUs into laptops, with a 5070-based laptop priced at $1299. This further expands the reach of high-performance computing. The RTX 5000 series incorporates AI to generate additional pixels and frames, significantly boosting rendering efficiency and enabling higher performance with lower power consumption.
Relevance across AI Wars—Datacenter Hardware: While not exclusively designed for datacenters, the RTX 5000 series is relevant due to its substantial compute and AI processing capabilities. Datacenters can leverage these GPUs for various workloads, including AI inference, and even AI training for certain use cases, especially smaller models or models that need to be fine-tuned for specific edge deployments. The ability of these GPUs to handle both graphics and AI workloads means that they can serve as versatile accelerators in a datacenter. The use of AI to improve rendering efficiency (DLSS) can reduce costs by decreasing computation time without sacrificing visual quality.
Edge Hardware: The RTX 5000 series is highly relevant to edge AI due to its high performance and improved power efficiency. The integration of these GPUs into laptops, and the ability to generate pixels and frames using AI, means that high-powered AI compute can be deployed to a wide range of edge devices and scenarios. This also enables the local deployment of AI models for real-time processing in industrial and other use cases where low latency is essential. The ability of the shader to process neural networks and the invention of neuro-texture compression and neural material shading enable amazing image quality that would have been impossible without AI. This technology also helps enable smaller form factors and reduced power consumption, making them ideal for edge deployments.
Consumer Gaming: Of course, these GPUs will be widely used in gaming, and this market continues to be important to Nvidia and the ecosystem. The RTX 5000 series delivers high-fidelity gaming experiences, integrating ray tracing and AI-enhanced graphics. The performance boosts and AI-driven rendering make these cards highly desirable for gamers who want cutting-edge visuals, performance, and power efficiency.
Physical AI and the Evolution of Digital Twins
Jensen Huang introduced "Physical AI," which centers around AI models understanding the physical world. This is enabled by Nvidia Cosmos, a World Foundation Model trained on a massive dataset of 20 million hours of video that focuses on physical dynamics, such as nature themes, human movement, and object manipulation. This training allows Cosmos to understand concepts like gravity, friction, inertia, spatial relationships, cause and effect, and object permanence, which are critical for physical AI applications.
When connected to Omniverse, Nvidia's physics-based simulation platform, Cosmos enables a physically grounded multiverse generator. This approach differs from previous digital twin and Metaverse concepts by integrating simulations to generate training data and using AI to learn from realistic physics-based scenarios.
Relevance to Industrial-Grade AI: This initiative directly addresses the Industrial-grade AI battlefront, where digital twins have been historically underutilized, or are only used for design and not as a learning and training tool. Nvidia is positioning its physical AI as a key technology to reinvigorate the value of digital twins by infusing them with AI. This enables the generation of synthetic data and facilitates the deployment of AI in real-world robotic applications and industrial scenarios and is grounded in real world physical limitations and properties. The pairing of Cosmos and Omniverse is essential to this approach.
Project DIGITS: A New Paradigm for Edge AI
ProjectDIGITS is a desktop-sized AI supercomputer powered by the new GB10 Grace Blackwell Superchip. Priced at $3,000, it offers a petaflop of AI computing performance and can run models with up to 200 billion parameters. It is designed to allow users to develop, fine-tune, and run AI models locally before deployment.
Relevance to Edge Hardware: Project DIGITS marks a significant leap in edge AI training and deployment. Unlike other companies focusing on edge inference, Project DIGITS empowers individual developers, researchers, and students with the ability to train and deploy powerful AI models locally. This addresses the need for accessible and powerful edge hardware, particularly for industrial applications where real-time processing and low latency are essential. The device also promotes a more democratic approach to AI development, reducing reliance solely on cloud resources, and securing Nvidia’s position as a leader in hardware for AI model training.
Industrial AI Use Cases for Project DIGITS: The potential applications of Project DIGITS across industries are vast, with examples including:
Real-Time Process Optimization: Training AI models on sensor data from factory machines to predict failures, optimize processes, and reduce waste without cloud reliance.
Robotics and Automation: Prototyping and testing AI models for robots on individual desktops, accelerating the development of new robotic systems for various industrial tasks.
Quality Control: Locally training AI models to analyze product images for defects, enabling more efficient and accurate quality control.
Custom AI Agent Development: Fine-tuning Nvidia Llama-based models to create AI agents that understand company-specific processes and terminologies using NIMS and Nemo, without a data center.
NVIDIA NIMS and Nemo: Democratizing AI Development
Nvidia is providing AI microservices (NIMS) and a digital employee onboarding and training system (Nemo) to simplify the development and deployment of AI agents. These tools allow for the packaging, optimization, and containerization of AI models for integration into various software packages. The NIMS offering includes models for vision, language, speech, animation, digital biology, and new models for physical AI, available across cloud, OEM, and on PCs through Windows Subsystem for Linux (WSL). The Nemo platform is designed to train AI agents with company-specific knowledge, processes, and guidelines, “transforming IT departments into the HR departments of AI agents”.
Nvidia is also releasing a family of open-source Llama-based models fine-tuned for enterprise use.
Relevance to General Purpose AI and Edge AI Software: These developments are crucial for both the general-purpose AI software platform and edge AI software battlefronts. NIMS and Nemo streamline the AI development and deployment process, particularly for industrial applications. The Llama-based models provide a flexible foundation for enterprise AI agents, and these models and systems can be deployed everywhere. This highlights Nvidia’s ambitions to deliver comprehensive, easily accessible AI platforms for developers, and software modernization is key to realizing the full potential of GenAI.
The AI Wars: A Fast-Paced Start to 2025
Nvidia’s announcements at CES 2025 underscore that the AI Wars are intensifying, with the company pushing the boundaries of AI hardware and software, empowering developers across all sectors, and demonstrating Nvidia's leadership position in AI and graphics. The new RTX 5000 series GPUs and Project DIGITS are poised to be significant steps forward in edge AI, and physical AI is revitalizing the potential of digital twins. These innovations are driving the evolution of robotics, digital twins (and the revival of metaverses), and industrial automation, and are creating a new era of computing.
The versatility of the RTX 5000 series is a strategic advantage, as it can be used in everything from consumer gaming to edge computing and even in datacenters. This broad applicability ensures that Nvidia's technology can penetrate multiple markets simultaneously, which will drive adoption at all levels. As 2025 progresses, it will be interesting to observe how other players respond to Nvidia’s multifaceted approach, and how quickly the industrial sector will adopt these new advancements.
For ARC Advisory Group recommendations for navigating the AI Wars, and closing the digital divide by embracing Industrial AI, and governing and guiding major decisions about enterprise, cloud, industrial edge and AI software, please contact Colin Masson at [email protected].