Nvidia GTC 2025：人工智能进入企业基础设施

Companies are changing the architecture for how By Sarah Elk and Eric Sheng Nvidia GTC 2025:AIMaturesintoEnterpriseInfrastructure We came away from Nvidia’s GTC event last week with a reinforced belief that AI has moved well beyondpilots and proofs of concept to scaled deployment and operational impact. Consistent with what we see inour work with clients, organizations are no longer testing AI at the edges. They are re-architecting how 1. “No data, no AI”—and now, AI is generating the next layer of data. Data remains the biggest challenge and the biggest opportunity. Every successful AI deploymentpresented at GTC rested on clean, connected, and accessible data. But the frontier is no longer just dataconsumption; it’s data creation. AI is surfacing previously invisible insights: operational patterns, best Data remains the biggest challenge and the biggestopportunity. Every successful AI deployment presented at 2. Smaller, specialized models are rewriting the economics of AI. The dominance of large, general-purpose models is giving way to smaller, fine-tuned systems built forspecific domains. Techniques like quantization, pruning, and retrieval-augmented generation (RAG) areunlocking cost savings without sacrificing performance. Enterprises are increasingly embracing model 3. Agentic AI is starting to gain traction—and trust depends on structure. The progression has moved from RAG to AI assistants and now agentic AI. Fully autonomous AI agents,however, are rarely deployed at scale. The biggest challenge remains evaluating the accuracy of an agent’s Nvidia GTC 2025:AIMaturesintoEnterpriseInfrastructure emerged: Structure matters. Enterprises are prioritizing transparency, escalation paths, redundancyguardrails, traceability and auditability in production, and predictable behavior. While fully autonomousagents remain rare, semiautonomous systems—with human oversight—are the pragmatic near-term 4. Digital twins and simulation are now everyday enterprise tools. Simulation has shifted from innovation showcase to standard operating practice. Teams are using digitaltwins to model factories, stores, and supply chains, testing changes virtually before implementing themphysically. The result is faster rollout cycles, lower risk, and more confident decision making. Executives 5. Video is becoming the next major dataset. Computer vision and video language models are transforming video from passive monitoring to activeintelligence. Organizations are using video to analyze customer behavior, product interaction, compliance,and safety in real time. These insights are now directly influencing merchandising, labor planning, and 6. Enterprises are shifting from build to buy for large language modeloperations (LLMOps) and infrastructure—and adoption is accelerating. Off-the-shelf tools like Nvidia DGX Cloud and Inference Microservices (NIM) are lowering the barriers toenterprise AI adoption. Companies are launching copilots, knowledge assistants, and other custom genAI applications without having to build deep machine learning operations or LLMOps stacks. We’re also 7. Simulation is emerging as the new collaboration layer. Beyond modeling, simulation is becoming a unifying platform for cross-functional teams. With NvidiaOmniverse integrating seamlessly into design and operations tools, teams can cocreate in virtualenvironments before making real-world changes. This is reducing iteration cycles, improving Nvidia GTC 2025:AIMaturesintoEnterpriseInfrastructure 8. Custom model deployment is the new organizational bottleneck. Fine-tuning foundation models is becoming easier, but deploying them into production is still hard.Teams face challenges around performance optimization, latency, hardware compatibility, and security. Fine-tuning foundation models is becoming easier, butdeploying them into production is still hard. 9. Multimodal AI is transforming creative workflows and brand expression. Tools like Nvidia Picasso, Adobe Firefly, and open-source diffusion models are enabling teams to generateproduct visuals, videos, 3D assets, and social content—all from natural language prompts. Companies arescaling content generation with creative pipelines from platforms like RunwayML, Canva, and Synthesia. Bold ideas. Bold teams. Extraordinary results. Bain & Company is a global consultancy that helps the world’s most Across the globe, we work alongside our clients as one team with a shared ambition to achieveextraordinary results, outperform the competition, and redefine industries. We complement our tailored,integrated expertise with a vibrant ecosystem of digital innovators to deliver better, faster, and more