NVIDIA GTC 2025 Keynote Recap
The NVIDIA GTC 2025 Keynote, delivered by CEO Jensen Huang, has concluded at the SAP Center in San Jose. The Stelia team is on the ground, capturing insights from GTC 2025 and available to chat all things AI and GTC at our booth #1033 in Expo Hall 3.
💡 Stelia is already partnering with the most advanced and ambitious companies where AI is core to their business. Drop by the GTC booth, grab some swag, and see how we can help you too!
Here are the top Keynote takeaways, industry insights and notable developments to watch as GTC 2025 moves full steam ahead this week. Our team is on the ground to bring you the latest from GTC. 
“100% of NVIDIA engineers will be AI assisted by the end of this year”
Top Insights and Takeaways from NVIDIA GTC Keynote 2025
- Blackwell Ultra GPUs (B300 Series): Powering the AI Boom
NVIDIA unveiled the Blackwell Ultra GPUs, specifically the B300 series, as a significant upgrade to the Blackwell architecture. Blackwell Ultra NVL72 is coming H2 2025.
- These GPUs deliver a reported 1.5x improvement in FP4 performance and feature 288GB of HBM3e memory, though they push power consumption to 1400W. Huang’s emphasized “accelerated computing” driving a renewed Industrial Revolution, with the B300 GPUs positioned as the backbone for trillion-parameter AI models. For B2B stakeholders, this cements NVIDIA’s leadership in data center compute, targeting hyperscalers and enterprises scaling AI workloads. Increasingly becoming a power limited industry.
“the more you save, the more you buy.”
Blackwell offers 40x the inference performance of Hopper GPU generation.
- GB300 Grace Blackwell Superchip: Integrated AI Supercomputing
The GB300 platform, pairing Blackwell Ultra GPUs with NVIDIA’s Grace CPU, was a centerpiece of the keynote. Huang’s stated that this system offers “40x the performance” of the previous Hopper architecture, a bold metric signaling NVIDIA’s focus on end-to-end AI infrastructure. This platform’s scalability and efficiency make it a compelling option for cloud providers and sovereign AI projects, potentially driving NVIDIA’s Data Center revenue past $300 billion by 2029.
- Rubin: Rubin’s new networking architecture doubling performance could have significant implications for AI model training efficiency.
- Rubin Ultra NVL576, Coming H2 2027. 100 PFLOPS of dense FP4. 1024GB HBM capacity.
- Vera Rubin NVL 144: Coming H2 2026. 50 PFLOPS of dense FP4. 3nm process. large jump in logic density going from 4NP on Blackwell to 3NP for Rubin. Estimated higher TDP (1800W from industry analysts).
- NVIDIA Dynamo: AI Inference Efficiency: The launch of NVIDIA Dynamo, an open-source inference software, is a strategic move. AI inference costs are climbing higher, with companies like OpenAI and Google investing heavily in optimizing model deployment. Dynamo’s ability to double performance on models like Llama is a direct challenge to rivals like AMD and Google, which are working on more power-efficient AI inference chips.
- Automotive Ambitions: GM Partnership and Halos Safety System
A major announcement was NVIDIA’s partnership with General Motors to power GM’s next-generation self-driving fleet using NVIDIA’s DRIVE platform. The reveal of “Halos,” a new safety system for autonomous vehicles (AVs), enhancing NVIDIA’s push into the multi-trillion-dollar AV market. For B2B clients in automotive, this partnership validates NVIDIA’s ecosystem as a critical enabler of safe, scalable AV deployment.
- cuOpt Goes Open-Source: Democratizing AI Optimization
NVIDIA announced the open-sourcing of cuOpt, an AI-driven decision optimization tool previously locked behind premium access. A strategic move to “broaden access to its ecosystem,” aligning with NVIDIA’s full-stack vision alongside tools like NIM and Omniverse. Businesses in logistics and operations can leverage cuOpt’s capabilities for free, lowering entry barriers and fostering broader CUDA adoption.
- AI’s Inflection Point: A New Computing Paradigm
Huang declared that computing has crossed an “inflection point,” shifting from hand-coded software on CPUs to machine-learning software on GPUs and accelerators. Huang saying, “We’ve passed that point now,” framing NVIDIA as the architect of this era. For B2B audiences, this narrative reinforces NVIDIA’s role in enabling AI-first strategies, with GPUs as the essential infrastructure.
- Quantum and Robotics Horizons
The keynote teased advancements in quantum computing and robotics. Notably Huang’s mention of a “Quantum Day” on March 20, suggesting deeper exploration with partners like Quantinuum, alongside robotics updates tied to 2024’s Project Groot. These long-term bets appeal to R&D-intensive sectors like healthcare and manufacturing, hinting at NVIDIA’s future growth vectors beyond traditional GPUs.
- Isaac GR00T N1: Robotics Foundation Model: Robotics has lagged behind AI in terms of real-world deployment, mainly due to the challenge of training models across diverse environments. Isaac GR00T N1, a pretrained robotics foundation model, is unlocking future growth. By standardizing training across companies like Boston Dynamics and Agility Robotics, NVIDIA is setting the stage for more capable humanoid robots. Oh, and it’s Open Source…
- Spectrum-X Silicon Photonics: Spectrum-X Silicon Photonics Ethernet Switch offers 1.6 Tbps per port, addressing bottlenecks in large-scale AI training and inference workloads. Silicon photonics (optical networking) is a further leap in reducing energy consumption and increasing bandwidth density.
- Enterprise AI and Personal AI Workstations: DGX Spark and DGX Station. DGX Spark and DGX Station signals NVIDIA’s push toward personal AI supercomputing. Project Digits is now called DGX Spark.
- Data Center Dominance: $1 Trillion Revenue Vision
Huang projected NVIDIA’s data center revenue could reach $1 trillion by 2028, driven by demand from hyperscalers like AWS, Google, Microsoft, and Meta. Focus on “inference scaling laws” and agentic AI, underscoring NVIDIA’s ambition to power the next wave of AI innovation. This aggressive forecast signals confidence in sustained enterprise adoption of NVIDIA’s hardware and software stack.
- Agentic AI and Reasoning Models: The Next Frontier
Huang’s introduction of “agentic AI” with advanced reasoning capabilities, showcased via a live demo of “AI nurses” answering medical queries. The evolution from generative to reasoning AI positions NVIDIA to address enterprise needs in healthcare, customer service, and beyond, leveraging its inference microservices and GPU power.
- Celebrating 25 Years of GeForce: A Nod to NVIDIA’s Legacy
Huang marked 25 years of GeForce, tying its legacy to AI-driven graphics advancements. Noting “AI has transformed graphics and computing,” a nod to NVIDIA’s roots that resonates with B2B clients in gaming and creative industries, even as the future focus shifts to AI.
Strategic Implications for B2B Stakeholders
- Competitive Edge: The B300 GPUs and GB300 platform widen NVIDIA’s moat against AMD and Intel, emphasizing performance and ecosystem lock-in. Competitors must innovate on cost or efficiency to counter this.
- Vertical Impact: Automotive (GM, Halos), logistics (cuOpt), and cloud (data center scale) are priority sectors. Businesses should align procurement and R&D with NVIDIA’s roadmap.
- Cost vs. Power: With B300’s 1400W draw and premium pricing, enterprises must balance performance gains against operational costs, a critical consideration for mid-tier adopters. Rack density is increasingly making AI a power limited industry.
The GTC 2025 keynote again provided exciting product reveals, gleans into future product roadmaps and the future trajectory of NVIDIA. Captivating the audience with demos like AI nurses and bold claims like the 40x Hopper comparison. This year’s event positions GTC as a showcase of NVIDIA’s “accelerated computing revolution,” with updates reflecting real-time excitement around Halos, cuOpt, and agentic AI.
As usual, delivered with high levels of energy and charisma, personified with a sartorial flair for black leather. NVIDIA’s CEO blend of innovation and charisma, no doubt a potent mix for inspiring enterprise confidence and captivating the wider technorati sphere.
NVIDIA GTC 2025 positions NVIDIA as the current unrivaled leader in AI infrastructure. From Blackwell Ultra’s raw power to strategic moves in AVs, open-source software, and reasoning AI, Huang’s keynote paints a future where NVIDIA isn’t just selling hardware, it’s shaping current and future industries. For B2B decision-makers, the message is clear: NVIDIA’s ecosystem is the go-to foundation for the AI-driven enterprise, with a trajectory that’s both technically groundbreaking and commercially impactful.
