It’s Monday March 17, 2025, and the streets of San Jose, Calif. are covered in green. You might assume that, on St. Patrick’s Day, the flashes of green on buildings, windows, cars, and buses celebrated the renowned Irish holiday. But you’d be wrong. San Jose was painted green for the annual NVIDIA GPU Technology Conference, and as Gen Z might say, the vibe “was lit!”
The week of GTC was not unlike previous conferences in the sense that it had a similar format of show-floor exhibits and breakout workshops. However, the atmosphere had a different energy. Exhibits showcased primarily established technology organizations, and the wares peddled were practical and solution-oriented, central to data management and infrastructure needs. This pivot from the exhaustive “what’s possible with AI” fringe discussion to the “practical implementation of AI” was surprising, reassuring, and reflective of the maturation of the industry and the company.
Here are five key takeaways from my time at GTC 2025:
Software development tools are as essential as advances in hardware acceleration
As expected, NVIDIA highlighted the leaps in processing power they’d achieved this year, boasting the capabilities of their latest AI chips, Blackwell Ultra (2025), Vera Rubin (2026) and Vera Rubin Ultra (2027). However, they also highlighted NVIDIA Dynamo, an open-source inference software designed to efficiently deploy and scale AI models. For the industry, this gives us the conviction that we are beyond the early stages of training models and ready to reap the benefits en masse (and ROI) via the outcomes and decisions these models infer, or inference. More importantly, being an open-source platform, we should expect accelerated adoption.
The need for dual operations
During his keynote, NVIDIA’s CEO Jensen Huang said, “Every industry, every company that has factories will have two factories in the future… the factory for what they build and the factory for the mathematics, the factory for the AI.” He gave the auto manufacturing industry as an example: “A factory for cars” and a “Factory for AI for the cars”. What does this mean? It indicates we should expect a boon in a new type of infrastructure central to proprietary data management and data protection. If you thought the Coca-Cola recipe was heavily guarded, imagine how companies will need to protect the intellectual property of their deployed AI models after investing in them. A unification process will need to occur downstream between these two factories. The infrastructure deployed to support inferencing of AI models at the edge will require sophisticated hardware and software integration to manage, maintain, and govern these solutions.
Robots, robots everywhere…
The idea of robots certainly isn’t new. Humans have been leveraging robotics to perform daily, repetitive tasks for decades. However, sprinkle in some AI and autonomy and suddenly it feels like we’re deep into Hollywood sci-fi territory. NVIDIA emphasized the significance of AI and robotic integration to key industries like industrial automation and med-tech, where these advances can be lifesaving. To further accelerate the adoption of robotics, NVIDIA introduced two open-source models. NVIDIA Isaac Gr00t N1, an open-source humanoid robotics foundation model, aimed to accelerate the adoption of humanoid robots and their Newton Physics Engine, an open-source physics engine designed to simulate training for robots to interact with the physical environment. We’re about to put AI to work, literally.
Tokens, Trillions and Trillions of Tokens
As with any industry show, you can bet that you will hear the emergence of this year’s bingo card of techie buzzwords. Through a plethora of “north stars,” “blue oceans,” “circle backs,” and “low hanging fruit,” two specific words stuck out to me more than the rest. Tokens and Trillions. Individually, tokens and trillions may not sound so impressive. But together? Now you’ve caught my ear. Let me explain why.
In AI terms, a token is how the AI thinks. AI models analyze words, phrases, and individual characters to understand the context and fulfill requests. These models rely on tokens, which comprise vast datasets of words, phrases, and characters. While some models train on billions of tokens, more advanced models like Gemini 1.5, Claude 3, and GPT-4 likely draw from trillions of tokens.
Of course, to process these trillions of tokens being used to train models, you need a heck of a lot of horsepower. Something like NVIDIA’s GB10 Grace Blackwell Super-Chip, capable of delivering up to 1,000 TRILLION operations per second of AI compute. To comprehend how much a TRILLION is, think of it this way. If I paid you $100,000 per day in 10 days, I would have given you $1M. In 27.4 YEARS, I would have paid you $1B. But $1T… a trillion dollars? It would take me 27,397 YEARS to pay you that $1T.
Relationships matter
Lastly, the aspect of GTC25 that resonated most with me was the emphasis on partnerships. You’ve likely heard an African proverb: “If you want to go fast, go alone. If you want to go far, go together.” This proverb highlights the need for collaboration, support, and partnerships to sustain success over the long run. NVIDIA has always fostered partnerships with its developer community to drive design, adoption, and technical advancements. But this year was FULL of announcements beyond their development and startup community.
Relationships with major companies signify the importance “going together,” even for large technology companies. As I made my way home after the show, my brain still overstimulated from the obscene amount of green before my eyes, I found myself blurring the lines between “what WILL be and what already IS.” So many of yesterday’s impossibilities have been achieved or even exceeded today. Many of them were on display last week in a San Jose convention center. The one abundantly clear thing is that we’re nowhere close to understanding the many ways AI will transform our lives. It may seem like we’re all knee-deep in the next space race and on the precipice of ‘groundbreaking’ every single day. That’s because we are. One can only imagine where our industry will be at next year’s GTC.
From Prototype to Market: Ready to bring your AI innovations to the world? Arrow’s end-to-end support ensures your solutions are optimized, integrated, and market-ready. Partner with us for seamless execution. Contact one of our experts today to learn more.
