Our team had the opportunity to attend NVIDIA GTC 2025 this year, and it was evident throughout the event that AI and high-performance computing are no longer emerging technologies — they're already being used to solve real problems in production environments.

We have moved from asking whether AI can help solve complex real-world problems to focusing on how to implement it faster, safer, and at a grander scale.

The sessions we attended showed how widely these tools are being applied, from automotive design and pharmaceutical research to financial systems and genomics. The conversation has shifted from questioning whether AI can make a difference to focusing on how to implement it efficiently, responsibly, and at scale.

Let's go through the most interesting takeaways we brought from NVIDIA GTC in 2025.

Open-Source AI at Scale

One of the most insightful sessions we attended was the panel discussion titled "Scaling Open Source AI: From Foundation Models to Ecosystem Success." With an all-star lineup featuring Ion Stoica (UC Berkeley), Joe Spisak (Meta), Matt White (PyTorch Foundation / Linux Foundation), Ankit Patel (NVIDIA), and Stephanie Zhan (Sequoia Capital), the conversation was as robust as it was timely.

Here are the top 4 insights from the talk.

1. Open Source Isn't One-Size-Fits-All

The term "open-source AI" is no longer a simple label. It now spans a broad spectrum, from frameworks like PyTorch and TensorFlow to models with varying levels of openness — some fully permissive, others with commercial restrictions. Panelists highlighted how businesses navigate the gray area between fostering innovation and protecting their IP.

There's no universal playbook for how "open" an AI model or ecosystem should be.  Importantly, open science goes even broader, encompassing architecture, weights, datasets, and training code. While this level of transparency benefits research, it often clashes with the practical and legal realities of enterprise deployment.

2. Licensing Can Make or Break Innovation

The licensing conversation was front and center. There has been a clear evolution from restrictive models to more permissive licenses that encourage experimentation and downstream innovation. For example, relaxed licensing enables techniques like model distillation, making large-scale models more accessible to smaller players.

However, this openness comes with a caveat: How do we balance innovation with protecting proprietary technology? Licensing isn't just a legal checkbox — it directly affects how freely developers can refine, deploy, and build upon models.

3. Community Governance is Everything

The sustainability of open-source AI hinges on community governance. Structured leadership, whether through foundations, core committees, or the "benevolent dictator" model, is essential for maintaining project momentum and integrity.

Foundations like the Linux Foundation (home to PyTorch) not only support technical development but also help ensure IP protection, vendor neutrality, and a long-term vision. Encouraging diverse contributors while keeping a clear leadership structure is a delicate but necessary balance.

4. The Innovation Engine

Open source has been the driving force behind many of the most exciting advances in AI. From content moderation tools to model optimization techniques, the ecosystem has matured quickly thanks to open collaboration.

But this growth also brings tension: how can we maintain transparency and community-driven progress while meeting the increasingly complex needs of businesses? It’s clear that open-source AI is at a crossroads, requiring new frameworks to balance accessibility, control, and real-world utility.

At Janea Systems, we've had the opportunity to contribute directly to this innovation cycle by porting PyTorch to Windows, helping make the framework more accessible across platforms. Our work also includes diagnosing a major matrix multiplication performance issue and enabling the MKL sparse matrix module on PyTorch — two challenges that reflect the needs of modern AI deployment.

In summary, this panel clarified the complexities and trade-offs of open-source AI today. The conversation underscored a central theme: the future of AI requires building ecosystems, not just fine-tuning models.

Edge Computing: Smarter AI at the Source

As a team invested in edge technologies, we were drawn to Chen Su's session, NVIDIA's Sr. Technical Product Marketing Manager, called "Edge Computing 101: Introduction to Smart Edge and Autonomous Robots." His talk laid out the case for why the future of AI is happening at the edge, not just in the cloud.

Real-Time > Round-Trip

Edge computing minimizes latency by processing data where it's generated. This shift is essential for robotics, autonomous checkout systems, industrial inspection, and smart city infrastructure, where split-second decisions are critical.

Privacy, Sovereignty, and Speed

Local processing isn't just about speed. It's about control. More organizations prioritize data sovereignty and security, making edge solutions increasingly attractive for privacy-sensitive applications.

Hardware to Match the Moment

NVIDIA's edge portfolio — including Jetson, IGX, and edge GPUs — brings data center-level performance to the edge. These platforms empower developers to run powerful inference models and perform real-time sensor fusion closer to the source.

Generative AI Meets the Physical World

A standout insight from Chen Su's talk was the concept of "physical generative AI." This involves bringing generative models to robotics and real-time environments, enabling systems to interact intelligently with the physical world. It's a breakthrough with massive implications for autonomy, human-machine interaction, and instant decision-making.

Diverse Use Cases, One Common Thread

From agriculture and healthcare to smart retail and intelligent traffic systems, the common denominator is clear: real-time AI is best deployed at the edge. This trend is fueling a new generation of business models and operational efficiencies.

AI-Ready Infrastructure

The demand for low-latency, high-throughput AI workloads at the edge is growing. Platforms like Jetson and IGX are stepping up with compute capabilities once reserved for centralized data centers, enabling sophisticated models to run anywhere.

End-to-End Stack Support

Tools like Isaac Sim, Metropolis, and Holoscan simplify the entire pipeline — from simulation and testing to real-world deployment. These frameworks help teams quickly move from idea to implementation across sectors like robotics, healthcare, and retail.

We're excited to continue leading the way in edge computing, and Chen Su's insights have only deepened our conviction that the smartest AI happens not in the cloud, but at the source.

AI Transforming Live Sports

One of the most unexpected — and incredibly engaging — highlights from our time at NVIDIA GTC was a standout panel featuring Javier Gil (Head of AI at La Liga, the Spanish football governing organization), Marta Mrak, and David Lehanski. This expert trio delivered a robust and forward-thinking session on how AI is revolutionizing the world of live sports.

Here's what really stuck with us.

Smarter Storytelling

One of the most fascinating takeaways was how real-time storyline generation is used to reshape the narrative of sports broadcasts. By dynamically generating context and drama around live events, AI is helping broadcasters create more immersive and emotionally engaging experiences for fans, both on-screen and across second-screen platforms.

Large Video Models (LVMs)

Large Video Models are pushing the boundaries of what's possible in sports content. From automated highlight reels to deep tactical analysis, LVMs enable hyper-personalized content creation and unlock insights that were previously impossible to scale. This tech isn't just about flashy edits — it's changing the way we understand and enjoy the game.

AI as a Competitive Edge

AI isn't just enhancing the fan experience — it's embedded directly into league-wide toolkits, offering strategic advantages to teams and organizations. From player performance tracking to game strategy optimization, clubs leverage these tools to stay ahead of the curve in an increasingly competitive landscape.

Fighting Audiovisual Fraud

With so much revenue at stake in live sports, audiovisual fraud detection has become a top priority. AI-powered systems can identify and combat piracy in real time, helping safeguard intellectual property and ensuring that organizations can retain and protect their broadcast revenue streams.

A Broader Look: What's Happening Across Industries

Beyond the individual sessions, GTC 2025 offered a clear view of how AI, edge computing, and simulation technologies come together across sectors. Here are a few key trends we saw repeated throughout the week:

Digital Twins, Edge AI, and HPC Are Merging

A recurring theme was the tight integration of real-time sensor data, fast AI inference, and high-performance simulation. Industries no longer use digital twins just for planning — they're now core to live operations. These virtual counterparts are continuously updated, creating a dynamic link between physical systems and digital reflections.

Autonomy Is Becoming the Default

What used to be separate, manual, or heavily siloed systems are being consolidated into unified platforms. We're seeing more examples of self-optimizing systems—from factory robots to financial algorithms to real-time diagnostics in healthcare. Human oversight still matters, but the gap between machine-led and human-led processes is closing fast.

Trust, Privacy, and Compliance Are Front and Center

As generative AI and real-time systems become more widespread, so do the challenges around data governance. Whether it's zero-trust architectures in finance or compliance standards in industrial safety, organizations are investing heavily in frameworks that ensure their AI implementations are secure, auditable, and compliant from day one.

Tangible Benefits for Society and the Economy

AI is already making a measurable impact. In healthcare, it helps bridge access gaps with tools like tele-surgery and AI-based diagnostics. In manufacturing, it supports more localized production, workforce upskilling, and stronger supply chains. In finance, better analytics improve resilience and open the door for smaller players to compete.

AI Innovation Is a Team Sport

One of the most exciting takeaways was how closely hardware and software development work together. From chipmakers and system designers to banks, automakers, and pharma companies — everyone is collaborating to build an ecosystem that can support AI at scale. The result is faster iteration, better performance, and more targeted solutions.

---

If any of these topics sparked ideas or align with challenges you tackle, we’d love to connect. Whether it’s exploring AI at the edge, open-source strategy, or how to operationalize AI safely and at scale — we’re always up for a conversation.

Get in touch with us via contact form or reach out through our LinkedIn.