Writer: Jordan Sprogis
Editor: Lillian Castro
Reviewer: Cristian Lopez
Key Takeaways
- NVIDIA’s annual GPU Technology Conference (GTC), held March 16-21 in San Jose, attracted more than 25,000 people in Silicon Valley.
- CEO Jensen Huang unveiled behind-the-scenes information, including the announcement of the new Blackwell GPU, which has already sold 3 million units.
- Hear from industry experts who attended the conference, including Octavian Tanase, Chadd Kenney, Michael McNerney, and Jake Madders.
The “Live at NVIDIA GTC With Acquired” broadcast aptly set the stage for what would become yet another defining moment for computing: “San Jose has been painted green to get ready for the thousands of folks who made the trek for this year’s GTC.”
More than 25,000 visitors traveled from far and wide to attend NVIDIA’s annual GTC held from March 16 to 21 in sunny Silicon Valley. Since it started in 2009, the convention has become one of the most-attended events for discussing, learning, and sharing all things AI, computing, and data center innovations.
NVIDIA GTC had a special keynote speaker, as it does every year: CEO Jensen Huang, who led a two-hour seminar in a sold-out auditorium.
Aside from the highly anticipated keynote were also several other workshops, trainings, speakers, and exhibits, drawing a medley of developers, engineers, researchers, and business leaders within the IT space.
HostingAdvice spoke with representatives from Hitachi Vantara, Pure Storage, Supermicro, and Hyve Managed Hosting, who attended this year’s GTC event.
They each touched on consistent themes across the conference — the emergence of agentic AI, quantum computing, storage/scalability requirements, and of course, NVIDIA’s newest GPU advancements.
Blackwell Architecture: Next-Generation Computing Power
As tradition dictates, CEO Jensen Huang delivered the keynote address, unveiling NVIDIA’s newest innovations to a sold-out audience.
Clad in his iconic leather jacket, Huang introduced the new Blackwell architecture, which is designed specifically for advanced AI and HPC workloads.
“This is a sight of beauty, wouldn’t you agree?” Huang asked, gesturing to an impressive array of servers from Microsoft, Google, Amazon, and others, meticulously arranged and connected.
He explained that Blackwell represents a “fundamental transition in computer architecture,” engineered for both power and efficiency at a massive scale.
The numbers tell a story of its own: Blackwell features up to 208 billion transistors and uses a custom-built TSMC 4NP process for optimal performance. Perhaps most impressively, thanks to Blackwell’s design and disaggregated NVLink switches, a full exaflop (1 quintillion operations per second) can now be packed into a single rack.
“And as a result, we have a one-exaflops computer in one rack. Isn’t it incredible?” Huang said.
Changes in Infrastructure for AI Demands?
Jake Madders, Director and Co-Founder of Hyve, is one of the many who said they look forward to Huang’s keynote every year.
“The highlight of GTC 2025 for me is definitely Jensen Huang’s keynote. Jensen is a visionary leader who is reshaping the industry,” said Madders. “This year, we expect the introduction of the Blackwell Ultra server computing accelerators and possibly news about the Rubin architecture that will replace Blackwell.”
With Blackwell, the keyword here seems to be scalability.
With NVIDIA holding a 98% revenue share in the data center GPU market, Blackwell’s improved scaling and performance capabilities (compared to its predecessor Hopper) come at a critical moment as computational demands grow thanks to the continued rise of genAI and agentic AI technologies.
“The latest GPUs are quickly expanding the ability for new, more complex models to be trained and perform complete inferencing at the edge,” noted Michael McNerney, SVP of Marketing and Network Security at Supermicro.
Octavian Tanase, Chief Product Officer at Hitachi Vantara, described Blackwell as “revolutionizing industry verticals by driving unprecedented levels of innovation and efficiency.”
For example, financial firms can run fast-paced trading systems and detect fraud in real time; healthcare organizations can accelerate drug discovery and personalize treatments; autonomous vehicle development gets better real-time data processing; and manufacturing operations can predict equipment issues before they occur.
The transformation extends to storage infrastructure as well. Chadd Kenney, VP of Technology at Pure Storage, pointed out that AI advancements are challenging traditional approaches.
“Bottlenecks in legacy architectures are no longer acceptable in an era where AI-driven innovation depends on keeping GPUs fully utilized across massive datasets,” said Kenney. “Enterprises need a storage platform engineered for the future, not one retrofitted from the past.”
For web hosts and data centers, moving to architectures purpose-built for AI workloads is becoming more important (and more common, evidently, as NVIDIA has already sold more than 3 million Blackwell GPUs).
The GPU’s ability to pack enormous computing power into a smaller, more energy-efficient footprint doesn’t just benefit the hyperscalers. It’s got the potential to be a foundation for smaller web hosts and providers to join the race in deploying AI-ready infrastructure using less space and energy than these mega-enterprises do.
NVIDIA GTC 2026 dates haven’t been announced yet, but in-person tickets go early, so keep an eye on NVIDIA’s official calendar toward the end of 2025.
